WO2023011381A1 - 数据处理方法及装置 - Google Patents

数据处理方法及装置 Download PDF

Info

Publication number
WO2023011381A1
WO2023011381A1 PCT/CN2022/109296 CN2022109296W WO2023011381A1 WO 2023011381 A1 WO2023011381 A1 WO 2023011381A1 CN 2022109296 W CN2022109296 W CN 2022109296W WO 2023011381 A1 WO2023011381 A1 WO 2023011381A1
Authority
WO
WIPO (PCT)
Prior art keywords
communication device
data
neural network
accuracy
parameter
Prior art date
Application number
PCT/CN2022/109296
Other languages
English (en)
French (fr)
Inventor
张公正
徐晨
王坚
李榕
王俊
Original Assignee
华为技术有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 华为技术有限公司 filed Critical 华为技术有限公司
Priority to EP22852094.6A priority Critical patent/EP4366253A1/en
Publication of WO2023011381A1 publication Critical patent/WO2023011381A1/zh
Priority to US18/431,054 priority patent/US20240214329A1/en

Links

Images

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L49/00—Packet switching elements
    • H04L49/55—Prevention, detection or correction of errors
    • H04L49/552—Prevention, detection or correction of errors by ensuring the integrity of packets received through redundant connections
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L1/00—Arrangements for detecting or preventing errors in the information received
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L25/00—Baseband systems
    • H04L25/02—Details ; arrangements for supplying electrical power along data transmission lines
    • H04L25/0202—Channel estimation
    • H04L25/024—Channel estimation channel estimation algorithms
    • H04L25/0254—Channel estimation channel estimation algorithms using neural network algorithms

Definitions

  • the present application relates to the technical field of communications, and in particular to a data processing method and device.
  • DNN deep neural networks
  • wireless transmission is designed to provide data transmission services with reliable transmission as the design criterion.
  • the physical layer may perform processing such as channel coding and modulation on the bit stream of the upper layer to obtain processed data, and then transmit the processed data through the air interface. If the channel quality (such as signal-to-noise ratio) is different, the corresponding data transmission capacity is also different.
  • the adaptive coding modulation and automatic retransmission request mechanism can adapt to channels of different qualities and provide dynamic transmission rates.
  • the performance index of transmission is mainly packet error rate, and transmission parameters (such as modulation and coding scheme (MCS)) are adjusted according to channel quality to maximize throughput or reliability requirements.
  • MCS modulation and coding scheme
  • the above wireless transmission method aims at error-free transmission, and takes the packet error rate as the basis for the adjustment of transmission parameters; while the main goal of neural network processing is accurate calculation, which has a good ability to tolerate errors, and error-free transmission may cause wireless resource loss. Inefficient use.
  • the present application provides a data processing method and device, which can effectively improve the utilization efficiency of wireless resources.
  • the embodiment of the present application provides a data processing method, the method comprising:
  • the first communication device obtains first data; the first communication device processes the first data according to a first parameter to obtain a first transmission symbol, and the first parameter is determined according to a first accuracy and first channel information , the first channel information is the channel information between the first communication device and the second communication device, and the first accuracy is used to represent the accuracy of data processed by the neural network in the second communication device; The first communication device sends the first transmission symbol to the second communication device.
  • the above-mentioned first accuracy can also be understood as: the accuracy of the processing results output by the neural network multiple times; or, the average of the accuracy of the processing results output by the neural network; or, the average performance of the neural network processing data.
  • the average performance includes average accuracy, average accuracy, or average accuracy, etc.
  • the above-mentioned first parameter is determined according to the first accuracy and the first channel information. It can also be understood as: the first parameter is related to the first accuracy and the first channel information, or the first parameter is determined according to the first accuracy and the first channel information. channel information is obtained. Determining the parameters used by the first communication device when processing data according to the first accuracy and the first channel information may enable the first communication device to process data according to the requirements of the neural network for processing data. Therefore, the first data is processed by making full use of the accuracy requirements of the neural network to process data and the channel quality, thereby improving the utilization rate of wireless resources.
  • the first communication device may process the first data according to the first parameter, where the first parameter is determined according to the first accuracy and the first channel information. Therefore, when the first communication device transmits data, the data processing only needs to comply with the first accuracy, and the second communication device can perform neural network processing on the data, which improves the accuracy of data retransmission due to the need to ensure the correctness of data transmission. In this case, the utilization efficiency of wireless resources is effectively improved.
  • the first parameter includes a first transformation coefficient, and the first transformation coefficient is used to represent a dimension ratio between the first data and the first transmission symbol; or, the The first parameter includes a first modulation order and a first coding rate.
  • the above-mentioned first transform coefficient is used to represent the dimensional ratio between the first data and the first transmission symbol.
  • the number of elements of the first data and the number of elements of the first transmission symbol may be used to represent the dimension ratio between the first data and the first transmission symbol.
  • a larger first transformation coefficient indicates that the first communication device retains fewer effective features when processing the first data. Therefore, the less transmission resources used when transmitting the first transmission symbol, the lower the first accuracy of the neural network for processing data.
  • the channel information is the same (for example, the channel quality is the same)
  • the smaller the first transform coefficient is the higher the requirement for the first accuracy is.
  • the better the channel quality represented by the first channel information the larger the first transformation coefficient.
  • the above method can support dynamic first conversion coefficients and first accuracy in a wide range of channel quality intervals, thereby improving wireless resource utilization efficiency.
  • the first accuracy is represented by any one or more of the following: the confidence degree of the processing result of the neural network; the classification accuracy of the processing result of the neural network; Mean square error (mean squared error, MSE) between the input data of network and described first data; Mean absolute error (mean absolute error, MAE) between the input data of described neural network and described first data .
  • the first accuracy may be represented by an average of confidence levels corresponding to multiple processing results of the neural network.
  • the first accuracy may be represented by an average of probabilities corresponding to multiple processing results of the neural network.
  • the first accuracy may be represented by an average of classification accuracy corresponding to multiple processing results of the neural network.
  • the first accuracy can be represented by the average MSE between the data input by the neural network multiple times and the real data, or the root mean square error between the data input by the neural network and the real data (root mean square error, RMSE), or the average representation of the MAE between the data input by the neural network and the real data, etc. It can be understood that the expression manner of the first accuracy shown here is only an example, and other types of output of the neural network are not limited in this embodiment of the present application.
  • the processing the first data by the first communication device according to the first parameter includes: transforming the first data by the first communication device according to the first conversion coefficient , to obtain the first transmission symbol; or, the first communication device encodes and modulates the first data according to the first modulation order and the first coding rate to obtain the first transmission symbol symbol.
  • the method further includes: the first communication device sends a first request message to the second communication device, the first request message is used to request the neural network to perform data processing , the first request message includes first indication information, and the first indication information is used to indicate the first accuracy; the first communication device receives the first response message from the second communication device, the The first response message includes second indication information, where the second indication information is used to indicate the first parameter.
  • the method further includes: the first communication device receives a second request message from the second communication device, the second request message is used to request data, and the second The request message includes third indication information, where the third indication information is used to indicate the first parameter.
  • the method further includes: the first communication device receiving a feedback message from the second communication device, where the feedback message includes information used to indicate the processing result.
  • the method further includes: the first communication device receives retransmission instruction information from the second communication device, and the retransmission instruction information is used to instruct retransmission of the first Data: the first communication device retransmits the first data according to the retransmission indication information.
  • the retransmission indication information is further used to indicate a second parameter, where the second parameter is an updated parameter of the first parameter.
  • one or more of the following information is carried in the neural network processing control channel: the first indication information, the second indication information, the third indication information or the retransmission Instructions.
  • the first transmission symbol is carried on a neural network processing shared channel.
  • the neural network processing control channel and the neural network processing shared channel are designed based on the requirements of neural network processing.
  • the neural network processing shared channel transmits the first transmission symbol, even if an error occurs, the second communication device can still process the received transmission symbol, thereby improving the utilization rate of wireless resources.
  • the neural network processing control channel and the neural network processing shared channel shown in the embodiment of the present application can not only be used to transmit parameters or data related to neural network processing, but also can be used to transmit other information that does not require error-free transmission. data etc.
  • an embodiment of the present application provides a data processing method, the method comprising:
  • the second communication device receives the first transmission symbol from the first communication device; the second communication device processes the first transmission symbol according to the first parameter to obtain the input of the neural network, and the first parameter is according to the first Accuracy and first channel information are determined, the first channel information is channel information between the first communication device and the second communication device, and the first accuracy is used to represent the neural network processing data accuracy; the second communication device processes the input according to the neural network to obtain a processing result.
  • the first parameter includes a first transformation coefficient, and the first transformation coefficient is used to represent a dimension ratio between the first data and the first transmission symbol; or, the The first parameter includes a first modulation order and a first coding rate.
  • the first accuracy is represented by any one or more of the following: the confidence degree of the processing result of the neural network; the classification accuracy of the processing result of the neural network; The mean square error MSE between the input data of the network and the first data; the mean absolute error MAE between the input data of the neural network and the first data.
  • the second communication device processes the first transmission symbol according to the first parameter, and obtaining the input of the neural network includes: the second communication device processes the first transmission symbol according to the first transformation coefficient performing an inverse transformation on the first transmission symbol to obtain an input of the neural network; or, the second communication device performs an inverse transformation on the first transmission symbol according to the first modulation order and the first coding rate demodulation and decoding to obtain the input of the neural network.
  • the method further includes: the second communication device receives a first request message from the first communication device, the first request message is used to request the neural network to process, The first request message includes first indication information, and the first indication information is used to indicate the first accuracy; the second communication device sends a first response message to the first communication device, and the first A response message includes second indication information, where the second indication information is used to indicate the first parameter.
  • the method further includes: the second communication device sends a second request message to the first communication device, the second request message is used to request data, and the second request The message includes third indication information, where the third indication information is used to indicate the first parameter.
  • the method further includes: when the second accuracy of the input processing result satisfies a preset condition, the second communication device sends feedback to the first communication device message, the feedback message includes information for indicating the processing result.
  • the method further includes: when the second accuracy of the input processing result does not meet a preset condition, the second communication device sends to the first communication device Retransmission indication information, where the retransmission indication information is used to indicate retransmission of the first data, and the first transmission symbol is obtained according to the first data.
  • the retransmission indication information is further used to indicate a second parameter, where the second parameter is an updated parameter of the first parameter.
  • one or more of the following information is carried in the neural network processing control channel: the first indication information, the second indication information, the third indication information or the retransmission Instructions.
  • the first transmission symbol is carried on a neural network processing shared channel.
  • the embodiment of the present application provides a communication device, configured to execute the method in the first aspect or any possible implementation manner of the first aspect.
  • the communication device includes a corresponding unit for performing the method in the first aspect or any possible implementation manner of the first aspect.
  • the communication device may be a first communication device or a chip, and the chip may be applied (or referred to as being set) in the first communication device.
  • the embodiment of the present application provides a communication device, configured to execute the method in the second aspect or any possible implementation manner of the second aspect.
  • the communication device includes a corresponding method for performing the method in the second aspect or any possible implementation manner of the second aspect.
  • the communication device may be a second communication device or a chip, and the chip may be applied (or referred to as being set) in the second communication device.
  • the above communication device may include a transceiver unit and a processing unit.
  • a transceiver unit and a processing unit For the specific description of the transceiver unit and the processing unit, reference may also be made to the device embodiments shown below.
  • an embodiment of the present application provides a communication device, where the communication device includes a processor, configured to execute the method described in the first aspect or any possible implementation manner of the first aspect.
  • the processor is used to execute a program stored in the memory, and when the program is executed, the method shown in the first aspect or any possible implementation manner of the first aspect is executed.
  • the sending process in the above method can be understood as a process output by the processor.
  • the processor when a processor outputs data, the processor outputs the data to a transceiver for transmission by the transceiver. After the data is output by the processor, additional processing may be required before reaching the transceiver.
  • a processor receives incoming data, a transceiver receives that data and inputs it to the processor.
  • the data may require other processing before being input to the processor. It can be understood that with respect to this description, the sixth aspect shown below is also applicable.
  • the above-mentioned processor may be a processor dedicated to performing these methods, or may be a processor that executes computer instructions in a memory to perform these methods, such as a general-purpose processor.
  • the above-mentioned memory can be a non-transitory (non-transitory) memory, such as a read-only memory (read only memory, ROM), which can be integrated with the processor on the same chip, or can be respectively arranged on different chips.
  • ROM read-only memory
  • the embodiment does not limit the type of the memory and the arrangement of the memory and the processor. It can be understood that the description of the processor and the memory is also applicable to the sixth aspect shown below, and will not be described in detail below.
  • the memory is located outside the communication device.
  • the memory is located in the above communication device.
  • the processor and the memory may also be integrated into one device, that is, the processor and the memory may also be integrated together.
  • the communication device further includes a transceiver, where the transceiver is configured to receive a signal or send a signal.
  • the communication device may be a first communication device or a chip, and the chip may be applied in the first communication device.
  • an embodiment of the present application provides a communication device, where the communication device includes a processor configured to execute the method described in the second aspect or any possible implementation manner of the second aspect.
  • the processor is used to execute the program stored in the memory, and when the program is executed, the method shown in the above second aspect or any possible implementation manner of the second aspect is executed.
  • the memory is located outside the communication device.
  • the memory is located in the above communication device.
  • the processor and the memory may also be integrated into one device, that is, the processor and the memory may also be integrated together.
  • the communication device further includes a transceiver, where the transceiver is configured to receive a signal or send a signal.
  • the communication device may be a second communication device or a chip, and the chip may be applied in the second communication device.
  • the embodiment of the present application provides a communication device, the communication device includes a logic circuit and an interface, the logic circuit is coupled to the interface; the interface is used to input the first data; the logic circuit is used to The first transmission symbol is obtained by processing the first data according to the first parameter; the interface is further configured to output the first transmission symbol.
  • the embodiment of the present application provides a communication device, the communication device includes a logic circuit and an interface, the logic circuit is coupled to the interface; the interface is used to input a first transmission symbol; the logic circuit, It is used for processing the first transmission symbol according to the first parameter to obtain an input of a neural network; and processing the input according to the neural network to obtain a processing result.
  • the embodiment of the present application provides a computer-readable storage medium, which is used to store a computer program, and when it is run on a computer, any of the above-mentioned first aspect or the first aspect is possible The method shown in the implementation is executed.
  • the embodiment of the present application provides a computer-readable storage medium, which is used to store a computer program, and when it is run on a computer, it makes possible any of the above-mentioned second aspect or the second aspect.
  • the method shown in the implementation is executed.
  • the embodiment of the present application provides a computer program product
  • the computer program product includes a computer program or a computer-executable instruction, and when it is run on a computer, any of the above-mentioned first aspect or the first aspect is possible
  • the method shown in the implementation is executed.
  • the embodiment of the present application provides a computer program product
  • the computer program product includes a computer program or a computer-executable instruction, and when it is run on a computer, any of the above-mentioned second aspect or the second aspect is possible
  • the method shown in the implementation is executed.
  • an embodiment of the present application provides a computer program.
  • the computer program When the computer program is run on a computer, the method shown in the above-mentioned first aspect or any possible implementation manner of the first aspect is executed.
  • an embodiment of the present application provides a computer program.
  • the computer program When the computer program is run on a computer, the method shown in the second aspect or any possible implementation manner of the second aspect is executed.
  • the embodiment of the present application provides a communication system, the communication system includes a first communication device and a second communication device, and the first communication device is used to implement the above first aspect or any possible method of the first aspect The method shown in the implementation manner, the second communication device is configured to execute the method shown in the second aspect or any possible implementation manner of the second aspect.
  • FIG. 1 is a schematic diagram of a communication system provided by an embodiment of the present application.
  • FIG. 2 is an interactive schematic diagram of a data processing method provided by an embodiment of the present application
  • FIG. 3 is a schematic diagram of the relationship between a first accuracy and a first signal-to-noise ratio provided by an embodiment of the present application;
  • FIG. 4 to FIG. 6 are interactive schematic diagrams of a data processing method provided by an embodiment of the present application.
  • FIGS. 7a to 7d are schematic diagrams of a frame structure provided by an embodiment of the present application.
  • Fig. 8a is a schematic diagram of channels corresponding to neural network processing provided by the embodiment of the present application.
  • Fig. 8b is a schematic diagram of layers corresponding to neural network processing provided by the embodiment of the present application.
  • Fig. 8c is a schematic diagram of channels corresponding to neural network processing provided by the embodiment of the present application.
  • FIG. 9a to FIG. 9c are schematic diagrams of scenarios of a data processing method provided by an embodiment of the present application.
  • FIG. 10 is a schematic diagram of a simulation result provided by an embodiment of the present application.
  • 11 to 13 are schematic structural diagrams of a communication device provided by an embodiment of the present application.
  • an embodiment means that a particular feature, structure, or characteristic described in connection with the embodiment may be included in at least one embodiment of the present application.
  • the occurrences of this phrase in various places in the specification are not necessarily all referring to the same embodiment, nor are separate or alternative embodiments mutually exclusive of other embodiments. It is understood explicitly and implicitly by those skilled in the art that the embodiments described herein can be combined with other embodiments.
  • At least one (item) means one or more
  • “multiple” means two or more
  • “at least two (items)” means two or three and three
  • “and/or” is used to describe the association relationship of associated objects, which means that there can be three kinds of relationships, for example, "A and/or B” can mean: only A exists, only B exists, and A and B exist at the same time A case where A and B can be singular or plural.
  • the character “/” generally indicates that the contextual objects are an "or” relationship.
  • “At least one of the following” or similar expressions refer to any combination of these items. For example, at least one item (piece) of a, b or c can mean: a, b, c, "a and b", “a and c", “b and c", or "a and b and c ".
  • the method provided by this application can be applied to various communication systems, for example, it can be an Internet of Things (Internet of Things, IoT) system, a narrowband Internet of Things (NB-IoT) system, a long term evolution (long term evolution) , LTE) system, or a fifth-generation (5th-generation, 5G) communication system, and a new communication system (such as 6G) that will appear in future communication development.
  • IoT Internet of Things
  • NB-IoT narrowband Internet of Things
  • LTE long term evolution
  • 5th-generation, 5G fifth-generation
  • 6G new communication system
  • the method provided in this application can also be applied to a wireless local area network (wireless local area network, WLAN) system, such as wireless-fidelity (wireless-fidelity, Wi-Fi) and the like.
  • WLAN wireless local area network
  • the technical solution provided by this application can also be applied to machine type communication (machine type communication, MTC), inter-machine communication long term evolution technology (long term evolution-machine, LTE-M), device-to-device (device-to-device, D2D) network , machine to machine (machine to machine, M2M) network, Internet of things (internet of things, IoT) network or other networks.
  • MTC machine type communication
  • LTE-M long term evolution-machine
  • D2D device-to-device
  • M2M machine to machine
  • IoT Internet of things
  • the IoT network may include, for example, the Internet of Vehicles.
  • V2X vehicle-to-everything
  • X can represent anything
  • the V2X can include: vehicle-to-vehicle (V2V) communication, Vehicle to infrastructure (V2I) communication, vehicle to pedestrian (V2P) or vehicle to network (V2N) communication, etc.
  • V2V vehicle-to-vehicle
  • V2I Vehicle to infrastructure
  • V2P vehicle to pedestrian
  • V2N vehicle to network
  • terminal devices may communicate with each other through D2D technology, M2M technology, or V2X technology.
  • Fig. 1 is a schematic diagram of a communication system provided by an embodiment of the present application. The method embodiments shown below in this application may be applicable to the communication system shown in FIG. 1 , and will not be described in detail below.
  • the communication system may include at least one access network device and at least one terminal device.
  • the access network device may be a next generation node B (next generation node B, gNB), a next generation evolved base station (next generation evolved nodeB, ng-eNB) (may be referred to as eNB for short), or a access network equipment, etc.
  • the access network device may be any device with a wireless transceiver function, including but not limited to the above-mentioned base station.
  • the base station may also be a base station in a future communication system such as a sixth generation communication system.
  • the access network device may be an access node, a wireless relay node, a wireless backhaul node, etc. in a wireless local area network (wireless fidelity, WiFi) system.
  • the access network device may be a wireless controller in a cloud radio access network (cloud radio access network, CRAN) scenario.
  • the access network device may be a wearable device or a vehicle-mounted device.
  • the access network device may also be a small cell, a transmission reception point (transmission reception point, TRP) (or may also be called a transmission point or a reception point), and the like. It can be understood that the access network device may also be a base station in a future evolving public land mobile network (public land mobile network, PLMN), etc.
  • PLMN public land mobile network
  • the terminal equipment may also be called user equipment (user equipment, UE), terminal, and so on.
  • a terminal device is a device with wireless transceiver function, which can be deployed on land, including indoor or outdoor, handheld, wearable or vehicle-mounted; it can also be deployed on water, such as on a ship; it can also be deployed in the air, such as on a Airplanes, balloons, or satellites, etc.
  • Terminal equipment can be mobile phone, tablet computer (Pad), computer with wireless transceiver function, virtual reality (virtual reality, VR) terminal equipment, augmented reality (augmented reality, AR) terminal equipment, industrial control (industrial control) ), wireless terminals in self driving, wireless terminals in remote medical, wireless terminals in smart grid, wireless terminals in transportation safety , wireless terminals in a smart city, wireless terminals in a smart home, etc.
  • the terminal device may also be a terminal device in a future 6G network or a terminal device in a future evolved PLMN.
  • the terminal equipment shown in this application may not only include vehicles (such as complete vehicles) in the Internet of Vehicles, but also include vehicle-mounted devices or vehicle-mounted terminals (including telematics boxes, T-boxes) in the Internet of Vehicles. or the host in the Internet of Vehicles system), etc., the present application does not limit the specific form of the terminal device when it is applied to the Internet of Vehicles.
  • the method involved in this application will be introduced below by taking the terminal device as an example.
  • the communication system shown in FIG. 1 includes one access network device and four UEs, such as UE1 to UE4 in FIG. 1 .
  • UE1, UE3, and UE4 may be mobile phones, and UE2 may be a car, etc.; or, UE3 and UE4 are cars, and UE1 and UE2 are mobile phones, etc.
  • FIG. 1 exemplarily shows one access network device, four UEs, and communication links between the communication devices.
  • the communication system may include multiple access network devices, and the coverage of each access network device may include other numbers of UEs, such as more or fewer UEs, etc.
  • This application does not make any limited.
  • the communication system shown in FIG. 1 may also include core network equipment, such as an access and mobility management function (access and mobility management function, AMF), etc., which is not limited in this application.
  • AMF access and mobility management function
  • reliable transmission is used as the design criterion in wireless transmission design. For example, if a first communication device sends data to a second communication device, if the second communication device cannot correctly decode the received data, the first communication device needs to repeat transfer data. After the second communication device decodes correctly, the second communication device performs neural network (neural network, NN) processing on the received data. However, since the neural network has a good tolerance to errors, the second communication device may not need to receive data completely correctly to perform neural network processing. That is to say, the aforementioned method that the second communication device needs to completely and correctly receive the data from the first communication device before performing the neural network processing will lead to a low utilization rate of wireless resources.
  • neural network neural network
  • the present application provides a data transmission method and device, which uses the accuracy of the data processed by the neural network in the second communication device as the index of transmission design, so as to determine the relevant parameters of the data processed by the first communication device. Furthermore, it can make full use of the trade-off between data transmission errors under different channel information and the accuracy of neural network processing data (such as the lower the signal-to-noise ratio, the lower the accuracy, so the requirement for wireless resources can be reduced by reducing the accuracy), The utilization efficiency of wireless resources is effectively improved.
  • the first accuracy shown in this application can be used to represent the accuracy of neural network processing data. Further, the first accuracy may be used to represent an average of the accuracy of the processing results output by the neural network. Alternatively, the first accuracy can be used to represent the average performance of the neural network on processing data. For example, the average performance includes average correct rate, average accuracy or average accuracy rate, etc.
  • the output of the neural network includes processing results.
  • the processing result may correspond to the processing type of the neural network.
  • the processing result may be the category corresponding to the input of the neural network.
  • the processing type of the neural network includes pattern recognition, the processing result may be a pattern corresponding to the input of the neural network.
  • the output of the neural network also includes the accuracy of the processing result obtained by the neural network, which may also be referred to as the accuracy corresponding to the processing result.
  • the output may be the probability that the input belongs to each category, and the category corresponding to the highest probability is used as the processing result of the input.
  • the accuracy of the processing result can be expressed by the probability of the category corresponding to the output of the neural network.
  • the probability of the predicted output of the neural network may also be a modified probability, such as by modifying the original output probability, so that the modified probability is close to the accuracy. That is, the first accuracy can be represented by a modified probability.
  • the output of the neural network may also include classification accuracy or confidence. Still taking the classification task as an example, if the probability distribution of the neural network output is relatively concentrated (or the entropy is small), it means that the confidence of the output is high. Classification accuracy indicates the proportion of correct results in the output.
  • the first accuracy can be represented by an average of confidence levels corresponding to multiple processing results of the neural network.
  • the first accuracy may be represented by an average of probabilities corresponding to multiple processing results of the neural network.
  • the first accuracy may be represented by an average of classification accuracy corresponding to multiple processing results of the neural network. It can be understood that the confidence, classification accuracy or probability shown above are only examples, and the output of the neural network may also include other types, such as recall rate (recall).
  • the output of the neural network shown above can be understood as an example of a discrete output, however, the output of the neural network in this application may also include other types of output.
  • the output of the neural network can be the result of channel estimation.
  • the output of the neural network may also be a numerical operation result or the like.
  • the output of the neural network may also be the result of parameter fusion.
  • the first accuracy may be represented by an average of errors between data input by the neural network multiple times and real data. That is to say, the first accuracy can be represented by an average of errors of the transmitted data.
  • the error includes the deviation between the received gradient of the neural network (ie, the input obtained by the neural network) and the real gradient (ie, the data sent by the transmitter).
  • errors include the deviation between fusion parameters and original parameters in federated learning.
  • the error may include any one or more items of mean square error (mean squared error, MSE), root mean square error (root mean square error, RMSE) or mean absolute error (mean absolute error, MAE).
  • MSE mean square error
  • RMSE root mean square error
  • MAE mean absolute error
  • the first accuracy can be expressed by the average MSE between the data input by the neural network multiple times and the real data, or by the average RMSE between the data input by the neural network multiple times and the real data, or , the average representation of the MAE between the data input by the neural network multiple times and the real data, etc.
  • the first accuracy shown here can also be understood as being represented by a distance in a feature space.
  • the distance of the feature space is the distance between features, which can be expressed by any one or more of MSE, RMSE or MAE.
  • the second accuracy shown in this application indicates the accuracy of the processing result output by the neural network.
  • the second accuracy can be expressed by the confidence of the processing result, or the second accuracy can be expressed by the probability (or probability entropy) of the processing result, or, the second accuracy The degree can be expressed by the classification accuracy of the processing results, etc.
  • the second accuracy may be represented by an error between the input data of the neural network and the real data, and the error includes one or more of MSE, RMSE, or MSE.
  • the second accuracy may represent the accuracy of the processing result output by the neural network for a single time.
  • the first accuracy may represent the accuracy of the processing results output by the neural network multiple times.
  • FIG. 2 is a schematic flowchart of a data processing method provided by an embodiment of the present application.
  • the method can be applied to the first communication device and the second communication device.
  • the second communication device may perform neural network processing on data from the first communication device. That is to say, the first communication device may provide data to the second communication device, and the second communication device processes the data provided by the first communication device according to the neural network.
  • the first communication device may be a terminal device
  • the second communication device may be an access network device.
  • the first communication device is an access network device
  • the second communication device is a terminal device.
  • both the first communication device and the second communication device may be terminal devices, etc., and the embodiment of the present application does not limit specific forms of the first communication device and the second communication device.
  • the method includes:
  • the first communication device obtains first data.
  • the first data may be data to be processed by the first communication device.
  • the first data may be data to be sent to the second communication device or the like.
  • the first data may include raw data obtained by the first communication device.
  • the neural network in the second communication device includes model forward reasoning.
  • the first data may include any type of image data, voice data, or text data.
  • a complete neural network can also be cut at a certain layer and divided into two parts of the neural network, such as the first part of the neural network and the second part of the neural network, so as to be deployed in the first communication device and the second communication device respectively. on the device.
  • the first communication device may use the first partial neural network to partially process the input data to obtain an intermediate result, and then send the intermediate result to the second communication device. After the second communication device obtains the intermediate result, it may continue to process the intermediate result by using the second part of the neural network to obtain an inference result.
  • the neural network in the second communication device may include a partial layer neural network, so as to be used for model forward reasoning.
  • the first data may include data obtained by the first communication device and processed by partial layers of the neural network.
  • the second communication device may serve as a parameter server for federated learning, so as to fuse parameters such as weights updated by neural networks of multiple first communication devices. That is to say, the neural network in the second communication device includes federated learning model fusion, that is, the first data may include model parameters or intermediate gradient data obtained by the first communication device.
  • the specific type of the first data may be determined according to the processing type of the neural network in the second communication device, and this embodiment of the present application does not limit the specific type of the first data.
  • all data that can be processed by the neural network in the second communication device falls within the protection scope of the embodiments of the present application.
  • the embodiment of the present application does not limit the specific method for the first communication device to obtain the first data.
  • the first communication device processes the first data according to a first parameter to obtain a first transmission symbol, where the first parameter is determined according to a first accuracy and first channel information.
  • the first transmission symbol may be understood as a symbol that can be used for transmission and obtained after the first communication device processes the first data.
  • the first transmission symbol may include a modulation symbol, a coded symbol, or a coded modulation symbol, and the like.
  • the embodiment of the present application does not limit the specific type of the first transmission symbol.
  • the first parameter can be understood as a parameter used when the first communication device processes data (such as the first data shown in this application).
  • the first parameter may also be called a first transmission parameter or a first processing parameter, etc., which is not limited in this embodiment of the present application.
  • the first parameter may include a first transformation coefficient (for example, reference may be made to the method shown in FIG. 4 below).
  • the first communication device codes and modulates the first data, then the first parameter may include a first modulation order and a first coding rate (for example, refer to the method shown in FIG. 5 below).
  • the first parameter may include a first modulation order.
  • the first parameter may include a first encoding rate.
  • the first channel information is channel information between the first communication device and the second communication device.
  • the first channel information may be channel state information (channel state information, CSI) between the first communication device and the second communication device.
  • the first channel information may be information indicating channel quality between the first communication device and the second communication device.
  • the first parameter shown in the embodiment of the present application is determined according to the first accuracy and the first channel information, which can also be understood as: the first parameter is related to the first accuracy and the first channel information, or the first parameter is determined according to the first An accuracy and a first channel information are obtained. Determining the parameters (such as the first parameter) used by the first communication device when processing data according to the first accuracy and the first channel information may enable the first communication device to process data according to the requirements of the neural network for processing data. Therefore, the first data is processed by making full use of the accuracy requirements of the neural network to process data and the channel quality, thereby improving the utilization rate of wireless resources.
  • the first communication device may relax the data processing requirement (for example, fewer effective features are retained) when processing according to the transformation coefficient. It can be understood that for the relationship among the first parameter, the first accuracy requirement, and the first channel information, reference may also be made to the following (such as related descriptions in Table 1 to Table 9, etc.), and details will not be detailed here.
  • the first communication device may directly process the first data according to the first parameter.
  • the method shown in FIG. 2 further includes: the first communication device embeds the first data into a continuous space.
  • the above steps can also be understood as: embedding the first data into the continuous space according to the type of the first data; or, according to the type of neural network processing in the second communication device, the data to be sent to the second communication device (ie the first data) embedded in the continuous space.
  • the embedding operation may not be performed.
  • raw data such as when the first data includes image data, voice data or wireless data (such as channel information)
  • voice data or wireless data such as channel information
  • the first data itself is in the form of a continuous space (such as image data or voice data, etc. )
  • you don’t need to embed it.
  • the first data includes text data or other discrete data (for example, its value is not in the form of numbers, but in the form of letters or characters), it is necessary to embed the first data in a continuous space.
  • the original data can be embedded into a space that reflects the semantic distance of the input according to the tasks processed by the neural network (such as natural language processing tasks, etc.), such as the discrete space of language is embedded into the continuous space through learning, and the distance of the continuous space is It can reflect the semantic distance of words.
  • the neural network such as natural language processing tasks, etc.
  • the first communication device may perform an embedding operation through a neural network, which is not limited in this embodiment of the present application.
  • the embedding operation can be implemented through a pre-trained network, or through a pre-trained mapping relationship.
  • the first communication device may map words (words) to real number vectors (eg, each word may correspond to a vector, etc.).
  • the first communication device sends the first transmission symbol to the second communication device, and correspondingly, the second communication device receives the first transmission symbol.
  • the first communication device may also perform other processing, which is not limited in this embodiment of the present application.
  • the first communication device may map the first transmission symbol to a transmission resource (such as a time-frequency resource), and send the first transmission symbol through the transmission resource.
  • the first communication device may perform power allocation, such as sending the first transmission symbol with a certain transmission power.
  • power allocation is mainly to deal with fading channel scenarios, so as to realize channel whitening or feature extraction.
  • channel whitening may be understood as that the first communication device allocates power according to different fading degrees of subcarriers, so as to ensure that the second communication device sees the same fading.
  • Feature extraction can be understood as a weighted convolution operation through power distribution, such as feature extraction can be a layer in a convolutional neural network. It can be understood that specific descriptions about channel whitening and feature extraction are not limited in this embodiment of the present application.
  • the second communication device processes the first transmission symbol according to the first parameter to obtain an input of the neural network.
  • the second communication device may inversely process the first transmission symbol according to the first parameter to obtain data conforming to the input of the neural network.
  • the input in the above step 204 can be understood as the data conforming to the input of the neural network.
  • the second communication device performs inverse transformation on the first transmission symbol according to the first transformation coefficient, for example, the inverse transformation may include linear transformation or nonlinear transformation.
  • the second communication device may demodulate the first transmission symbol according to the first modulation order.
  • the second communication device may perform decoding processing on the first transmission symbol according to the first coding rate.
  • the second communication device may demodulate and decode the first transmission symbol according to the first modulation order and the first coding rate.
  • the second communication device processes the input according to the neural network to obtain a processing result.
  • the second communication device may also obtain the accuracy of the processing result.
  • the second communication device may perform an inverse embedding operation through a neural network, or, after obtaining a processing result through a neural network, the second communication device may perform
  • the inverse embedding operation and the like are not limited in this embodiment of the present application.
  • the second communication device may map the real number vectors output by the neural network to words or the like.
  • the first communication device embeds the graph structure into the real number vector space through graph embedding, and correspondingly, the second communication device can recover the graph structure from the real number vector space.
  • the first communication device may process the first data according to the first parameter, where the first parameter is determined according to the first accuracy and the first channel information. Therefore, when the first communication device transmits data, the processing of the data only needs to comply with the first accuracy and the first channel information, and the second communication device can perform neural network processing on the data, which improves the accuracy of data transmission due to the need to ensure correctness. In the case of retransmitting data due to certain characteristics, the utilization efficiency of wireless resources is effectively improved.
  • the first parameter includes a first transform coefficient representing a dimensional ratio between the first data and the first transmission symbol.
  • the number of elements of the first data and the number of elements of the first transmission symbol may be used to represent the dimension ratio between the first data and the first transmission symbol.
  • the form of the first data may include a vector or a matrix, etc., which is not limited in this embodiment of the present application.
  • the first communication device processing the first data includes the first communication device performing transformation processing on the first data, and the transformation includes linear transformation, such as discrete Fourier transform or discrete cosine transform, etc., or includes neural network conversion, etc., which are not limited in this embodiment of the present application.
  • the dimension adaptation of the first data and the first transmission symbol can be realized, and the dimension adaptation can be understood as being determined according to the requirements of the first accuracy of the neural network processing data and the dimension of the first data. The number of transmitted symbols.
  • the relationship between the first transform coefficient, the first accuracy, and the first channel information may be as follows:
  • a larger first transformation coefficient indicates that the first communication device retains fewer effective features when processing the first data. Therefore, the less transmission resources used when transmitting the first transmission symbol, the lower the first accuracy of the neural network for processing data.
  • the first channel information is the same (for example, the channel quality is the same)
  • the smaller the first transform coefficient is the higher the requirement for the first accuracy is.
  • the better the channel quality represented by the first channel information the larger the first transformation coefficient.
  • wireless transmission is aimed at maximizing throughput, and corresponding parameters are set according to reliability requirements, such as channel quality indication (CQI) or modulation and coding scheme (MCS).
  • CQI channel quality indication
  • MCS modulation and coding scheme
  • the method for setting the above parameters may include the channel quality (such as the signal-to-noise ratio (signal-to-noise ratio, SNR) corresponding to the MCS under the condition that the block error rate (block error rate, BLER) is equal to 0.1 or 0.001.
  • SNR signal-to-noise ratio
  • BLER block error rate
  • the embodiments of the present application can support dynamic first transform coefficients and first accuracy in a wide range of channel quality intervals.
  • the relationship among the first transform coefficient, the first accuracy, and the first channel information may be expressed in the form of a curve.
  • Fig. 3 is a schematic diagram of the relationship between the first accuracy and the first signal-to-noise ratio provided by the embodiment of the present application.
  • the neural network including reasoning it can be seen from the two curves shown in FIG. 3 that as the first signal-to-noise ratio increases, the first accuracy of reasoning increases. It can be seen from the lines parallel to the horizontal axis shown in FIG.
  • Corresponding transform coefficients can be selected under corresponding signal-to-noise ratio conditions to meet accuracy requirements.
  • the current Under the condition of signal-to-noise ratio the balance between accuracy and transmission resources is achieved by controlling the transformation coefficient. It can be understood that the unit of the horizontal axis shown in FIG. 3 may be dB, and the unit of the vertical axis may be percentage.
  • the relationship among the first transform coefficient, the first accuracy, and the first channel information may be expressed in a function form.
  • the first transform coefficient f(first accuracy, first channel information).
  • the relationship among the first transform coefficient, the first accuracy, and the first channel information may be represented in a table form.
  • Table 1 shows the relationship between the first transform coefficient and the first channel information when the first accuracy is equal to 99%.
  • Table 2 shows the relationship between the first transform coefficient and the first channel information when the first accuracy is equal to 90%.
  • Table 3 shows the relationship between the first transform coefficient and the first channel information when the first accuracy is equal to 80%.
  • SNR signal to interference plus noise ratio
  • SINR signal to interference plus noise ratio
  • the first transformation coefficient can be determined according to different types or stages of neural network processing.
  • the type of neural network processing may include radar perception or automatic driving analysis, etc.
  • the first conversion coefficient may be determined according to a relatively high first accuracy and first channel information.
  • the type of neural network processing includes image recognition, foreign object detection, or machine translation, etc. In this case, it can be based on the lower first accuracy (that is, it can be understood that there is no strict requirement for accuracy) and transmission resources, etc.
  • a first transform coefficient is determined.
  • there may be large errors in the early stage of neural network training so the table with the first lower accuracy can be selected to improve the utilization efficiency of wireless resources.
  • a more refined parameter update is required in the later stage of neural network training, so the first table with higher accuracy can be selected.
  • Tables 1 to 3 are shown as examples with the first accuracy being 99%, 90% and 80% respectively.
  • the corresponding table may also be selected according to the method of upward alignment required by the actual accuracy.
  • Table 1 may be selected according to the principle of upward alignment.
  • Table 2 can be selected according to the principle of upward alignment.
  • the corresponding table may also be selected according to the method of downward alignment required by the actual accuracy.
  • Table 1 may correspond to an accuracy of 99% or above
  • Table 2 may correspond to an accuracy of 90% to 99%
  • Table 3 may correspond to an accuracy of 80% to 89%, and so on.
  • the specific relationships shown in Table 1 to Table 3 are only examples, and in a specific implementation, the relationship between the first transform coefficient, the first accuracy requirement, and the first channel information is not limited thereto.
  • the first accuracy may also include 95%, 80% or 70% and so on.
  • the first accuracy may also be flexible, and in this case, the first transformation coefficient may be determined according to the first channel information.
  • the relationship between the first transform coefficient, the first accuracy and the first channel information may also be as shown in Table 4.
  • Table 4 shows optional first conversion coefficients for each channel quality, and different first conversion coefficients may correspond to different first accuracies of data processed by the neural network.
  • a flexible first accuracy may be selected according to the usage of transmission resources or the requirement of time delay, so as to determine the first transform coefficient. For example, if the time delay is strictly required (for example, the time delay is required to be small), the first transformation coefficient may be relatively large in order to process the data in a more timely manner. For another example, if the time delay requirement is not strict (for example, the time delay may be relatively large), the first transform coefficient may be relatively small. It can be understood that the time delay shown here may represent the processing and feedback time delay of the neural network.
  • the table may be formulated with reference to the relationship between the first accuracy, the first channel information, and the first transform coefficient.
  • the first transform coefficients may be 1/8, 1/6, and 1/4 in sequence, respectively.
  • the corresponding first transform coefficients may be 1/8, 1/6, and 1/4.
  • the first transformation coefficient may be selected according to the usage of the transmission resource. If there are fewer transmission resources, the first transform coefficient may be larger, such as 1/6 or 1/4.
  • the first transformation coefficient and the like are determined according to the time delay requirement, which will not be described in detail here.
  • the relationship among the first transform coefficient, the first accuracy, and the first channel information may also be expressed in a form of a table combined with a formula.
  • the relationship between the first transform coefficient, the first accuracy and the first channel information may be represented by a basic table and a correction formula.
  • the representation forms among the first transform coefficient, the first accuracy and the first channel information shown above are only examples.
  • the first accuracy shown in the embodiment of the present application may be expressed by confidence, classification accuracy, probability, etc., and may also be expressed by any one or more of MSE, RMSE, or MAE.
  • the first parameter includes a first coding rate.
  • the first parameter includes a first modulation order.
  • the first parameter includes a first modulation order and a first coding rate.
  • the relationship between the first encoding rate and the first accuracy may be as follows: for example, the greater the first encoding rate, the higher the first accuracy.
  • the first channel information is the same (for example, the channel quality is the same)
  • the higher the first coding rate the higher the first accuracy.
  • the better the channel quality indicated by the first channel information the smaller the first coding rate.
  • the relationship among the first coding rate, the first accuracy and the first channel information may be expressed in the form of a function.
  • the first coding rate f 8 (first accuracy, first channel information).
  • the relationship between the first coding rate, the first accuracy, and the first channel information may be represented in a table form, and so on.
  • the relationship between the first coding rate, the first accuracy and the first channel information please refer to the relationship between the first coding rate, the first modulation order, the first accuracy and the first channel information shown below Relationship.
  • the relationship among the first coding rate, the first modulation order, and the first accuracy requirement may be expressed in the form of a function.
  • (first modulation order, first coding rate) f 9 (first accuracy, first channel information).
  • the relationship between the first coding rate, the first modulation order, the first accuracy, and the first channel information can be expressed in the form of a table, as shown in Table 5 and Table 6 .
  • Table 5 shows the relationship between the first coding rate, the first modulation order and the first channel information (such as represented by SNR) when the first accuracy is equal to 99%.
  • Table 6 shows the relationship between the first coding rate, the first modulation order and the first channel information when the first accuracy is equal to 90%.
  • the relationship between the first coding rate, the first modulation order, the first accuracy and the first channel information can also be as shown in Table 7 and Table 8 shown.
  • first modulation order 2 2 2 4 4 4 6 ... the first code rate 1/2 2/3 3/4 1/2 2/3 3/4 2/3 the the first SNR -4 -3 -2 -1 0 1 2 ... the
  • the relationship between the first modulation order, the first coding rate, the first accuracy, and the first channel information may also be as shown in Table 9.
  • the table shown in Table 9 shows the first modulation order and the first coding rate corresponding to the lowest acceptable first accuracy under each channel quality. Therefore, the first modulation order and the first coding rate not lower than those shown in Table 9 can be selected according to conditions such as transmission resource usage and time delay requirements. Or, for services with flexible accuracy, only the level indication of channel quality, the first modulation order and the first coding rate corresponding to the distance of the lowest acceptable feature space (such as MSE, etc.) may be defined.
  • general transmission parameter tables (such as tables corresponding to MCS, etc.) can be reused to provide reliable transmission services.
  • the tables shown in this application are only examples, and the tables shown in this application should not be construed as limiting the application.
  • the first signal-to-noise ratio may not be included (it can also be understood that the first signal-to-noise ratio is only a reference when formulating the tables).
  • Table 1 to Table 4 may include the first transform coefficient and the corresponding index
  • Table 5 to Table 9 may include the first modulation order, the first coding rate and the corresponding index.
  • the embodiments of the present application also provide the following data transmission methods.
  • FIG. 4 is a schematic flowchart of a data processing method provided by an embodiment of the present application. As shown in Figure 4, the method includes:
  • the first communication device sends a first request message to a second communication device, where the first request message is used to request neural network processing.
  • the second communication device receives the first request message.
  • the method shown in FIG. 4 can be understood as that the first communication device requests the neural network in the second communication device to process data (it can also be referred to simply as a node requesting a neural network processing mode). That is, the first communication device provides data, and the second communication device provides neural network processing services.
  • the first communication device shown in the embodiment of the present application can also be called a first node or a sending end (that is, a device that provides data), and the second communication device can also be called a second node or a receiving end (that is, a data processing device).
  • devices for neural network processing For example, the first communication device may be a terminal device or the like, and the second communication device may be an access network device or the like.
  • the second communication device sends a first response message to the first communication device.
  • the first communication device receives the first response message.
  • the first request message includes first indication information, where the first indication information is used to indicate the first accuracy.
  • the first accuracy may be a requirement of the first communication device on the accuracy of data processed by the neural network, or an expected accuracy of the first communication device.
  • the first accuracy may include any one of 99%, 95%, 90%, 85%, 80%, 75%, 70%, or 65%, among others.
  • the first accuracy may also have other division manners, for example, the first accuracy includes any one of 90%, 80%, 70%, 60%, 50%, and so on.
  • the first indication information may also indicate the first accuracy by means of an index.
  • the index may be an index of a table as shown in Table 1 to Table 9 (that is, the first accuracy is indicated by the index of the table), or it may be another type of index, etc., which is not limited in the embodiment of the present application.
  • the first request message includes first indication information, where the first indication information is used to indicate the first parameter.
  • the first parameter may include a first transform coefficient, or include a first coding rate and a first modulation order, and the like.
  • the first indication information may indicate the first parameter by means of an index.
  • the first indication information is used to indicate the first index.
  • the first index may be an index shown in Table 1 to Table 9, such as any one of indexes 0 to 6 in Table 1, and so on. Alternatively, it may also be other types of indexes, etc., which are not limited in this embodiment of the present application.
  • indexes of the tables shown above in this application are only examples, and the indexes shown in Table 1 to Table 3 may also be continuous indexes, etc., which are not limited in this embodiment of the present application.
  • the indexes shown in Table 1 to Table 9 may be consecutive indexes or the like.
  • the indexes shown in Table 5 and Table 6 may be consecutive indexes or the like.
  • the first indication information is used to indicate the first parameter or the first index, it means that the first communication device has obtained the first channel information in advance.
  • the first response message includes confirmation indication information, where the confirmation indication information is used to confirm the first indication information.
  • the confirmation indication information is used to confirm the first accuracy indicated by the first communication device (or confirm that the first communication device uses the first accuracy to determine the first parameter).
  • the access network device may independently determine the first parameter according to the first accuracy according to the confirmation indication information fed back by the terminal device.
  • the confirmation indication information may be used to confirm that the first communication device uses the first parameter or the first index, and the like.
  • the confirmation indication information shown here may be a bit, and when the value of the bit is 1, it can be used to confirm that the first communication device can use the content indicated by the first indication information; the one bit When the value of the bit is 0, it means that the first communication device cannot use the content indicated by the first indication information.
  • the first response message may further include second indication information and the like.
  • the first response message includes second indication information, and when the first indication information is used to indicate the first accuracy, the second indication information may be used to indicate the first parameter.
  • the access network device may determine the first parameter according to the first accuracy sent by the terminal device. For example, if the access network device has obtained the first channel information in advance, the access network device may determine the first parameter according to the first channel information and the first accuracy, so as to send the first response message to the terminal device.
  • the first request message may further include a reference signal (reference signal, RS).
  • the access network device can perform channel estimation according to the reference signal, so as to obtain the first channel information. Then determine the first parameter according to the first channel information and the first accuracy.
  • the first response message includes second indication information, where the second indication information is used to indicate the second parameter.
  • the second communication device may re-determine the transformation coefficient or the coding bit rate according to its requirement on the accuracy of the neural network.
  • the access network device may re-determine the second parameter according to the first indication information sent by the terminal device.
  • the second parameter includes a second variation coefficient; or, includes a second modulation order, a second coding rate, and the like.
  • the above-mentioned first indication information, second indication information or acknowledgment indication information may be carried in control information
  • the control information may include downlink control information (downlink control information, DCI), uplink control information (uplink control information) , UCI) or any one or more items of sidelink control information (SCI).
  • the above-mentioned first indication information, second indication information, confirmation indication information, etc. may be carried in radio resource control (radio resource control, RRC) signaling.
  • the foregoing first indication information, second indication information, or confirmation indication information may be carried in artificial intelligence (artificial intelligence, AI) dedicated control information or on a dedicated control channel.
  • AI artificial intelligence
  • the first communication device and the second communication device may pre-negotiate specific types of parameters.
  • the type of parameter includes one or more items of transform coefficient, coding rate or modulation order.
  • the type of the parameter may be determined by the access network device.
  • data processing methods based on coded modulation can provide reliable transmission and achieve higher inference accuracy at high signal-to-noise ratios. However, the performance degrades faster at low signal-to-noise ratios, so it is more suitable for data processing methods based on direct modulation. Therefore, optionally, the first communication device or the second communication device may automatically switch the parameter type according to the magnitude of the channel information.
  • the parameter type when the channel quality is greater than the quality threshold, the parameter type includes one or more items of coding rate or modulation order. When the channel quality is less than the quality threshold, the parameter type includes transform coefficients. It can be understood that when the channel quality is equal to the quality threshold, the type of the parameter is not limited, for example, it can be set in advance, which is not limited in this embodiment of the present application.
  • the channel quality may include SNR or SINR, etc. If the channel quality includes SNR, the quality threshold may include any one of 5dB, 8dB, 10dB or 15dB.
  • the first communication device sends the first transmission symbol to the second communication device, and correspondingly, the second communication device receives the first transmission symbol.
  • the above step 403 can also be understood as: sending the first transmission symbol to the second communication device according to the first response message.
  • the first communication device may obtain the first transmission symbol according to the first response message, and then send the first transmission symbol and so on.
  • the second communication device processes the input according to the neural network, and obtains a processing result and a second accuracy of the processing result.
  • the second accuracy of the processing result may be represented by any one or more of the probability, confidence, classification accuracy, or MSE of the processing result. Therefore, for the convenience of description, the feedback method of the second communication device will be described below by taking the second accuracy represented by a probability and the preset condition represented by a preset probability as an example.
  • the second communication device feeds back the processing result to the first communication device.
  • the preset probability may be a numerical value. If the probability of the processing result is greater than or equal to the preset probability, it means that it meets the preset probability.
  • the preset probability may be a probability range, and if the probability of the processing result is within the probability range, it means that it satisfies the preset probability. It can be understood that the embodiment of the present application does not limit the specific manner of the preset probability.
  • the second communication device sends retransmission indication information to the first communication device, where the retransmission indication information is used to instruct retransmission of the first data.
  • the second communication device may also indicate a new parameter or an adjustment direction of the parameter, etc. to the first communication device.
  • the new parameter may be a second parameter (such as a second transformation coefficient, or a second coding rate and a second modulation order, etc.) and the like.
  • the adjustment direction of the parameter may be, for example, increasing the transform coefficient or decreasing the transform coefficient.
  • the second communication device may also indicate the new parameter through the index of the second parameter. It can be understood that after the first communication device acquires the new parameter, it can also retransmit the first data according to the new parameter.
  • the second communication device obtains the retransmitted data and obtains the processing result and the probability of the processing result again, when the difference between the probability obtained again and the probability obtained last time is less than a certain threshold, it can be considered that the retransmitted Data transmission does not improve the performance of the neural network, so retransmission can be terminated.
  • the above-mentioned preset probability can be set by the first communication device, or set by the second communication device, or negotiated by the first communication device and the second communication device, etc.
  • the specific setting of the preset probability in the embodiment of the present application The method is not limited.
  • the output of the neural network may also be the entropy and confidence of the processing results, which will not be described in detail here.
  • the first request message may include an RS, so that the second communication device may perform channel estimation.
  • the first response message may include an RS, so that the first communication device performs channel estimation and obtains channel information between the first communication device and the second communication device.
  • the RS may be included.
  • the first communication device retransmits the first data, it may also send the RS and the like at the same time, which is not limited in this embodiment of the present application.
  • the channel information obtained through the RS is updated, the first parameter may be updated accordingly.
  • the data processing when the first communication device transmits data, the data processing only needs to comply with the first accuracy and the first channel information, and the second communication device can perform neural network processing on the data, which improves due to the need to ensure The correctness of data transmission and the situation of retransmitting data effectively improve the utilization efficiency of wireless resources.
  • Fig. 5 is a schematic flow chart of a data processing method provided in the embodiment of the present application. As shown in Fig. 5, the method includes:
  • the second communication device sends a second request message to the first communication device, where the second request message is used to request data.
  • the first communication device receives the second request message.
  • the method shown in FIG. 5 can be understood as that the second communication device requests data from the first communication device for neural network processing (also referred to as the node request processing data mode for short). That is, the second communication device needs the data of the first communication device to be processed by the neural network.
  • the second communication device requests data from the first communication device for neural network processing (also referred to as the node request processing data mode for short). That is, the second communication device needs the data of the first communication device to be processed by the neural network.
  • the second request message includes third indication information, where the third indication information is used to indicate the first parameter.
  • the third indication information is used to indicate the first parameter.
  • the first communication device sends a first transmission symbol to a second communication device, and the second communication device receives the first transmission symbol.
  • the above step 502 can also be understood as: the first communication device sends the first transmission symbol to the second communication device according to the second request message.
  • the first communication device sends the first transmission symbol to the second communication device according to the second request message.
  • the second communication device processes the input according to the neural network, and outputs a processing result and a second accuracy of the processing result.
  • the second communication device feeds back the processing result to the first communication device.
  • the second communication device sends a retransmission indication to the first communication device, where the retransmission indication is used to instruct retransmission of the first data.
  • the data processing when the first communication device transmits data, the data processing only needs to comply with the first accuracy and the first channel information, and the second communication device can perform neural network processing on the data, which improves due to the need to ensure The correctness of data transmission and the situation of retransmitting data effectively improve the utilization efficiency of wireless resources.
  • Fig. 6 is a schematic flow chart of a data processing method provided by an embodiment of the present application. As shown in Fig. 6, the method includes:
  • the first communication device sends a third request message to the second communication device, where the third request message is used to request neural network processing.
  • the second communication device receives the third request message.
  • the third request message includes the first transmission symbol.
  • the third request message further includes a reference signal. That is to say, for the elastic accuracy service, the first communication device may send the first transmission symbol to the second communication device while initiating a neural network request.
  • the first communication device may also send a fourth request message to the second communication device, where the fourth request message is used to request neural network processing.
  • the fourth request message may include RS.
  • the second communication device After receiving the fourth request message, the second communication device sends a fourth response message to the first communication device, where the fourth response message may include information indicating the first parameter.
  • the third request message may include information for indicating the first parameter.
  • the third request message shown in FIG. 6 is shown as an example including the first transmission symbol and the information used to indicate the first parameter, but the content included in the third request message should not be understood as the implementation of the present application. Example limitations.
  • the second communication device processes the input according to the neural network, and outputs a processing result.
  • the second communication device may directly feed back processing results and the like to the first communication device.
  • the neural network may also output the second accuracy of the processing result.
  • the second communication device may directly feed back the processing result and the second accuracy to the first communication device.
  • this embodiment of the present application does not limit whether the neural network feeds back the retransmission indication according to the second accuracy.
  • the first communication device when transmitting data, may determine the first parameter autonomously, or the second communication device may determine the first parameter autonomously, so that the first communication device or the second communication device does not need to negotiate the first parameter with each other. parameter, which reduces the signaling overhead; moreover, the second communication device can perform neural network processing after receiving the first transmission symbol, which improves the situation of retransmitting data due to the need to ensure the correctness of data transmission, and effectively improves wireless resources. utilization efficiency.
  • the information used to carry the first parameter may also be referred to as second control information.
  • the first indication information in the first request message shown in FIG. 4 may be called second control information.
  • the second indication information in the first response message may also be called second control information.
  • the third indication information in the second request message may also be called second control information.
  • the information used to indicate the first parameter in the third request message may also be called second control information.
  • the information used to indicate the first parameter in the fourth request message may also be called second control information. Therefore, the present application also provides a structural schematic diagram of a frame structure.
  • the frame may include first control information and second control information.
  • the second control information shown in FIG. 7a can be understood as the first indication information sent by the first communication device to a second communication device, or the second instruction information sent by the second communication device to a first communication device.
  • the two second control information shown in FIG. 7b can be understood as the first instruction information sent by the first communication device to the two second communication devices respectively, or the information sent by the second communication device to the two first communication devices respectively. Second instruction message.
  • FIG. 7c or 7d shows a schematic diagram of a frame structure when control information and data are sent simultaneously.
  • the frame structure shown in Figure 7c or Figure 7d may be applicable to the scenario where the third request message shown above includes both the first transmission symbol and the information for indicating the first parameter.
  • the foregoing second control information may be used to indicate the first parameter or the first accuracy, and the like.
  • the first control information may be used to indicate resource locations occupied by data to be processed by the neural network.
  • the first control information may be used to indicate the time-frequency resource occupied by the first transmission symbol. That is, separate the traditional resource scheduling control from the neural network processing control shown in this application. That is, the first control information may be used to indicate that the content included in the frame is related parameters (such as the first parameter) and related data (such as the first transmission symbol) involved in the data processing method shown in this application.
  • the first control information may be used to indicate that the second control information is control information of traditional data transmission or control information of neural network processing.
  • the control information of traditional data transmission can be understood as downlink control information or uplink control information in wireless transmission design
  • the control information processed by neural network can be understood as the information processed by neural network related to the embodiment shown in this application, such as First parameter or first accuracy etc.
  • the separate first A control information indication is to separate resource scheduling control from neural network processing control.
  • the first control information in the header of the frame structure can be used to indicate the resource location occupied by the data processed by the neural network
  • the second control information is used to indicate related parameters. or accuracy information.
  • Another implementation manner is to multiplex the control information, for example, to indicate by marking that the content in the first control information is the control information of traditional data processing or the control information of neural network processing.
  • the embodiment of the present application also provides a transmission channel or a physical channel corresponding to neural network processing.
  • the transmission symbols (such as including the first transmission symbol) obtained according to the embodiment shown above in the present application may be carried on the transmission channel, such as for the first data directly
  • the modulated first transmission symbol may be carried on an uplink and downlink neural network shared channel (NNSCH) (also called a neural network shared channel).
  • NNSCH uplink and downlink neural network shared channel
  • the first transmission symbol may be carried on the uplink NNSCH; when the access network device sends the first transmission symbol to the terminal device, the first transmission symbol may be carried on the Downlink NNSCH.
  • the relevant parameters obtained according to the embodiments shown above in the present application such as the first parameter (or the second parameter, etc.) channel) on the transmission.
  • the NNCCH may include an uplink NNCCH and a downlink NNCCH.
  • the uplink mentioned above may refer to the terminal device sending information to the access network device
  • the downlink may refer to the access network device sending information to the terminal device.
  • the NNSCH shown above may not distinguish between uplink and downlink, and the NNCCH may not distinguish between uplink and downlink.
  • the aforementioned NNSCH may also be called a sidelink NNSCH
  • the NNCCH may also be called a sidelink NNCCH.
  • the NNSCH can not only be used to transmit data related to neural network processing, but also can be used to transmit neural network model data and the like.
  • the NNCCH can be used not only to transmit control information related to neural network processing, but also to transmit related parameters of the NNSCH or other signaling related to neural network processing, which is not limited in this embodiment of the present application.
  • NNCCH requires reliable transmission, it can be carried on a dedicated physical neural network control channel (PNNCCH), or it can be carried on a traditional data transmission physical channel, such as a physical downlink shared channel (physical downlink shared channel, PDSCH) or physical uplink shared channel (physical uplink shared channel, PUSCH).
  • NNSCH can be carried on the physical neural network shared channel (PNNSCH), so as to provide flexible and controllable transmission.
  • a dedicated traffic channel (dedicated traffic channel, DTCH) can be carried on a downlink shared channel (downlink shared channel, DL-SCH) and an uplink shared channel (uplink shared channel, UL-SCH).
  • DL-SCH and UL-SCH can be carried by PDSCH and PUSCH respectively.
  • the reference signal in this embodiment of the present application may include a neural network processing reference signal, and the neural network processing reference signal may be used for channel estimation and accuracy estimation during neural network processing.
  • FIG. 8b is a schematic diagram of layers corresponding to neural network processing provided by the embodiment of the present application.
  • the terminal device and the access network device may include a layer corresponding to the neural network processing service, and the layer corresponding to the neural network processing service is located in the physical layer (physical, PHY) above.
  • the layer corresponding to the neural network processing service may be used to process the first data, or process the first transmission symbols, so as to be input to the neural network or the like.
  • the embodiment of the present application does not limit the specific name of the layer corresponding to the neural network processing service.
  • the first data obtained by the first communication device from the layer corresponding to the neural network processing service may be directly transmitted through the physical layer. After the second communication device obtains the first transmission symbol, it may be processed by the physical layer and then directly transmitted to the layer corresponding to the neural network processing service.
  • the neural network processing logic channel may include NNCCH or NNTCH and the like.
  • the neural network processing transmission channel may include NNSCH (including uplink and downlink), and the neural network processing logical channel may be carried on DL-SCH, UL-SCH or NNSCH.
  • the neural network processing physical channel may include the PNNSCH, and the neural network transmission channel may be carried on the PNNSCH.
  • the NNCCH shown in Fig. 8c may be a neural network control channel, carried on DL-SCH and UL-SCH.
  • the neural network processing logical channel is used to carry neural network processing data.
  • the logical channel may be used to carry data processed by the neural network (such as the first transmission symbol).
  • the logical channel may be used to bear the processing results returned to the neural network and the like.
  • the neural network processing transmission channel may be used to bear the neural network processing logical channel
  • the neural network processing physical channel may be used to bear the neural network processing transmission channel.
  • the channel provided by this application can provide corresponding error control level and error level detection according to the requirement of accuracy, even if there is a transmission error, it can still provide data transmission service to the upper layer.
  • the neural network processing logic channel may be borne on a data shared channel (such as PDSCH or PUSCH) providing error-free services, or may be carried on a neural network processing shared channel (such as NNSCH).
  • the data transmission may also be carried on the neural network processing transmission channel.
  • the neural network processing control channel can be used to transmit configurations and parameters related to neural network processing, for example, the neural network processing control channel can be carried on a traditional data sharing channel (such as DL-SCH or UL-SCH).
  • Fig. 9a is a schematic diagram of a scenario of a data processing method provided by an embodiment of the present application.
  • the processing process of the data by the first communication device may include: embedding and preprocessing, etc.; the processing process of the data by the second communication device includes: post-processing, reasoning, etc.
  • FIG. 9a can also be understood as: the first communication device includes an embedding module, a preprocessing module, and the like, and the second communication device includes a postprocessing module, an inference module, and the like.
  • each module shown here may be a functional module, and the functional module may be implemented in the form of hardware, or may also be implemented in the form of a software functional module, etc., which is not limited in this embodiment of the present application. Since the first communication device may not perform the embedding operation, the embedding shown in Fig. 9a is represented by a dotted line.
  • Fig. 9a can be understood as a data processing method based on direct modulation.
  • the first communication device first embeds the input (such as the first data) into a continuous (such as continuous real number or complex number, etc.) space, and performs preprocessing according to the first accuracy of the neural network processing data (that is, it can be understood In order to determine the first parameter according to the first accuracy, so as to perform preprocessing), the input is transformed into the first transmission symbol, so as to realize dimension adaptation.
  • the second communication device performs post-processing on the data transmitted through the channel (such as the first transmission symbol) to obtain the input of the neural network. Then the neural network (inference as shown in Fig.
  • Post-processing the data transmitted through the channel by the second communication device includes: the second communication device performs inverse transformation on the first transmission symbol transmitted through the channel according to the first parameter, such as performing linear transformation or nonlinear transformation.
  • the first communication device transforms the first data into the first transmission symbol according to the first transformation coefficient
  • the second communication device can inversely transform the received first transmission symbol through the channel transmission into the first transmission symbol according to the first transformation coefficient. data.
  • Fig. 9b is a schematic diagram of another data processing method provided by the embodiment of the present application.
  • the data processing process by the first communication device may include: embedding, source channel coding, quadrature amplitude modulation (quadrature amplitude modulation, QAM) and so on.
  • the processing process of the data by the second communication device may include: QAM demodulation, information source channel decoding and reasoning, and the like.
  • FIG. 9b can also be understood as: the first communication device includes an embedding module, a source channel coding module, and a QAM modulation module; the second communication device includes a QAM demodulation module, a source channel decoding module, and an inference module.
  • QAM modulation shown here is only an example, and the embodiment of the present application is also applicable to other modulation methods, such as binary phase shift keying (binary phase shift keying, BPSK) or quadrature phase shift keying (quadrature phase shift keying) , QPSK) etc.
  • binary phase shift keying binary phase shift keying
  • BPSK binary phase shift keying
  • quadrature phase shift keying quadrature phase shift keying
  • QPSK quadrature phase shift keying
  • Fig. 9b is an example of source channel coding.
  • source channel coding can implement transformation (such as real number to real number transformation) and quantization of embedded data, and map to modulation symbols.
  • transformation such as real number to real number transformation
  • quantization such as quantization of embedded data
  • map to modulation symbols if there is no source channel coding, the method provided in the embodiment of the present application may include: mapping embedded data to modulation symbols through coding, and the like.
  • Figure 9b can be understood as part or all of the modules based on traditional data transmission, but the relevant parameters or retransmissions are determined according to the data processing method provided by the embodiment of the present application (such as based on the first accuracy or the second accuracy) of).
  • the first communication device first embeds the input into a continuous space, and performs encoding and modulation according to the first accuracy of the neural network processing data, so as to obtain the first transmission symbol.
  • the second communication device demodulates and decodes the data transmitted through the channel (such as the first transmission symbol) to obtain the input of the neural network. Then the neural network processes the input to obtain a processing result and a second accuracy corresponding to the processing result.
  • the second communication device may also adjust related parameters (for example, adjust the first parameter) or instruct retransmission according to the second accuracy corresponding to the processing result; or, the second communication device feeds back the processing result and the like.
  • Fig. 9c is a schematic diagram of another data processing method provided by the embodiment of the present application.
  • Figure 9c can be understood as a data processing method based on direct modulation, or a data processing method that multiplexes part of the tradition (for example, pre-processing and post-processing can be traditional modulation and demodulation, etc.).
  • related parameters or retransmissions are determined according to the data processing method provided in the embodiment of the present application (for example, determined based on the distance in the feature space). It is understandable that reference may be made to FIG. 9a for a specific description of FIG. 9c, and details are not repeated here.
  • the error estimation may be used to represent the first accuracy of the neural network input data, that is, the error estimation may be used to estimate the first accuracy.
  • the error estimation can also be used to represent the second accuracy of the processing result, so that the second communication device can determine whether retransmission is required, adjust parameters, and the like.
  • the first communication device adopts direct modulation (such as the direct modulation is a transformation in a continuous space)
  • the input the input of the first communication device as shown in Figure 9a
  • the input can be embedded in the continuous space (as shown in Figure 9a embedding operation), and then input the embedded output to the preprocessing, so as to obtain the first transmission symbol.
  • the second communication device can estimate the channel or the influence of the transmission on the semantic distance according to the received first transmission symbol transmitted through the channel, and then adjust the transmission configuration according to the second accuracy of reasoning (such as adjusting the first parameter, And optionally, feedback the adjustment of the first parameter to the first communication device, etc.).
  • the embodiment of the present application defines as follows: the input after embedding (that is, the preprocessed input) is a feature space, that is, the space where the features that need to be inferred by the neural network are located.
  • the first transmission symbol is in the channel space, that is, the space where the channel is located.
  • the output of neural network inference is the inference space, which is the space where the processing results are located.
  • the distance (or loss) of the channel space can be the distance between the transmitted symbols before and after the channel
  • the distance (or loss) of the feature space is the distance between the features before preprocessing and postprocessing
  • the distance of the inference space (or Loss) is the distance between the input of the neural network after transmission and the output of inference without transmission.
  • the relationship between the distance of the channel space, the distance of the feature space, and the inference distance is as follows:
  • the distance of the channel space can be deduced from the distance of the feature space after transmission preprocessing and postprocessing, and the distance of the feature space can be obtained by neural Network processing can get the distance of inference space.
  • the transmission and inference can be decoupled, and the corresponding inference distance (or loss) can be defined according to different tasks; the distance of the feature space is calculated from the inference distance, which is the demand for transmission; and then it can be calculated corresponding channel distance.
  • the transmission parameter table takes the distance of the feature space as the performance index, and defines the transmission parameters under the corresponding channel quality. That is to say, the data transmission can correspond to various neural network processing, as long as these neural network processing have the same distance requirements on the feature space, the same transmission configuration (such as the first parameter or the first accuracy, etc.) can be reused.
  • the distance of the feature space can be used as an indicator for retransmission or parameter adjustment, and the distance of the feature space can determine the distance of neural network reasoning.
  • the distance requirement of neural network inference can determine the distance of the feature space.
  • the relationship among MSE, coding rate, adjustment order and channel information may be as shown in Table 7 to Table 9.
  • FIG. 10 is a schematic diagram of a simulation result provided by an embodiment of the present application.
  • Figure 10 takes the classification task as an example.
  • the abscissa is the signal-to-noise ratio (dB), and the ordinate is the neural network reasoning, that is, the classification accuracy (accuracy, that is, the proportion of correct classification).
  • the dotted line represents the performance of the direct modulation data processing method, and the solid line represents the joint photographic experts group (joint photographic experts group, JPEG) (that is, a compression method for continuous tone still images) compression plus coding modulation
  • JPEG joint photographic experts group
  • JPEG joint photographic experts group
  • compression ratio compression ratio
  • the performance upper limit, the conversion coefficient (compression ratio, CR) is 12, and the solid line parallel to the horizontal axis represents the performance when the data is directly processed by the neural network.
  • the inference accuracy varies smoothly with the signal-to-noise ratio, so the required accuracy can be achieved by adjusting parameters (such as the first parameter or the second parameter, etc.), or the accuracy can be reduced to save transmission resources .
  • the data processing method based on direct modulation such as the data processing method based on the first transform coefficient
  • the data processing method based on coded modulation higher accuracy. Therefore, different data processing methods can be used under different signal-to-noise ratio conditions to achieve better performance.
  • relevant parameters for processing data are set according to the first accuracy of neural network processing data, so that the processing of data only needs to comply with the first accuracy of neural network processing data.
  • the situation of continuously retransmitting data due to the need to ensure the correctness of data transmission is improved, and the utilization efficiency of wireless resources is effectively improved. And since there is no need to retransmit data until there is no error, the end-to-end delay is also reduced.
  • the method provided in this application does not require bit-level processing, and it is easy to evaluate the impact of transmission errors on the processing accuracy of the neural network.
  • the present application divides the communication device into functional modules according to the above method embodiments.
  • each functional module may be divided corresponding to each function, or two or more functions may be integrated into one processing module.
  • the above-mentioned integrated modules can be implemented in the form of hardware or in the form of software function modules. It should be noted that the division of modules in this application is schematic, and is only a logical function division, and there may be other division methods in actual implementation.
  • the communication device according to the embodiment of the present application will be described in detail below with reference to FIG. 11 to FIG. 13 .
  • FIG. 11 is a schematic structural diagram of a communication device provided by an embodiment of the present application. As shown in FIG. 11 , the communication device includes a processing unit 1101 and a transceiver unit 1102 .
  • the communication device may be the first communication device or chip shown above, and the chip may be applied to (or set in) the first communication device and the like. That is, the communication device may be used to perform the steps or functions performed by the first communication device in the above method embodiments.
  • a processing unit 1101 configured to obtain first data, and process the first data according to a first parameter to obtain a first transmission symbol
  • the transceiver unit 1102 is configured to send the first transmission symbol to the second communication device.
  • the processing unit 1101 is specifically configured to transform the first data according to the first transform coefficient to obtain the first transmission symbol; or, transform the first data according to the first modulation order and the first coding rate A data is encoded and modulated to obtain a first transmission symbol.
  • the transceiver unit 1102 is further configured to send a first request message to the second communication device, where the first request message is used to request neural network processing, where the first request message includes first indication information, The first indication information is used to indicate a first accuracy;
  • the transceiver unit 1102 is further configured to receive a first response message from the second communication device, where the first response message includes second indication information, where the second indication information is used to indicate the first parameter.
  • the transceiver unit 1102 is further configured to receive a second request message from the second communication device, where the second request message is used to request data, and the second request message includes third indication information, the The third indication information is used to indicate the first parameter.
  • the transceiver unit 1102 is further configured to receive a feedback message from the second communication device when the second accuracy of the processing result of the first data satisfies a preset condition, the feedback message includes Information indicating the result of the processing.
  • the transceiver unit 1102 is further configured to receive retransmission indication information from the second communication device when the second accuracy of the processing result of the first data does not meet a preset condition, the The retransmission instruction information is used to instruct retransmission of the first data;
  • the transceiver unit 1102 is further configured to retransmit the first data according to the retransmission indication information.
  • the transceiver unit and the processing unit shown in the embodiments of the present application are only examples.
  • the processing unit 1101 can also be used to execute step 201 and step 202 shown in FIG. 2
  • the transceiver unit 1102 can also be used to execute the sending step in step 203 shown in FIG. 2 .
  • the transceiver unit 1102 may also be configured to perform the sending step in step 401 shown in FIG. 4 , the receiving step in step 402 and the sending step in step 403 .
  • the transceiver unit 1102 may also be configured to perform the receiving step in step 501 and the sending step in step 502 shown in FIG. 5 .
  • the transceiver unit 1102 may also be configured to perform the sending step in step 601 shown in FIG. 6 . It can be understood that the transceiving unit and the receiving unit can also be used to execute the methods shown in FIG. 9a to FIG. 9c, etc., which will not be described in detail here.
  • the communication device may be the second communication device or chip shown above, and the chip may be applied to (or provided in) the second communication device. That is, the communication device can be used to perform the steps or functions performed by the second communication device in the above method embodiments.
  • a transceiver unit 1102 configured to receive a first transmission symbol from a first communication device
  • a processing unit 1101, configured to process the first transmission symbol according to the first parameter, to obtain the input of the neural network
  • the processing unit 1101 is further configured to process the input according to the neural network to obtain a processing result.
  • the processing unit 1101 is specifically configured to inversely transform the first transmission symbol according to the first transformation coefficient to obtain the input of the neural network; or, according to the first modulation order and the first coding rate The first transmission symbol is demodulated and decoded to obtain the input of the neural network.
  • the transceiver unit 1102 is further configured to receive a first request message from the first communication device, where the first request message is used to request neural network processing, and the first request message includes first indication information , the first indication information is used to indicate the first accuracy;
  • the transceiver unit 1102 is further configured to send a first response message to the first communication device, where the first response message includes second indication information, where the second indication information is used to indicate the first parameter.
  • the transceiver unit 1102 is further configured to send a second request message to the first communication device, the second request message is used to request data, the second request message includes third indication information, and the first The three indication information are used to indicate the first parameter.
  • the transceiving unit 1102 is further configured to send a feedback message to the first communication device when the second accuracy of the input processing result satisfies a preset condition, the feedback message includes instructions for indicating Information about the results of this processing.
  • the transceiver unit 1102 is further configured to send retransmission instruction information to the first communication device when the second accuracy of the input processing result does not meet the preset condition, the retransmission instruction The information is used to indicate retransmission of the first data, and the first transmission symbol is obtained according to the first data.
  • the specific descriptions of the transceiver unit and the processing unit shown in the embodiments of the present application are only examples.
  • the transceiver unit 1102 can also be used to execute the receiving step in step 203 shown in FIG. 2 .
  • the processing unit 1101 may also be used to execute steps 404 to 406 shown in FIG. 4 (for example, the processing unit 1101 may be used to control the transceiver unit 1102 to output processing results, or retransmission instructions, etc.).
  • the transceiver unit 1102 can also be used to execute the sending step in step 501 shown in FIG. 5 and the receiving step in step 502, and the processing unit 1101 can also be used to execute steps 503 to 505 shown in FIG. 5 .
  • the transceiver unit 1102 may also be used to perform the receiving step in step 601 shown in FIG. 6
  • the processing unit 1101 may also be used to perform step 602 shown in FIG. 6 . It can be understood that the transceiving unit and the receiving unit can also be used to execute the methods shown in FIG. 9a to FIG. 9c, etc., which will not be described in detail here.
  • first communication device and the second communication device are described above, and possible product forms of the first communication device and the second communication device are introduced below. It should be understood that any product of any form having the function of the first communication device described above in FIG. 11 , or any product of any form having the function of the second communication device described in FIG. 11 above, falls within the scope of this application. The scope of protection of the embodiment. It should also be understood that the following introduction is only an example, and product forms of the first communication device and the second communication device in the embodiment of the present application are not limited thereto.
  • the processing unit 1101 may be one or more processors
  • the transceiver unit 1102 may be a transceiver, or the transceiver unit 1102 may also be a sending unit and a receiving unit
  • the sending unit may be a transmitter
  • the receiving unit may be a receiver
  • the sending unit and the receiving unit are integrated into one device, such as a transceiver.
  • the processor and the transceiver may be coupled, and the connection manner of the processor and the transceiver is not limited in the embodiment of the present application.
  • the communication device 120 includes one or more processors 1220 and a transceiver 1210 .
  • the processor 1220 is configured to process the first data according to the first parameter after obtaining the first data, and obtain The first transmission symbol; the transceiver 1210, configured to send the first transmission symbol to the second communication device.
  • the transceiver 1210 is used to receive the first transmission symbol from the first communication device;
  • the processor 1220 is used to receive the first transmission symbol from the first communication device;
  • the first parameter processes the first transmission symbol to obtain the input of the neural network, and processes the input according to the neural network to obtain a processing result.
  • the transceiver may include a receiver and a transmitter, the receiver is used to perform a function (or operation) of reception, and the transmitter is used to perform a function (or operation) of transmission ). And the transceiver is used to communicate with other devices/devices through the transmission medium.
  • the communication device 120 may further include one or more memories 1230 for storing program instructions and/or data.
  • the memory 1230 is coupled to the processor 1220 .
  • the coupling in the embodiments of the present application is an indirect coupling or a communication connection between devices, units or modules, which may be in electrical, mechanical or other forms, and is used for information exchange between devices, units or modules.
  • Processor 1220 may cooperate with memory 1230 .
  • Processor 1220 may execute program instructions stored in memory 1230 .
  • at least one of the above one or more memories may be included in the processor.
  • the memory 1230 may store the first parameter, the relationship between the first accuracy and the first channel information, and the like.
  • a specific connection medium among the transceiver 1210, the processor 1220, and the memory 1230 is not limited.
  • the memory 1230, the processor 1220, and the transceiver 1210 are connected through a bus 1240.
  • the bus is represented by a thick line in FIG. 12, and the connection between other components is only for schematic illustration. , is not limited.
  • the bus can be divided into address bus, data bus, control bus and so on. For ease of representation, only one thick line is used in FIG. 12 , but it does not mean that there is only one bus or one type of bus.
  • the processor may be a general-purpose processor, a digital signal processor, an application-specific integrated circuit, a field programmable gate array or other programmable logic device, a discrete gate or transistor logic device, a discrete hardware component, etc., and may realize Or execute the methods, steps and logic block diagrams disclosed in the embodiments of the present application.
  • a general purpose processor may be a microprocessor or any conventional processor or the like. The steps of the methods disclosed in connection with the embodiments of the present application may be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules in the processor.
  • the memory may include but not limited to hard disk drive (hard disk drive, HDD) or solid-state drive (solid-state drive, SSD) and other non-volatile memory, random access memory (Random Access Memory, RAM), Erasable Programmable ROM (EPROM), Read-Only Memory (ROM) or Portable Read-Only Memory (Compact Disc Read-Only Memory, CD-ROM), etc.
  • the memory is any storage medium that can be used to carry or store program codes in the form of instructions or data structures, and can be read and/or written by a computer (such as the communication device shown in this application, etc.), but is not limited thereto.
  • the memory in the embodiment of the present application may also be a circuit or any other device capable of implementing a storage function, and is used for storing program instructions and/or data.
  • the processor 1220 is mainly used to process communication protocols and communication data, control the entire communication device, execute software programs, and process data of the software programs.
  • the memory 1230 is mainly used to store software programs and data.
  • the transceiver 1210 may include a control circuit and an antenna, and the control circuit is mainly used for converting a baseband signal to a radio frequency signal and processing the radio frequency signal.
  • Antennas are mainly used to send and receive radio frequency signals in the form of electromagnetic waves.
  • Input and output devices, such as touch screens, display screens, and keyboards, are mainly used to receive data input by users and output data to users.
  • the processor 1220 can read the software program in the memory 1230, interpret and execute the instructions of the software program, and process the data of the software program.
  • the processor 1220 performs baseband processing on the data to be sent (as shown above in this application to process the first data according to the first parameter), and then outputs the baseband signal to the radio frequency circuit, and the radio frequency circuit will After the baseband signal is subjected to radio frequency processing, the radio frequency signal is sent out in the form of electromagnetic waves through the antenna.
  • the radio frequency circuit When data is sent to the communication device, the radio frequency circuit receives the radio frequency signal through the antenna, converts the radio frequency signal into a baseband signal, and outputs the baseband signal to the processor 1220, and the processor 1220 converts the baseband signal into data and processes the data deal with.
  • the radio frequency circuit and the antenna can be set independently from the processor for baseband processing.
  • the radio frequency circuit and antenna can be arranged remotely from the communication device. .
  • the communication device shown in the embodiment of the present application may have more components than those shown in FIG. 12 , which is not limited in the embodiment of the present application.
  • the method performed by the processor and the transceiver shown above is only an example, and for the specific steps performed by the processor and the transceiver, reference may be made to the method introduced above.
  • the processing unit 1101 may be one or more logic circuits, and the transceiver unit 1102 may be an input-output interface, or a communication interface, or an interface circuit , or interfaces and so on.
  • the transceiver unit 1102 may also be a sending unit and a receiving unit, the sending unit may be an output interface, and the receiving unit may be an input interface, and the sending unit and the receiving unit are integrated into one unit, such as an input and output interface.
  • the communication device shown in FIG. 13 includes a logic circuit 1301 and an interface 1302 .
  • the above-mentioned processing unit 1101 can be realized by a logic circuit 1301
  • the transceiver unit 1102 can be realized by an interface 1302 .
  • the logic circuit 1301 may be a chip, a processing circuit, an integrated circuit or a system on chip (SoC) chip, etc.
  • the interface 1302 may be a communication interface, an input/output interface, or a pin.
  • FIG. 13 takes the aforementioned communication device as a chip as an example, and the chip includes a logic circuit 1301 and an interface 1302 .
  • the logic circuit and the interface may also be coupled to each other.
  • the embodiment of the present application does not limit the specific connection manner of the logic circuit and the interface.
  • the communication device shown in FIG. 13 may further include a memory 1303, where the memory may be used to store a relationship among the first parameter, the first accuracy, and the first channel information.
  • the memory can be used to store Tables 1 to 9, or can also be used to store the formulas shown above, etc. Because the memory shown in FIG. 13 may not be integrated with the processor, but located outside the chip, the memory shown in FIG. 13 is represented by a dotted line.
  • the logic circuit 1301 when the communication device is used to perform the steps, methods or functions performed by the above-mentioned first communication device, the logic circuit 1301 is configured to process the first data according to the first parameter after obtaining the first data, and obtain A first transmission symbol; an interface 1302, configured to output the first transmission symbol.
  • the interface 1302 when the communication device is used to perform the steps or methods or functions performed by the above-mentioned second communication device, the interface 1302 is used to input the first transmission symbol; the logic circuit 1301 is used to input the first transmission symbol according to the first parameter.
  • the transmission symbol is processed, the input of the neural network is obtained, and the input is processed according to the neural network to obtain the processing result.
  • the communication device shown in the embodiment of the present application may implement the method provided in the embodiment of the present application in the form of hardware, or may implement the method provided in the embodiment of the present application in the form of software, which is not limited in the embodiment of the present application.
  • An embodiment of the present application further provides a communication system, where the communication system includes a first communication device and a second communication device, and the first communication device and the second communication device may be used to execute the method in any of the foregoing embodiments.
  • the present application further provides a computer program, which is used to implement the operations and/or processing performed by the first communication device in the method provided in the present application.
  • the present application also provides a computer program, which is used to implement the operations and/or processing performed by the second communication device in the method provided in the present application.
  • the present application also provides a computer-readable storage medium, the computer-readable storage medium stores computer programs or computer-executable instructions, and when the computer programs or computer-executable instructions are run on the computer, the computer executes the Operations and/or processing performed by the first communication device in the method.
  • the present application also provides a computer-readable storage medium, the computer-readable storage medium stores computer programs or computer-executable instructions, and when the computer programs or computer-executable instructions are run on the computer, the computer executes the Operations and/or processing performed by the second communication device in the method.
  • the present application also provides a computer program product, which includes computer-executable instructions or computer programs.
  • a computer program product which includes computer-executable instructions or computer programs.
  • the method provided by the present application is executed by the first communication device Performed The operation and/or processing is performed.
  • the present application also provides a computer program product, which includes computer-executable instructions or computer programs, and when the computer-executable instructions or computer programs are run on a computer, the method provided by the present application is executed by the second communication device Performed The operation and/or processing is performed.
  • the disclosed systems, devices and methods may be implemented in other ways.
  • the device embodiments described above are only illustrative.
  • the division of the units is only a logical function division. In actual implementation, there may be other division methods.
  • multiple units or components can be combined or May be integrated into another system, or some features may be ignored, or not implemented.
  • the mutual coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices or units, and may also be electrical, mechanical or other forms of connection.
  • the units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place, or may be distributed to multiple network units. Part or all of the units can be selected according to actual needs to realize the technical effects of the solutions provided by the embodiments of the present application.
  • each functional unit in each embodiment of the present application may be integrated into one processing unit, each unit may exist separately physically, or two or more units may be integrated into one unit.
  • the above-mentioned integrated units can be implemented in the form of hardware or in the form of software functional units.
  • the integrated unit is realized in the form of a software function unit and sold or used as an independent product, it can be stored in a computer-readable storage medium.
  • the computer software products are stored in a readable storage medium, including a number of instructions to make a computer device (which can be a personal computer, A server, or a network device, etc.) executes all or part of the steps of the methods described in the various embodiments of the present application.
  • the aforementioned readable storage media include: U disk, mobile hard disk, read-only memory (read-only memory, ROM), random access memory (random access memory, RAM), magnetic disk or optical disc, etc., which can store program codes. medium.

Landscapes

  • Engineering & Computer Science (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Signal Processing (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Computation (AREA)
  • Power Engineering (AREA)
  • Computer Security & Cryptography (AREA)
  • Mobile Radio Communication Systems (AREA)

Abstract

本申请公开了一种数据处理方法及装置，该方法包括：第一通信装置获得第一数据之后，根据第一参数对该第一数据进行处理，获得第一传输符号。然后，向第二通信装置发送该第一传输符号，对应的，第二通信装置接收该第一传输符号。第二通信装置根据第一参数对该第一传输符号进行处理，获得神经网络的输入，以及根据神经网络对该输入进行处理，获得处理结果。其中，第一参数根据第一准确度和第一信道信息确定，第一信道信息为第一通信装置与第二通信装置之间的信道信息，第一准确度用于表示神经网络处理数据的准确度。本申请提供的方法有效提高了无线资源的利用效率。

Description

数据处理方法及装置
本申请要求于2021年08月04日提交中国国家知识产权局、申请号为202110892508.7、申请名称为“数据处理方法及装置”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本申请涉及通信技术领域，尤其涉及一种数据处理方法及装置。
背景技术
以深度神经网络(deep neural networks，DNN)为代表的人工智能(artificial intelligence，AI)技术近年来在机器视觉、自然语言处理等领域取得重大进展，并且在实际生活中逐渐开始普及。DNN的主要特征之一是数值计算，即将各种不同类型的数据嵌入到实数空间，通过卷积、矩阵运算等数值运算，进行推理或者训练等。
一般的，无线传输设计以可靠传输为设计准则提供数据传输服务。示例性的，物理层可以对上层的比特流进行信道编码和调制等处理，获得处理后的数据，然后通过空口传输该处理后的数据。如信道质量(如信噪比)不同，相应的数据传输容量也不同，自适应编码调制和自动重传请求机制可以适应不同质量的信道，提供动态的传输速率。具体来说，传输的性能指标主要是误包率，传输参数(如调制与编码策略(modulation and coding scheme，MCS))以最大化吞吐量或者可靠度的要求，根据信道质量调整。
上述无线传输方法以无差错传输为目标，以误包率作为传输参数调整的依据；而神经网络处理的主要目标是准确计算，具有很好的容忍差错的能力，无差错传输可能会造成无线资源利用效率低下。
发明内容
本申请提供一种数据处理方法及装置，可以有效提高无线资源利用效率。
第一方面，本申请实施例提供一种数据处理方法，所述方法包括：
第一通信装置获得第一数据；所述第一通信装置根据第一参数对所述第一数据进行处理，获得第一传输符号，所述第一参数根据第一准确度以及第一信道信息确定，所述第一信道信息为所述第一通信装置与第二通信装置之间的信道信息，所述第一准确度用于表示所述第二通信装置中神经网络处理数据的准确度；所述第一通信装置向所述第二通信装置发送所述第一传输符号。
上述第一准确度还可以理解为：神经网络多次输出的处理结果的准确度；或者，神经网络输出的处理结果的准确度的平均；或者，神经网络处理数据的平均性能。示例性的，该平均性能包括平均正确率、平均准确度或平均准确率等。上述第一参数根据第一准确度以及第一信道信息确定，还可以理解为：第一参数与第一准确度和第一信道信息有关，或者，第一参数是根据第一准确度和第一信道信息得到的。根据第一准确度和第一信道信息确定第一通信装置处理数据时所使用的参数，可使得第一通信装置能够结合神经网络处理数据的需求来处理数据。从而，充分利用神经网络处理数据的准确度需求以及信道质量对第一数据进行处理，提高了无线资源的利用率。
一般的，通过无线传输方案传输数据时，是在数据传输完全正确后，才会进行神经网络的处理。然而，本申请实施例提供的方法中，第一通信装置可以根据第一参数对第一数据进行处理，该第一参数根据第一准确度和第一信道信息确定。从而，第一通信装置在传输数据时，对数据的处理只需要符合第一准确度，第二通信装置就可以对数据进行神经网络处理，改善了由于需要保证数据传输的正确性而重传数据的情况，有效提高了无线资源的利用效率。
在一种可能的实现方式中，所述第一参数包括第一变换系数，所述第一变换系数用于表示所述第一数据与所述第一传输符号之间的维度比；或者，所述第一参数包括第一调制阶数和第一编码码率。
上述第一变换系数用于表示第一数据与第一传输符号之间的维度比。例如，可以用第一数据的元素个数与第一传输符号的元素个数(如第一传输符号的个数)表示第一数据与第一传输符号之间的维度比。例如，第一变换系数越大，则表示第一通信装置在对第一数据进行处理时所保留的有效特征越少。由此，传输第一传输符号时所使用的传输资源越少，神经网络处理数据的第一准确度就越低。又例如，在信道信息相同(如信道质量相同)的情况下，第一变换系数越小，第一准确度的要求就越高。又例如，在第一准确度的要求相同的情况下，第一信道信息所表示的信道质量越好，第一变换系数就越大。
由于神经网络对数据的误差有灵活的容忍度，因此通过上述方法，可支持大范围信道质量区间内支持动态的第一变换系数与第一准确度，从而提高无线资源利用效率。
在信道信息相同的情况下，所述第一变换系数越小，所述第一准确度越高。
在一种可能的实现方式中，所述第一准确度用以下任一项或多项表示：所述神经网络的处理结果的置信度；所述神经网络的处理结果的分类精度；所述神经网络的输入数据与所述第一数据之间的均方误差(mean squared error，MSE)；所述神经网络的输入数据与所述第一数据之间的平均绝对误差(mean absolute error，MAE)。
本申请实施例中，第一准确度可以用神经网络多次处理结果对应的置信度的平均表示。或者，第一准确度可以用神经网络多次处理结果对应的概率的平均表示。或者，第一准确度可以用神经网络多次处理结果对应的分类精度的平均表示等。第一准确度可以用神经网络多次输入的数据与真实数据之间的MSE的平均表示，或者，用神经网络多次输入的数据与真实数据之间的均方根误差(root mean squard error，RMSE)的平均表示，或者，用神经网络多次输入的数据与真实数据之间的MAE的平均表示等。可理解，这里所示的第一准确度的表示方式仅为示例，对于神经网络的其他类型的输出，本申请实施例不作限定。
在一种可能的实现方式中，所述第一通信装置根据第一参数对所述第一数据进行处理包括：所述第一通信装置根据所述第一变换系数对所述第一数据进行变换，获得所述第一传输符号；或者，所述第一通信装置根据所述第一调制阶数和所述第一编码码率对所述第一数据进行编码和调制，获得所述第一传输符号。
在一种可能的实现方式中，所述方法还包括：所述第一通信装置向所述第二通信装置发送第一请求消息，所述第一请求消息用于请求所述神经网络进行数据处理，所述第一请求消息包括第一指示信息，所述第一指示信息用于指示所述第一准确度；所述第一通信装置接收来自所述第二通信装置的第一响应消息，所述第一响应消息包括第二指示信息，所述第二指示信息用于指示所述第一参数。
在一种可能的实现方式中，所述方法还包括：所述第一通信装置接收来自所述第二通信装置的第二请求消息，所述第二请求消息用于请求数据，所述第二请求消息包括第三指示信息，所述第三指示信息用于指示所述第一参数。
在一种可能的实现方式中，所述方法还包括：所述第一通信装置接收来自所述第二通信装置的反馈消息，所述反馈消息包括用于指示所述处理结果的信息。
在一种可能的实现方式中，所述方法还包括：所述第一通信装置接收来自所述第二通信装置的重传指示信息，所述重传指示信息用于指示重传所述第一数据；所述第一通信装置根据所述重传指示信息重传所述第一数据。
在一种可能的实现方式中，所述重传指示信息还用于指示第二参数，所述第二参数为所述第一参数进行更新后的参数。
在一种可能的实现方式中，以下一项或多项信息承载于神经网络处理控制信道中：所述第一指示信息、所述第二指示信息、所述第三指示信息或所述重传指示信息。
在一种可能的实现方式中，所述第一传输符号承载于神经网络处理共享信道。
本申请实施例中，神经网络处理控制信道和神经网络处理共享信道是基于神经网络处理的需求设计的。该神经网络处理共享信道在传输第一传输符号时，即使出现了差错，仍然可以使得第二通信装置对接收到的传输符号进行处理，提高了无线资源利用率。可选的，本申请实施例所示的神经网络处理控制信道和神经网络处理共享信道不仅可以用于传输与神经网络处理相关的参数或数据等，还可以用于传输其他不要求无差错传输的数据等。
第二方面，本申请实施例提供一种数据处理方法，所述方法包括：
第二通信装置接收来自第一通信装置的第一传输符号；所述第二通信装置根据第一参数对所述第一传输符号进行处理，获得神经网络的输入，所述第一参数根据第一准确度以及第一信道信息确定，所述第一信道信息为所述第一通信装置与所述第二通信装置之间的信道信息，所述第一准确度用于表示所述神经网络处理数据的准确度；所述第二通信装置根据所述神经网络对所述输入进行处理，获得处理结果。
在一种可能的实现方式中，所述第一参数包括第一变换系数，所述第一变换系数用于表示所述第一数据与所述第一传输符号之间的维度比；或者，所述第一参数包括第一调制阶数和第一编码码率。
在信道信息相同的情况下，所述第一变换系数越小，所述第一准确度越高。
在一种可能的实现方式中，所述第一准确度用以下任一项或多项表示：所述神经网络的处理结果的置信度；所述神经网络的处理结果的分类精度；所述神经网络的输入数据与所述第一数据之间的均方误差MSE；所述神经网络的输入数据与所述第一数据之间的平均绝对误差MAE。
在一种可能的实现方式中，所述第二通信装置根据第一参数对所述第一传输符号进行处理，获得神经网络的输入包括：所述第二通信装置根据所述第一变换系数对所述第一传输符号进行逆变换，获得所述神经网络的输入；或者，所述第二通信装置根据所述第一调制阶数和所述第一编码码率对所述第一传输符号进行解调和解码，获得所述神经网络的输入。
在一种可能的实现方式中，所述方法还包括：所述第二通信装置接收来自所述第一通信装置的第一请求消息，所述第一请求消息用于请求所述神经网络处理，所述第一请求消息包括第一指示信息，所述第一指示信息用于指示所述第一准确度；所述第二通信装置向所述第一通信装置发送第一响应消息，所述第一响应消息包括第二指示信息，所述第二指示信息用于指示所述第一参数。
在一种可能的实现方式中，所述方法还包括：所述第二通信装置向所述第一通信装置发送第二请求消息，所述第二请求消息用于请求数据，所述第二请求消息包括第三指示信息，所述第三指示信息用于指示所述第一参数。
在一种可能的实现方式中，所述方法还包括：在所述输入的处理结果的第二准确度满足预设条件的情况下，所述第二通信装置向所述第一通信装置发送反馈消息，所述反馈消息包括用于指示所述处理结果的信息。
在一种可能的实现方式中，所述方法还包括：在所述输入的处理结果的第二准确度不满足预设条件的情况下，所述第二通信装置向所述第一通信装置发送重传指示信息，所述重传指示信息用于指示重传第一数据，所述第一传输符号根据所述第一数据得到。
在一种可能的实现方式中，所述重传指示信息还用于指示第二参数，所述第二参数为所述第一参数进行更新后的参数。
在一种可能的实现方式中，以下一项或多项信息承载于神经网络处理控制信道中：所述第一指示信息、所述第二指示信息、所述第三指示信息或所述重传指示信息。
在一种可能的实现方式中，所述第一传输符号承载于神经网络处理共享信道。
可理解，关于第二方面的描述或有益效果可以参考第一方面，这里不再赘述。
第三方面，本申请实施例提供一种通信装置，用于执行第一方面或第一方面的任意可能的实现方式中的方法。该通信装置包括具有执行第一方面或第一方面的任意可能的实现方式中的方法的相应单元。
示例性的，该通信装置可以为第一通信装置或芯片等，该芯片可以应用于(或称为设置于)第一通信装置中。
第四方面，本申请实施例提供一种通信装置，用于执行第二方面或第二方面的任意可能的实现方式中的方法。该通信装置包括具有执行第二方面或第二方面的任意可能的实现方式中的方法的相应方法。
示例性的，该通信装置可以为第二通信装置或芯片等，该芯片可以应用于(或称为设置于)第二通信装置中。
在第三方面或第四方面中，上述通信装置可以包括收发单元和处理单元。对于收发单元和处理单元的具体描述还可以参考下文示出的装置实施例。
第五方面，本申请实施例提供一种通信装置，该通信装置包括处理器，用于执行上述第一方面或第一方面的任意可能的实现方式所示的方法。或者，该处理器用于执行存储器中存储的程序，当该程序被执行时，上述第一方面或第一方面的任意可能的实现方式所示的方法被执行。
在执行上述方法的过程中，上述方法中有关发送的过程，可以理解为由处理器输出的过程。例如，处理器输出数据时，处理器将该数据输出给收发器，以便由收发器进行发射。该数据在由处理器输出之后，还可能需要进行其他的处理，然后才到达收发器。类似的，处理器接收输入的数据时，收发器接收该数据，并将其输入处理器。更进一步的，在收发器收到该数据之后，该数据可能需要进行其他的处理，然后才输入处理器。可理解，关于该说明，下文示出的第六方面，同样适用。
对于处理器所涉及的发射、发送和接收等操作，如果没有特殊说明，或者，如果未与其在相关描述中的实际作用或者内在逻辑相抵触，则均可以更加一般性的理解为处理器输出和接收、输入等操作，而不是直接由射频电路和天线所进行的发射、发送和接收操作。
在实现过程中，上述处理器可以是专门用于执行这些方法的处理器，也可以是执行存储器中的计算机指令来执行这些方法的处理器，例如通用处理器。上述存储器可以为非瞬时性(non-transitory)存储器，例如只读存储器(read only memory，ROM)，其可以与处理器集成在同一块芯片上，也可以分别设置在不同的芯片上，本申请实施例对存储器的类型以及存 储器与处理器的设置方式不做限定。可理解，对于处理器和存储器的说明同样适用于下文示出的第六方面，下文不再详述。
在一种可能的实现方式中，存储器位于上述通信装置之外。
在一种可能的实现方式中，存储器位于上述通信装置之内。
本申请实施例中，处理器和存储器还可以集成于一个器件中，即处理器和存储器还可以被集成在一起。
在一种可能的实现方式中，通信装置还包括收发器，该收发器，用于接收信号或发送信号。
本申请实施例中，该通信装置可以为第一通信装置或芯片等，该芯片可以应用于第一通信装置中。
第六方面，本申请实施例提供一种通信装置，该通信装置包括处理器，用于执行上述第二方面或第二方面的任意可能的实现方式所示的方法。或者，处理器用于执行存储器中存储的程序，当该程序被执行时，上述第二方面或第二方面的任意可能的实现方式所示的方法被执行。
在一种可能的实现方式中，存储器位于上述通信装置之外。
在一种可能的实现方式中，存储器位于上述通信装置之内。
在本申请实施例中，处理器和存储器还可以集成于一个器件中，即处理器和存储器还可以被集成在一起。
在一种可能的实现方式中，通信装置还包括收发器，该收发器，用于接收信号或发送信号。
本申请实施例中，该通信装置可以为第二通信装置或芯片等，该芯片可以应用于第二通信装置中。
第七方面，本申请实施例提供一种通信装置，该通信装置包括逻辑电路和接口，所述逻辑电路和所述接口耦合；所述接口，用于输入第一数据；所述逻辑电路，用于根据根据第一参数对所述第一数据进行处理，获得第一传输符号；所述接口，还用于输出所述第一传输符号。
可理解，关于第一参数、第一准确度和第一信道信息等的描述，可以参考上述第一方面或第二方面的描述；或者，还可以参考下文示出的各个实施例，这里不再详述。
第八方面，本申请实施例提供一种通信装置，该通信装置包括逻辑电路和接口，所述逻辑电路和所述接口耦合；所述接口，用于输入第一传输符号；所述逻辑电路，用于根据第一参数对所述第一传输符号进行处理，获得神经网络的输入；以及根据所述神经网络对所述输入进行处理，获得处理结果。
可理解，关于第一参数、第一准确度和第一信道信息等的描述，可以参考上述第一方面或第二方面的描述；或者，还可以参考下文示出的各个实施例，这里不再详述。
第九方面，本申请实施例提供一种计算机可读存储介质，该计算机可读存储介质用于存储计算机程序，当其在计算机上运行时，使得上述第一方面或第一方面的任意可能的实现方式所示的方法被执行。
第十方面，本申请实施例提供一种计算机可读存储介质，该计算机可读存储介质用于存储计算机程序，当其在计算机上运行时，使得上述第二方面或第二方面的任意可能的实现方式所示的方法被执行。
第十一方面，本申请实施例提供一种计算机程序产品，该计算机程序产品包括计算机程 序或计算机可执行指令，当其在计算机上运行时，使得上述第一方面或第一方面的任意可能的实现方式所示的方法被执行。
第十二方面，本申请实施例提供一种计算机程序产品，该计算机程序产品包括计算机程序或计算机可执行指令，当其在计算机上运行时，使得上述第二方面或第二方面的任意可能的实现方式所示的方法被执行。
第十三方面，本申请实施例提供一种计算机程序，该计算机程序在计算机上运行时，上述第一方面或第一方面的任意可能的实现方式所示的方法被执行。
第十四方面，本申请实施例提供一种计算机程序，该计算机程序在计算机上运行时，上述第二方面或第二方面的任意可能的实现方式所示的方法被执行。
第十五方面，本申请实施例提供一种通信系统，该通信系统包括第一通信装置和第二通信装置，所述第一通信装置用于执行上述第一方面或第一方面的任意可能的实现方式所示的方法，所述第二通信装置用于执行上述第二方面或第二方面的任意可能的实现方式所示的方法。
附图说明
图1是本申请实施例提供的一种通信系统示意图；
图2是本申请实施例提供的一种数据处理方法的交互示意图；
图3是本申请实施例提供的一种第一准确度与第一信噪比的关系示意图；
图4至图6是本申请实施例提供的一种数据处理方法的交互示意图；
图7a至图7d是本申请实施例提供的一种帧结构示意图；
图8a是本申请实施例提供的与神经网络处理对应的信道示意图；
图8b是本申请实施例提供的与神经网络处理对应的层次示意图；
图8c是本申请实施例提供的与神经网络处理对应的信道示意图；
图9a至图9c是本申请实施例提供的一种数据处理方法的场景示意图；
图10是本申请实施例提供的一种仿真结果示意图；
图11至图13是本申请实施例提供的一种通信装置的结构示意图。
具体实施方式
为了使本申请的目的、技术方案和优点更加清楚，下面将结合附图对本申请作进一步地描述。
本申请的说明书、权利要求书及附图中的术语“第一”和“第二”等仅用于区别不同对象，而不是用于描述特定顺序。此外，术语“包括”和“具有”以及它们的任何变形，意图在于覆盖不排他的包含。例如包含了一系列步骤或单元的过程、方法、系统、产品或设备等，没有限定于已列出的步骤或单元，而是可选地还包括没有列出的步骤或单元等，或可选地还包括对于这些过程、方法、产品或设备等固有的其它步骤或单元。
在本文中提及的“实施例”意味着，结合实施例描述的特定特征、结构或特性可以包含在本申请的至少一个实施例中。在说明书中的各个位置出现该短语并不一定均是指相同的实施例，也不是与其它实施例互斥的独立的或备选的实施例。本领域技术人员可以显式地和隐式地理解的是，本文所描述的实施例可以与其它实施例相结合。
在本申请中，“至少一个(项)”是指一个或者多个，“多个”是指两个或两个以上， “至少两个(项)”是指两个或三个及三个以上，“和/或”，用于描述关联对象的关联关系，表示可以存在三种关系，例如，“A和/或B”可以表示：只存在A，只存在B以及同时存在A和B三种情况，其中A，B可以是单数或者复数。字符“/”一般表示前后关联对象是一种“或”的关系。“以下至少一项(个)”或其类似表达，是指这些项中的任意组合。例如，a，b或c中的至少一项(个)，可以表示：a，b，c，“a和b”，“a和c”，“b和c”，或“a和b和c”。
本申请提供的方法可以应用于各类通信系统，例如，可以是物联网(internet of things，IoT)系统、窄带物联网(narrow band internet of things，NB-IoT)系统、长期演进(long term evolution，LTE)系统，也可以是第五代(5th-generation，5G)通信系统，以及未来通信发展中出现的新的通信系统(如6G)等。以及本申请提供的方法还可以应用于无线局域网(wireless local area network，WLAN)系统，如无线保真(wireless-fidelity，Wi-Fi)等。
本申请提供的技术方案还可以应用于机器类通信(machine type communication，MTC)、机器间通信长期演进技术(long term evolution-machine，LTE-M)、设备到设备(device-todevice，D2D)网络、机器到机器(machine to machine，M2M)网络、物联网(internet of things，IoT)网络或者其他网络。其中，IoT网络例如可以包括车联网。其中，车联网系统中的通信方式统称为车与任何事物(vehicle-to-everything，V2X，X可以代表任何事物)，例如，该V2X可以包括：车辆到车辆(vehicle to vehicle，V2V)通信，车辆与基础设施(vehicle to infrastructure，V2I)通信、车辆与行人之间的通信(vehicle to pedestrian，V2P)或车辆与网络(vehicle to network，V2N)通信等。示例性的，下文示出的图1中，终端设备与终端设备之间便可以通过D2D技术、M2M技术或V2X技术通信等。
图1是本申请实施例提供的一种通信系统的示意图。本申请下文示出的方法实施例可以适用于图1所示的通信系统，下文不再赘述。
示例性的，该通信系统可以包括至少一个接入网设备以及至少一个终端设备。
示例性的，接入网设备可以是下一代节点B(next generation node B，gNB)、下一代演进型基站(next generation evolved nodeB，ng-eNB)(可以简称为eNB)、或者未来6G通信中的接入网设备等。接入网设备可以是任意一种具有无线收发功能的设备，包括但不限于以上所示的基站。该基站还可以是未来通信系统如第六代通信系统中的基站。可选的，该接入网设备可以为无线局域网(wireless fidelity，WiFi)系统中的接入节点、无线中继节点、无线回传节点等。可选的，该接入网设备可以是云无线接入网络(cloud radio access network，CRAN)场景下的无线控制器。可选的，该接入网设备可以是可穿戴设备或车载设备等。可选的，该接入网设备还可以是小站，传输接收节点(transmission reception point，TRP)(或也可以称为传输点或接收点)等。可理解，该接入网设备还可以是未来演进的公共陆地移动网络(public land mobile network，PLMN)中的基站等等。
示例性的，该终端设备也可称为用户设备(user equipment，UE)、终端等。终端设备是一种具有无线收发功能的设备，可以部署在陆地上，包括室内或室外、手持、穿戴或车载；也可以部署在水面上，如轮船上等；还可以部署在空中，例如部署在飞机、气球或卫星上等。终端设备可以是手机(mobile phone)、平板电脑(Pad)、带无线收发功能的电脑、虚拟现实(virtual reality，VR)终端设备、增强现实(augmented reality，AR)终端设备、工业控制(industrial control)中的无线终端、无人驾驶(self driving)中的无线终端、远程医疗(remote  medical)中的无线终端、智能电网(smart grid)中的无线终端、运输安全(transportation safety)中的无线终端、智慧城市(smart city)中的无线终端、智慧家庭(smart home)中的无线终端等等。可理解，该终端设备还可以是未来6G网络中的终端设备或者未来演进的PLMN中的终端设备等。
可理解，本申请示出的终端设备不仅可以包括车联网中的车(如整车)、而且还可以包括车联网中的车载设备或车载终端(包括车载信息盒子(telematics box，T-box)或车联网系统中的主机)等，本申请对于该终端设备应用于车联网时的具体形态不作限定。为便于描述，下文中将以终端设备为UE为例，介绍本申请所涉及的方法。
图1所示的通信系统中，包括一个接入网设备和四个UE如图1中的UE1至UE4。如UE1、UE3和UE4可以是手机，UE2可以是车等；或者，UE3和UE4是车，UE1和UE2是手机等。可理解，关于UE和接入网设备的具体说明可以参考上文，这里不再赘述。应理解，图1示例性地示出了一个接入网设备和四个UE，以及各通信设备之间的通信链路。可选地，该通信系统可以包括多个接入网设备，并且每个接入网设备的覆盖范围内可以包括其它数量的UE，例如更多或更少的UE等，本申请对此不做限定。可选的，图1所示的通信系统还可以包括核心网设备，如接入与移动性管理功能(access and mobility management function，AMF)等，本申请对此不作限定。
一般的，在无线传输设计以可靠性传输为设计准则，如第一通信装置向第二通信装置发送数据，如果第二通信装置无法正确译码其接收到的数据，则第一通信装置需要重传数据。在第二通信装置正确译码之后，该第二通信装置对其接收到的数据进行神经网络(neural network，NN)的处理等。然而，由于神经网络具有很好的容忍差错的能力，因此第二通信装置可能不需要完全正确接收数据，就可以进行神经网络处理。也就是说，上述第二通信装置需要完全正确接收到来自第一通信装置的数据之后，才会进行神经网络处理的方法会使得无线资源利用率低下。
鉴于此，本申请提供一种数据传输方法及装置，以第二通信装置中神经网络处理数据的准确度为传输设计的指标，从而确定第一通信装置处理数据的相关参数。进而能够充分利用不同信道信息下数据的传输误差与神经网络处理数据的准确度的折衷(如信噪比越低，准确度越低，因此可以通过降低准确度来降低对无线资源的要求)，有效提高了无线资源的利用效率。
在介绍本申请提供的方法之前，先对本申请涉及的术语进行说明。
1、第一准确度：
本申请所示的第一准确度可以用于表示神经网络处理数据的准确度。进一步的，第一准确度可以用于表示神经网络输出的处理结果的准确度的平均。或者，第一准确度可以用于表示神经网络处理数据的平均性能。例如，该平均性能包括平均正确率、平均准确度或平均准确率等。
示例性的，神经网络的输出包括处理结果。该处理结果可以与神经网络的处理类型对应。例如，神经网络的处理类型包括分类类型，则该处理结果可以是神经网络的输入所对应的类别。又例如，神经网络的处理类型包括模式识别，则该处理结果可以是神经网络的输入所对应的模式。可选的，神经网络的输出还包括神经网络所获得的处理结果的准确度，获得也可以称为与处理结果对应的准确度。例如，对于分类任务来说，输出的可以是输入属于各个类别的概率，概率最大所对应的类别作为输入的处理结果。该情况下，处理结果的准确度则可 以用神经网络的输出所对应的类别的概率表示。可选的，神经网络预测输出的概率也可以是修正的概率，如通过对原始输出概率进行修正，以使该修正的概率接近准确度。即第一准确度可以用修正的概率表示。又例如，神经网络的输出还可以包括分类精度(precision)或置信度等。仍以分类任务为例，若神经网络输出的概率分布比较集中(或者熵较小)，则表示输出的置信度较高。分类精度表示输出中判决为正确的结果所占的比例。
根据上述介绍，第一准确度可以用神经网络多次处理结果对应的置信度的平均表示。或者，第一准确度可以用神经网络多次处理结果对应的概率的平均表示。或者，第一准确度可以用神经网络多次处理结果对应的分类精度的平均表示等。可理解，以上所示的置信度或分类精度或概率仅为示例，对于神经网络的输出可能还包括其他类型，如召回率(recall)等。
以上所示的神经网络的输出可以理解为是以离散的输出为例，然而，本申请中神经网络的输出还可以包括其他类型的输出。例如，神经网络的输出可以信道估计的结果。又例如，神经网络的输出还可以是数值的运算结果等。又例如，神经网络的输出还可以是参数融合的结果等。该情况下，第一准确度可以用神经网络多次输入的数据与真实数据之间的误差的平均表示。也就是说，第一准确度可以用传输数据的误差的平均表示。例如，误差包括神经网络的接收梯度(即神经网络所获得的输入)与真实梯度(即发送端所发送的数据)之间的偏差。又例如，误差包括联邦学习中融合参数与原始参数的偏差等。示例性的，误差可以包括均方误差(mean squared error，MSE)、均方根误差(root mean squard error，RMSE)或平均绝对误差(mean absolute error，MAE)中的任一项或多项。也就是说，第一准确度可以用神经网络多次输入的数据与真实数据之间的MSE的平均表示，或者，用神经网络多次输入的数据与真实数据之间的RMSE的平均表示，或者，用神经网络多次输入的数据与真实数据之间的MAE的平均表示等。可理解，这里所示的第一准确度还可以理解为可以用特征空间的距离表示。特征空间的距离即特征之间的距离，该距离可以用MSE、RMSE或MAE中的任一项或多项表示。
2、第二准确度：
本申请所示的第二准确度表示神经网络输出的处理结果的准确度。根据上述介绍的神经网络不同类型的输出，如第二准确度可以用处理结果的置信度表示，或者，第二准确度可以用处理结果的概率(或概率的熵)表示，或者，第二准确度可以用处理结果的分类精度表示等。或者，第二准确度可以用神经网络输入的数据与真实数据之间的误差表示，该误差包括MSE、RMSE或MSE中的一项或多项。
也就是说，第二准确度可以表示神经网络单次输出的处理结果的准确度。第一准确度可以表示神经网络多次输出的处理结果的准确度。
可理解，以上所示的神经网络的输出仅为示例，本申请对于神经网络的其他类型的输出不作限定。
图2是本申请实施例提供的一种数据处理方法的流程示意图。该方法可以应用于第一通信装置和第二通信装置。该第二通信装置可以对来自第一通信装置的数据进行神经网络处理。也就是说，第一通信装置可以为第二通信装置提供数据，第二通信装置根据神经网络对第一通信装置提供的数据进行处理。示例性的，第一通信装置可以为终端设备，第二通信装置可以为接入网设备。或者，第一通信装置为接入网设备，第二通信装置为终端设备。或者，第一通信装置和第二通信装置可以都是终端设备等，本申请实施例对于第一通信装置和第二通信装置的具体形态不作限定。可理解，关于终端设备和接入网设备的具体说明可以参考图1， 这里不再详述。如图2所示，该方法包括：
201、第一通信装置获得第一数据。
该第一数据可以是待第一通信装置处理的数据。或者，该第一数据可以是待发送给第二通信装置的数据等。示例性的，第一数据可以包括第一通信装置所获得的原始数据。例如，第二通信装置中的神经网络包括模型前向推理，该情况下，第一数据可以包括图像数据、语音数据或文字数据中的任一项类型等。示例性的，还可以将一个完整的神经网络在某一层切割，分割成两个部分神经网络，如第一部分神经网络和第二部分神经网络，从而分别部署在第一通信装置和第二通信装置上。如第一通信装置可以利用第一部分神经网络对输入数据进行部分处理，得到中间结果，然后将该中间结果发送给第二通信装置。第二通信装置在获得中间结果后，可以接着利用第二部分神经网络对该中间结果进行继续处理，得出推理结果。也就是说，第二通信装置中神经网络可以包括部分层神经网络，从而用于模型前向推理。即第一数据可以包括第一通信装置所获得的经过神经网络部分层处理后的数据。示例性的，第二通信装置可以作为联邦学习的参数服务器，从而对多个第一通信装置的神经网络更新的权重等参数进行融合。也就是说，第二通信装置中的神经网络包括联邦学习的模型融合，即第一数据可以包括第一通信装置所获得的模型参数或中间梯度数据等。
可理解，第一数据的具体类型可以根据第二通信装置中神经网络的处理类型确定，本申请实施例对于第一数据的具体类型不作限定。但凡是第二通信装置中神经网络能够处理的数据均属于本申请实施例保护的范围。对于第一通信装置获得第一数据的具体方法，本申请实施例不作限定。
202、第一通信装置根据第一参数对第一数据进行处理，获得第一传输符号，该第一参数根据第一准确度以及第一信道信息确定。
第一传输符号可以理解为第一通信装置对第一数据进行处理之后，所获得的能够用于传输的符号。例如，该第一传输符号可以包括调制符号、编码符号或编码调制符号等。本申请实施例对于该第一传输符号的具体类型不作限定。
第一参数可以理解为第一通信装置对数据(如本申请所示的第一数据)进行处理时所使用到的参数。示例性的，该第一参数还可以称为第一传输参数或第一处理参数等，本申请实施例对此不作限定。作为示例，如第一通信装置对第一数据进行线性变换或非线性变换，则第一参数可以包括第一变换系数(如可以参考下文关于图4所示的方法)。又如第一通信装置对第一数据进行编码调制，则第一参数可以包括第一调制阶数和第一编码码率(如可以参考下文图5所示的方法)。又如第一通信装置需要对第一数据进行调制，则第一参数可以包括第一调制阶数。又如第一通信装置需要对第一数据进行编码，则第一参数可以包括第一编码码率。可理解，关于步骤202的具体说明还可以参考下文，这里先不一一详述。
第一信道信息为第一通信装置与第二通信装置之间的信道信息。例如，第一信道信息可以为第一通信装置与第二通信装置之间的信道状态信息(channel state information，CSI)。又例如，该第一信道信息可以为用于表示第一通信装置与第二通信装置之间信道质量的信息。
本申请实施例所示的第一参数根据第一准确度以及第一信道信息确定，还可以理解为：第一参数与第一准确度和第一信道信息有关，或者，第一参数是根据第一准确度和第一信道信息得到的。根据第一准确度和第一信道信息确定第一通信装置处理数据时所使用的参数(如第一参数)，可使得第一通信装置能够结合神经网络处理数据的需求来处理数据。从而，充分利用神经网络处理数据的准确度需求以及信道质量对第一数据进行处理，提高了无线资源的利用率。如准确度需求稍差，则第一通信装置根据变换系数进行处理时，可以放松对数据的 处理要求(如保留的有效特征较少)。可理解，关于第一参数、第一准确度需求以及第一信道信息之间的关系还可以参考下文(如表1至表9的相关描述等)，这里先不一一详述。
在一种可能的实现方式中，第一通信装置获得第一数据之后，可以直接根据第一参数对第一数据进行处理。
在另一种可能的实现方式中，第一通信装置根据第一参数对第一数据进行处理之前，图2所示的方法还包括：第一通信装置将第一数据嵌入到连续空间。
可理解，上述步骤还可以理解为：根据第一数据的类型将该第一数据嵌入到连续空间；或者，根据第二通信装置中神经网络处理的类型，将待发送给第二通信装置的数据(即第一数据)嵌入到连续空间。
示例性的，当第一数据包括神经网络部分层处理后的数据、模型参数或中间梯度数据等时，由于其本身是在连续空间，则也可以不进行嵌入操作。示例性的，对于原始数据，如第一数据包括图像数据、语音数据或无线数据(如信道信息)时，如果该第一数据本身是在连续空间(如图像数据或语音数据等是数值的形式)，则可以不对其进行嵌入操作。当第一数据包括文本数据或其他离散类数据(如其值不是数值的形式，而是字母或文字的形式等)，则需要将第一数据嵌入到连续空间。可选的，原始数据可以根据神经网络处理的任务(如自然语言处理的任务等)，嵌入到反映输入语义距离的空间，如语言的离散空间通过学习嵌入到连续空间，该连续空间的距离则可以反映单词的语义距离。
可选的，第一通信装置可以通过神经网络进行嵌入操作等，本申请实施例对此不作限定。例如，嵌入操作可以通过预训练后的网络实现，或者，通过预训练后的映射关系实现等。例如，通过预训练的映射，第一通信装置可以将单词(word)映射到实数向量(如每个word可以对应一个向量等)。
203、第一通信装置向第二通信装置发送第一传输符号，相应的，第二通信装置接收该第一传输符号。
可理解，在将第一传输符号发送出去之前，第一通信装置还可以进行其他处理，本申请实施例对此不作限定。例如，第一通信装置可以将第一传输符号映射到传输资源(如时频资源)上，通过该传输资源发送第一传输符号。又如，第一通信装置可以进行功率分配，如通过一定的发送功率发送第一传输符号。一般的，功率分配主要是应对衰落信道场景，从而实现信道白化或者特征抽取。示例性的，信道白化可以理解为是第一通信装置根据各子载波衰落程度的不同分配功率，从而保证第二通信装置看到的是同样的衰落。特征抽取可以理解为是通过功率分配，实现权重的卷积操作，如特征抽取可以是卷积神经网络中的一层。可理解，关于信道白化和特征抽取的具体说明本申请实施例不作限定。
204、第二通信装置根据第一参数对第一传输符号进行处理，获得神经网络的输入。
也就是说，第二通信装置在获取到第一传输符号之后，可以根据第一参数对第一传输符号进行逆处理，获得符合神经网络输入的数据。上述步骤204中的输入可以理解是符合神经网络输入的数据。
示例性的，第二通信装置根据第一变换系数对第一传输符号进行逆变换，如该逆变换可以包括线性变换或非线性变换。又例如，第二通信装置可以根据第一调制阶数对第一传输符号进行解调处理。又例如，第二通信装置可以根据第一编码码率对第一传输符号进行解码处理。又例如，第二通信装置可以根据第一调制阶数和第一编码码率对第一传输符号进行解调和解码。
205、第二通信装置根据神经网络对输入进行处理，获得处理结果。
可选的，第二通信装置还可以获得处理结果的准确度。
可选的，在第一数据被嵌入到连续空间的情况下，第二通信装置可以通过神经网络进行逆嵌入操作，或者，通过神经网络获得处理结果后，第二通信装置可以对该处理结果进行逆嵌入操作等，本申请实施例不作限定。例如，第二通信装置可以将神经网络输出的实数向量映射到单词等。又例如，第一通信装置通过图嵌入将图结构嵌入到实数向量空间，则相应的，第二通信装置可以从实数向量空间恢复出图结构。
可理解，关于神经网络输出的具体说明可以参考上文，这里不再赘述。关于第二通信装置获取到处理结果之后的方法可以参考下文关于图4至图6的说明，这里先不一一详述。
一般的，通过无线传输方案传输数据时，是在数据传输完全正确后，才会进行神经网络的处理。然而，本申请实施例提供的方法中，第一通信装置可以根据第一参数对第一数据进行处理，该第一参数根据第一准确度和第一信道信息确定。从而，第一通信装置在传输数据时，对数据的处理只需要符合第一准确度以及第一信道信息，第二通信装置就可以对数据进行神经网络处理，改善了由于需要保证数据传输的正确性而重传数据的情况，有效提高了无线资源的利用效率。
为进一步理解本申请提供的方法，以下详细说明本申请所示的第一参数。
实现方式一、
第一参数包括第一变换系数，该第一变换系数用于表示第一数据与第一传输符号之间的维度比。例如，可以用第一数据的元素个数与第一传输符号的元素个数(如第一传输符号的个数)表示第一数据与第一传输符号之间的维度比。例如，第一数据的形式可以包括向量或矩阵等，本申请实施例对此不作限定。
示例性的，第一通信装置对第一数据进行处理包括第一通信装置对第一数据进行变换处理，该变换包括线性变换，如离散傅里叶变换或离散余弦变换等，或者包括神经网络的变换等，本申请实施例对此不作限定。通过变换处理，可以实现第一数据和第一传输符号的维度适配，该维度适配可以理解为是根据神经网络处理数据的第一准确度的要求和第一数据的维度所确定的合适的传输符号个数。
该实现方式中，第一变换系数、第一准确性和第一信道信息之间的关系可以如下所示：
例如，第一变换系数越大，则表示第一通信装置在对第一数据进行处理时所保留的有效特征越少。由此，传输第一传输符号时所使用的传输资源越少，神经网络处理数据的第一准确度就越低。又例如，在第一信道信息相同(如信道质量相同)的情况下，第一变换系数越小，第一准确度的要求就越高。又例如，在第一准确度的要求相同的情况下，第一信道信息所表示的信道质量越好，第一变换系数就越大。
一般的，无线传输是以最大化吞吐量为目标，根据可靠度要求设置相应的参数，如信道质量指示(channel quality indication，CQI)或者调制与编码策略(modulation and coding scheme，MCS)等。示例性的，上述参数的设置方法可以包括信道质量(如信噪比(signal-to-noise ratio，SNR)在误块率(block error rate，BLER)等于0.1或者0.001条件下对应的MCS。然而，本申请实施例中，由于神经网络对于数据的误差有灵活的容忍度，如随着误差的增大，处理结果的准确度可能会平稳下降。也就是说，神经网络的数据处理可以有灵活的准确度。由此，本申请实施例可以支持在大范围信道质量区间支持动态的第一变换系数与第一准确度。
在一种可能的实现方式中，第一变换系数、第一准确度和第一信道信息之间的关系可以用曲线的形式表示。示例性的，图3是本申请实施例提供的一种第一准确度与第一信噪比的 关系示意图。可理解，图3所示的第一变换系数＝6和第一变换系数＝12仅为示例，不应将其理解为对本申请实施例的限定。如神经网络包括推理为例，从图3所示的两条曲线可以看出，随着第一信噪比的增大，推理的第一准确度随着提升。从图3所示的与横轴平行的线条可以看出，当第一准确度相同的情况下，第一信噪比越大，第一变换系数越大。从图3所示的与纵轴平行的线条可以看出，当第一信噪比相同的情况下，第一变换系数越小，第一准确度就越高。
由此，示例性的，对于准确度有要求的服务(如自动驾驶服务或雷达感知等对准确度有较高要求的服务)，如图3所示的与横轴平行的线条所示，则可以在相应的信噪比条件下选择相应的变换系数，以满足准确度的要求。示例性的，对于准确度弹性可控的服务(如图像识别或异物检测等对准确度有较低要求的服务)，如图3所示的与纵轴平行的线条所示，则可以在当前的信噪比条件下，通过控制变换系数，实现准确度与传输资源的平衡。可理解，图3所示的横轴的单位可以是dB，纵轴的单位可以是百分比。
在另一种可能的实现方式中，第一变换系数、第一准确度和第一信道信息之间的关系可以用函数的形式表示。示例性的，第一变换系数＝f(第一准确度，第一信道信息)。如第一变换系数＝f 1(第一准确度，第一SNR)，或者，第一变换系数＝f 2(第一准确度，第一SINR)等。
在又一种可能的实现方式中，第一变换系数、第一准确度和第一信道信息之间的关系可以用表的形式表示。示例性的，表1是当第一准确度等于99％时，第一变换系数与第一信道信息之间的关系。表2是当第一准确度等于90％时，第一变换系数与第一信道信息之间的关系。表3是当第一准确度等于80％时，第一变换系数与第一信道信息之间的关系。可理解，表1至表3是以SNR为例表示信道质量的好坏，但是不应将其理解为对本申请实施例的限定。例如，还可以通过信号与干扰加噪声比(signal to interference plus noise ratio，SINR)表示信道质量的好坏等。
表1
索引 0 1 2 3 4 5 6 …  
第一变换系数 1/8 1/4 1/2 1 2 4 8 …  
第一信噪比 -4 -3 -2 -1 0 1 2 …  
表2
索引 0 1 2 3 4 5 6 …  
第一变换系数 1/6 1/3 1 3/2 3 6 12 …  
第一信噪比 -4 -3 -2 -1 0 1 2 …  
表3
索引 0 1 2 3 4 5 6 …  
第一变换系数 1/4 1/2 1 2 4 8 16 …  
第一信噪比 -4 -3 -2 -1 0 1 2 …  
通过表1至表3所示的不同的第一准确度，第一变换系数可以根据神经网络处理的不同类型或者不同阶段确定。例如，神经网络处理的类型可以包括雷达感知或自动驾驶分析等，该情况下，可以根据较高的第一准确度以及第一信道信息确定第一变换系数。又例如，神经网络处理的类型包括图像识别、异物检测或机器翻译等，该情况下，可以根据较低的第一准确度(即可以理解为对准确度没有很严格的要求)以及传输资源等确定第一变换系数。又例如，神经网络训练前期可以有较大的误差，由此可以选择第一准确度较低的表格以提升无线 资源的利用效率。又例如，神经网络训练后期需要更精细的参数更新，由此可以选择第一准确度较高的表格。
可理解，表1至表3是以第一准确度分别为99％、90％和80％为例示出的。然而，可选的，还可以根据实际的准确度要求向上对齐的方法，选择对应的表格。例如，第一准确度的要求为95％时，根据向上对齐的原则，可以选择表1。又例如，第一准确度的要求为89％，则根据向上对齐的原则，可以选择表2。可选的，还可以根据实际的准确度要求向下对齐的方法选择对应的表格。可选的，表1还可以对应99％以及以上的准确度，表2对应90％至99％的准确度，表3对应80％至89％的准确度等。
可理解，表1至表3所示的具体关系仅为示例，在具体实现中，第一变换系数、第一准确度要求和第一信道信息之间的关系不限于此。例如，第一准确度还可以包括95％、80％或70％等。
在又一种可能的实现方式中，第一准确度还可以是弹性的，该情况下，第一变换系数可以根据第一信道信息确定。如第一变换系数、第一准确度和第一信道信息之间的关系还可以如表4所示。
表4
Figure PCTCN2022109296-appb-000001
表4示出了各信道质量下可选的第一变换系数，不同的第一变换系数可以对应神经网络处理数据不同的第一准确度。示例性的，可以根据传输资源的使用情况或者时延的需求选择灵活的第一准确度，从而确定第一变换系数。例如，时延要求严格(如要求时延较小)，则为更及时地对数据进行处理，第一变换系数可以较大。又例如，时延要求不严格(如时延可以较大)，则第一变换系数可以较小。可理解，这里所示的时延可以表示神经网络的处理和反馈的时延。可理解，尽管表4未体现出第一准确度，但是，可以参考第一准确度、第一信道信息以及第一变换系数的关系制定表格。例如，当第一信噪比为-4dB时，根据信噪比越大，准确度可能会越大的原则，则对应的准确度可能为99％、90％和80％。因此，第一变换系数相应可以依次为1/8、1/6和1/4。例如，当第一信噪比为-4dB时，对应的第一变换系数可以是1/8、1/6和1/4。该情况下，可以根据传输资源的使用情况选择第一变换系数。如传输资源较少，则第一变换系数可以较大，如可以是1/6或1/4。又或者，根据时延要求确定第一变换系数等，这里不再详述。
在又一种可能的实现方式中，第一变换系数、第一准确度和第一信道信息之间的关系还可以用表格结合公式的形式表示。例如，可以通过基础表格以及修正公式表示第一变换系数、第一准确度和第一信道信息之间的关系。如以第一准确度＝99％的表格作为基础表格(即基础准确度＝99％)，修正公式可以满足如下公式：修正变换系数＝f 3(基础变换系数，第一准确度)，或者，修正信噪比＝f 4(基础信噪比，第一准确度)。或者，修正变换系数＝f 5(基础变换系数，相对准确度)，或者，修正信噪比＝f 6(基础信噪比，相对准确度)。或者，修正变换系数＝基础变换系数+f 7(第一准确度，基础准确度)，f 7(第一准确度，基础准确度)可以包括f 7(第一准确度-基础准确度)、f 7(第一准确度/基础准确度)或者f 7(1-第一准确度/基础准确度)中的任一项。即，修改变换系数可以根据是第一准确度与基础准确度的差的函数，或者，是第一准确度与基础准确度的比值的函数等。也就是说，修正变换系数或者修正信噪比可以是根据具体的准确度 对基础参数表中的变换系数或者信噪比进行修正的结果。例如，基础表格是以第一准确度＝99％制定的，修正时第一准确度为80％，则80％相对99％即为相对准确度。
可理解，以上所示的第一变换系数、第一准确度和第一信道信息之间的表示形式仅为示例。关于实现方式一的具体说明还可以参考下文图4至图6所示的方法。本申请实施例所示的第一准确度可以用置信度、分类精度或概率等表示，还可以用MSE、RMSE或MAE中的任一项或多项表示等。
实现方式二、
可选的，第一参数包括第一编码码率。可选的，第一参数包括第一调制阶数。可选的，第一参数包括第一调制阶数和第一编码码率。
示例性的，第一编码码率与第一准确度之间的关系可以如下所示：例如，第一编码码率越大，第一准确度就越高。又例如，在第一信道信息相同(如信道质量相同)的情况下，第一编码码率越大，第一准确度就越高。又例如，在第一准确度相同的情况下，第一信道信息所表示的信道质量越好，第一编码码率就越小。例如，第一编码码率、第一准确度和第一信道信息之间的关系可以用函数的形式表示。示例性的，第一编码码率＝f 8(第一准确度，第一信道信息)。又例如，第一编码码率、第一准确度和第一信道信息之间的关系可以用表的形式表示等。关于第一编码码率、第一准确度和第一信道信息之间的关系可以参考下文所示的关于第一编码码率、第一调制阶数、第一准确度和第一信道信息之间的关系。
在一种可能的实现方式中，第一编码码率、第一调制阶数和第一准确度要求之间的关系可以用函数的形式表示。示例性的，(第一调制阶数，第一编码码率)＝f 9(第一准确度，第一信道信息)。
在另一种可能的实现方式中，第一编码码率、第一调制阶数、第一准确度和第一信道信息之间的关系可以用表的形式表示，如表5和表6所示。表5示出的是当第一准确度等于99％时，第一编码码率、第一调制阶数和第一信道信息(如用SNR表示)之间的关系。表6示出的是当第一准确度等于90％时，第一编码码率、第一调制阶数和第一信道信息之间的关系。
表5
索引 0 1 2 3 4 5 6 …  
第一调制阶数 2 2 2 4 4 6 6 …  
第一编码码率 1/3 1/2 2/3 1/2 2/3 1/2 2/3    
第一信噪比 -4 -3 -2 -1 0 1 2 …  
表6
索引 0 1 2 3 4 5 6 …  
第一调制阶数 2 4 4 6 6 6 8 …  
第一编码码率 1/2 1/2 2/3 1/2 2/3 5/6 2/3    
第一信噪比 -4 -3 -2 -1 0 1 2 …  
示例性的，在上述第一准确度用MSE表示的情况下，第一编码码率、第一调制阶数、第一准确度和第一信道信息之间的关系还可以如表7和表8所示。示例性的，表7示出的是当MSE＝0.01时，调制阶数、编码码率和信道信息之间的关系。表8示出的是当MSE＝0.1时，调制阶数、编码码率和信道信息之间的关系。
表7
索引 0 1 2 3 4 5 6 …  
第一调制阶数 2 2 2 4 4 4 6 …  
第一编码码率 1/2 2/3 3/4 1/2 2/3 3/4 2/3    
第一信噪比 -4 -3 -2 -1 0 1 2 …  
表8
索引 0 1 2 3 4 5 6 …  
第一调制阶数 2 4 4 4 6 6 8 …  
第一编码码率 2/3 1/2 2/3 3/4 2/3 5/6 2/3    
第一信噪比 -4 -3 -2 -1 0 1 2 …  
对于弹性的第一准确度来说，第一调制阶数、第一编码码率、第一准确度和第一信道信息之间的关系还可以如表9所示。表9所示的表格示出的是各信道质量下最低可接受的第一准确度对应的第一调制阶数和第一编码码率。由此，可以根据传输资源的使用情况和时延的需求等条件，选择不低于表9所示的第一调制阶数和第一编码码率。或者，对于弹性准确度的服务，可以只定义信道质量的等级指示，和最低可接受特征空间的距离(如MSE等)对应的第一调制阶数和第一编码码率。
关于弹性的第一准确度的具体说明还可以参考下文图6所示的方法。
表9
索引 0 1 2 3 4 5 6 …  
第一调制阶数 2 2 2 4 4 6 6 …  
第一编码码率 1/3 1/2 2/3 1/2 2/3 1/2 2/3    
第一信噪比 -4 -3 -2 -1 0 1 2 …  
可理解，对于准确度要求为100％或特征空间距离为0的神经网络处理服务，可以复用一般的传输参数表(如MCS对应的表格等)，以提供可靠传输服务。
可理解，本申请所示的各个表格仅为示例，不应将本申请所示的表格理解为对本申请的限定。可选的，表1至表9中，还可以不包括第一信噪比(也可以理解为第一信噪比仅仅是制定表格时的参考)。如表1至表4中可以包括第一变换系数和对应的索引；表5至表9中可以包括第一调制阶数、第一编码码率和对应的索引。关于实现方式二的具体说明还可以参考下文图4至图6所示的方法。
结合上文所示的方法，本申请实施例还提供了以下几种数据传输方法。
图4是本申请实施例提供的一种数据处理方法的流程示意图。如图4所示，该方法包括：
401、第一通信装置向第二通信装置发送第一请求消息，该第一请求消息用于请求神经网络处理。对应的，第二通信装置接收第一请求消息。
也就是说，图4所示的方法可以理解为是第一通信装置请求第二通信装置中的神经网络处理数据(也可以简称为节点请求神经网络处理模式)。即第一通信装置提供数据，第二通信装置提供神经网络处理服务。
可理解，本申请实施例示出的第一通信装置还可以称为第一节点或发送端(即提供数据的设备)，第二通信装置还可以称为第二节点或接收端(即对数据进行神经网络处理的设备)。例如，第一通信装置可以为终端设备等，第二通信装置可以为接入网设备等。
402、第二通信装置向第一通信装置发送第一响应消息。对应的，第一通信装置接收第一响应消息。
关于第一请求消息和第一响应消息中的内容可以有如下几种实现方式：
1、第一请求消息包括第一指示信息，该第一指示信息用于指示第一准确度。第一准确度可以为第一通信装置对神经网络处理数据的准确度的要求，或者，第一通信装置预期的准确度等。例如，第一准确度可以包括99％、95％、90％、85％、80％、75％、70％或65％等中的任一项。或者，第一准确度还可以有其他的划分方式，如第一准确度包括90％、80％、70％、60％、50％等中的任一项。可理解，第一指示信息还可以通过索引的方式指示第一准确度。例如，该索引可以是如表1至表9所示的表格的索引(即通过表格的索引指示第一准确度)，或者，也可以是其他类型的索引等，本申请实施例不作限定。
2、第一请求消息包括第一指示信息，该第一指示信息用于指示第一参数。可选的，该第一参数可以包括第一变换系数，或者包括第一编码码率和第一调制阶数等。可选的，第一指示信息可以通过索引的方式指示第一参数。如该第一指示信息用于指示第一索引。例如，该第一索引可以是如表1至表9所示的索引，如表1中的索引0至6中的任一个等。或者，也可以是其他类型的索引等，本申请实施例不作限定。可理解，本申请上文所示的表格的索引仅为示例，如表1至表3所示的索引还可以为连续的索引等，本申请实施例对此不作限定。又如，表1至表9所示的索引可以为连续的索引等。又如，表5和表6所示的索引可以为连续的索引等。
可理解，当第一指示信息用于指示第一参数或第一索引时，则表示第一通信装置预先已经获得了第一信道信息。
3、第一响应消息包括确认指示信息，该确认指示信息用于确认第一指示信息。例如，确认指示信息用于确认第一通信装置所指示的第一准确度(或者确认第一通信装置使用第一准确度确定第一参数)。例如，第一通信装置为接入网设备，第二通信装置为终端设备，则接入网设备可以根据终端设备所反馈的确认指示信息，根据第一准确度自主确定第一参数。又例如，该确认指示信息可以用于确认第一通信装置使用第一参数或第一索引等。可理解，这里所示的确认指示信息可以是一个比特位，该一个比特位的取值为1时，可以用于确认第一通信装置可以使用通过第一指示信息所指示的内容；该一个比特位的取值为0时，则表示第一通信装置不可以使用通过第一指示信息所指示的内容。该情况下，第一响应消息中还可以包括第二指示信息等。
4、第一响应消息包括第二指示信息，在第一指示信息用于指示第一准确度时，第二指示信息可以用于指示第一参数。例如，第一通信装置为终端设备，第二通信装置为接入网设备，则接入网设备可以根据终端设备发送的第一准确度确定第一参数。例如，在接入网设备预先获得了第一信道信息的情况下，该接入网设备可以根据第一信道信息和第一准确度确定第一参数，从而向终端设备发送第一响应消息。然而，在接入网设备接收到第一请求消息之前，未获得第一信道信息的情况下，上述第一请求消息还可以包括参考信号(reference signal，RS)。由此，接入网设备可以根据该参考信号进行信道估计，从而获得第一信道信息。然后根据第一信道信息以及第一准确度确定第一参数。
5、第一响应消息包括第二指示信息，该第二指示信息用于指示第二参数。例如，第二通信装置在获得第一指示信息后，可以根据其对神经网络准确度的需求重新确定变换系数或编码码率等。例如，第一通信装置为终端设备，第二通信装置为接入网设备，则接入网设备可以根据终端设备发送的第一指示信息重新确定第二参数。该第二参数包括第二变化系数；或者，包括第二调制阶数和第二编码码率等。
可选的，上述第一指示信息、第二指示信息或确认指示信息等可以承载在控制信息中，如该控制信息可以包括下行控制信息(downlink control information，DCI)、上行控制信息 (uplink control information，UCI)或侧行链路控制信息(sidelink control information，SCI)中的任一项或多项。可选的，上述第一指示信息、第二指示信息或确认指示信息等可以承载在无线资源控制(radio resource control，RRC)信令中。可选的，上述第一指示信息、第二指示信息或确认指示信息等可以承载在人工智能(artificial intelligence，AI)专用控制信息中或者专用控制信道上等。
可选的，第一通信装置和第二通信装置可以预先协商参数的具体类型。例如，参数的类型包括变换系数、编码码率或调制阶数中的一项或多项。可选的，参数的类型可以由接入网设备确定。一般的，基于编码调制的数据处理方法可以提供可靠传输，在高信噪比时可以达到更高的推理准确度。但是，在低信噪比时性能下降更快，因此更适合基于直接调制的数据处理方法。因此，可选的，第一通信装置或第二通信装置可以根据信道信息的大小自动切换参数的类型。通过信道信息的大小切换参数的类型可以在不同的信噪比条件下提供更合适的数据处理方法，从而获得更高的推理准确度。示例性的，当信道质量大于质量阈值的情况下，参数的类型包括编码码率或调制阶数中的一项或多项。当信道质量小于质量阈值的情况下，参数的类型包括变换系数。可理解，当信道质量等于质量阈值时，对于参数的类型不作限定，如可以预先设置等，本申请实施例不作限定。如信道质量可以包括SNR或SINR等，如信道质量包括SNR时，质量阈值可以包括5dB、8dB、10dB或15dB中的任一项。
403、第一通信装置向第二通信装置发送第一传输符号，对应的，第二通信装置接收第一传输符号。
上述步骤403还可以理解为：根据第一响应消息向第二通信装置发送第一传输符号。关于第一通信装置获得第一传输符号的具体说明可以参考上文如图2或图3等的描述，这里不再详述。例如，第一通信装置可以根据第一响应消息获得第一传输符号，然后发送该第一传输符号等。
404、第二通信装置根据神经网络对输入进行处理，获得处理结果和该处理结果的第二准确度。
可理解，关于第二通信装置对第一传输符号进行处理，获得神经网络的输入的具体方法可以参考上文，这里不再详述。关于第二通信装置获得处理结果的具体方式可以参考上文，这里不再详述。
示例性的，处理结果的第二准确度可以用处理结果的概率、置信度、分类精度或MSE等中的任一项或多项表示。因此，为便于描述，下文将以第二准确度用概率表示、预设条件以预设概率为例说明第二通信装置的反馈方法。
405、在处理结果的概率满足预设概率的情况下，第二通信装置向第一通信装置反馈处理结果。
可选的，预设概率可以是一个数值，如处理结果的概率大于或等于该预设概率，则表示其满足预设概率。可选的，预设概率可以是一个概率范围，如处理结果的概率在这个概率范围之内，则表示其满足预设概率。可理解，关于预设概率的具体方式本申请实施例不作限定。
406、在处理结果的概率不满足预设概率的情况下，第二通信装置向第一通信装置发送重传指示信息，该重传指示信息用于指示重传第一数据。
可选的，在处理结果的概率不满足预设概率的情况下，第二通信装置也可以向第一通信装置指示新的参数或参数的调整方向等。例如，新的参数可以是第二参数(如第二变换系数，或者，第二编码码率和第二调制阶数等)等。参数的调整方向如可以是增大变换系数或减小变换系数等。可选的，第二通信装置还可以通过第二参数的索引指示新的参数。可理解，第 一通信装置获取到新的参数后，还可以根据该新的参数重传第一数据。
可选的，在第二通信装置得到重传的数据，再次得到处理结果和该处理结果的概率之后，当再次得到的概率与前一次得到的概率之间的差小于一定门限，则可以认为重传数据对神经网络的性能没有提升，由此可以终止重传。
示例性的，上述预设概率可以由第一通信装置设置、或者由第二通信装置设置，或者由第一通信装置与第二通信装置协商等，本申请实施例对于该预设概率的具体设置方法不作限定。
可理解，以上所示的概率仅为示例，如神经网络的输出还可以是处理结果的熵、置信度等，这里不再详述。
可理解，本申请实施例所示的各个消息中是否包括RS不作限定。例如，第一请求消息可以包括RS，以便于第二通信装置进行信道估计。又例如，第一响应消息可以包括RS，以便于第一通信装置进行信道估计，获得第一通信装置与第二通信装置之间的信道信息。又例如，第一通信装置发送第一传输符号时，可以包括RS。又例如，第一通信装置重传第一数据时，也可以同时发送RS等，本申请实施例对此不作限定。当通过RS获得的信道信息更新时，第一参数可以随着更新。
本申请实施例中，第一通信装置在传输数据时，对数据的处理只需要符合第一准确度以及第一信道信息，第二通信装置就可以对数据进行神经网络处理，改善了由于需要保证数据传输的正确性而重传数据的情况，有效提高了无线资源的利用效率。
图5是本申请实施例提供的一种数据处理方法的流程示意图，如图5所示，该方法包括：
501、第二通信装置向第一通信装置发送第二请求消息，该第二请求消息用于请求数据。对应的，第一通信装置接收第二请求消息。
也就是说，图5所示的方法可以理解为是第二通信装置向第一通信装置请求数据，以便进行神经网络处理(也可以简称为节点请求处理数据模式)。即第二通信装置需要第一通信装置的数据进行神经网络处理。
本申请实施例中，第二请求消息包括第三指示信息，该第三指示信息用于指示第一参数。可理解，关于第一参数的类型的确定方法可以参考图4，这里不再详述。可理解，关于第三指示信息用于指示第一参数的具体说明可以参考图4所示的相关描述，这里不再详述。
502、第一通信装置向第二通信装置发送第一传输符号，第二通信装置接收第一传输符号。
上述步骤502还可以理解为：第一通信装置根据第二请求消息向第二通信装置发送第一传输符号。关于第一通信装置获得第一传输符号的具体说明可以参考上文如图2至图4中的描述等，这里不再详述。
503、第二通信装置根据神经网络对输入进行处理，输出处理结果和该处理结果的第二准确度。
504、在处理结果的概率满足预设概率的情况下，第二通信装置向第一通信装置反馈处理结果。
505、在处理结果的概率不满足预设概率的情况下，第二通信装置向第一通信装置发送重传指示，该重传指示用于指示重传第一数据。
可理解，关于步骤503至步骤505的具体说明可以参考上文，这里不再详述。
本申请实施例中，第一通信装置在传输数据时，对数据的处理只需要符合第一准确度以及第一信道信息，第二通信装置就可以对数据进行神经网络处理，改善了由于需要保证数据 传输的正确性而重传数据的情况，有效提高了无线资源的利用效率。
图6是本申请实施例提供的一种数据处理方法的流程示意图，如图6所示，该方法包括：
601、第一通信装置向第二通信装置发送第三请求消息，该第三请求消息用于请求神经网络处理。对应的，第二通信装置接收第三请求消息。
该第三请求消息包括第一传输符号。可选的，该第三请求消息还包括参考信号。也就是说，对于弹性准确度的服务，第一通信装置发起神经网络请求的同时，可以将第一传输符号发送给第二通信装置。
可选的，第一通信装置向第二通信装置发送第三请求消息之前，第一通信装置还可以向第二通信装置发送第四请求消息，该第四请求消息用于请求神经网络处理。示例性的，该第四请求消息可以包括RS。第二通信装置在接收到第四请求消息之后，向第一通信装置发送第四响应消息，该第四响应消息可以包括用于指示第一参数的信息。
可选的，第三请求消息可以包括用于指示第一参数的信息。
可理解，图6所示的第三请求消息是以包括第一传输符号和用于指示第一参数的信息为例示出的，但是不应将该第三请求消息包括的内容理解为对本申请实施例的限定。
可理解，关于弹性准确度的具体说明可以参考上文，如表4或表9的相关描述等，这里不再详述。
602、第二通信装置根据神经网络对输入进行处理，输出处理结果。
可理解，关于步骤602的具体说明可以参考上文，如图4或图5的相关描述等，这里不再详述。例如，第二通信装置可以直接向第一通信装置反馈处理结果等。可选的，神经网络还可以输出处理结果的第二准确度，该情况下，第二通信装置可以直接向第一通信装置反馈处理结果和第二准确度。可选的，对于神经网络是否根据第二准确度反馈重传指示等，本申请实施例不作限定。
本申请实施例中，第一通信装置在传输数据时，可以自主确定第一参数，或者由第二通信装置自主确定第一参数，从而第一通信装置或第二通信装置不需要相互协商第一参数，降低了信令开销；而且，第二通信装置接收到第一传输符号便可以进行神经网络的处理，改善了由于需要保证数据传输的正确性而重传数据的情况，有效提高了无线资源的利用效率。
本申请所示的各个实施例中，用于承载第一参数的信息还可以称为第二控制信息。示例性的，上述图4所示的第一请求消息中的第一指示信息可以称为第二控制信息。又如，第一响应消息中的第二指示信息也可以称为第二控制信息。又如，第二请求消息中的第三指示信息也可以称为第二控制信息。又如，第三请求消息中用于指示第一参数的信息也可以称为第二控制信息。又如，第四请求消息中用于指示第一参数的信息也可以称为第二控制信息。由此，本申请还提供了一种帧结构的结构示意图。
可选的，如图7a或图7b所示，该帧中可以包括第一控制信息和第二控制信息。例如，图7a所示的一个第二控制信息可以理解为是第一通信装置向一个第二通信装置发送的第一指示信息，或者，是第二通信装置向一个第一通信装置发送的第二指示信息。图7b所示的两个第二控制信息可以理解为是第一通信装置向两个第二通信装置分别发送的第一指示信息，或者是第二通信装置向两个第一通信装置分别发送的第二指示信息。
可选的，如图7c或图7d所示，图7c或图7d示出的是控制信息和数据同时发送时的帧结构示意图。例如，图7c或图7d示出的帧结构可以适用于上文所示的第三请求消息同时包 括第一传输符号和用于指示第一参数的信息的场景。
示例性的，上述第二控制信息可以用于指示第一参数或第一准确度等。
在一种可能的实现方式中，第一控制信息可以用于指示待神经网络处理的数据占用的资源位置。例如，第一控制信息可以用于指示第一传输符号占用的时频资源。也就是说，将传统的资源调度控制与本申请所示的神经网络处理控制分开。即第一控制信息可以用于指示帧中包括的内容是本申请所示的数据处理方法中涉及的相关参数(如第一参数)和相关数据(如第一传输符号)。
在另一种可能的实现方式中，第一控制信息可以用于指示第二控制信息是传统数据传输的控制信息或神经网络处理的控制信息。传统数据传输的控制信息如可以理解为无线传输设计中的下行控制信息或上行控制信息等，神经网络处理的控制信息可以理解为与本申请所示的实施例相关的神经网络处理的信息，如第一参数或第一准确度等。
也就是说，由于本申请所提供的面向神经网络处理的数据处理方法，包括传输符号、第一参数或第一准确度等，因此在与传统数据传输共享无线频谱资源时，可以通过单独的第一控制信息指示。一种实现方式是将资源调度控制和神经网络处理控制分开，如帧结构头部的第一控制信息可以用于指示神经网络处理的数据所占用的资源位置，第二控制信息用于指示相关参数或准确度等信息。另一种实现方式是复用控制信息，如通过标识指示第一控制信息中的内容是传统数据处理的控制信息或神经网络处理的控制信息。
本申请所示的各个实施例中，由于本申请所示的数据处理方法的传输性能指标(如第一准确度等)和传统数据处理的指标(如误码率或误块率等)不同，因此，本申请实施例还提供了与神经网络处理对应的传输信道或物理信道等。
在一种可能的实现方式中，如图8a所示，根据本申请上文所示的实施例获得的传输符号(如包括第一传输符号)可以承载在传输信道上，如对于第一数据直接调制成的第一传输符号可以承载在上下行的神经网络共享信道(neural network shared channel，NNSCH)(也可以称为神经网络处理共享信道)。例如，终端设备向接入网设备发送第一传输符号时，该第一传输符号可以承载在上行的NNSCH；接入网设备向终端设备发送第一传输符号时，该第一传输符号可以承载在下行的NNSCH。根据本申请上文所示的实施例获得的相关参数，如第一参数(或第二参数等)可以承载在神经网络控制信道(neural network control channel，NNCCH)(也可以称为神经网络处理控制信道)上传输。该NNCCH可以包括上行的NNCCH和下行的NNCCH。可理解，以上所示的上行可以指终端设备向接入网设备发送信息，下行可以指接入网设备向终端设备发送信息。然而，当第一通信装置和第二通信装置都是终端设备时，以上所示的NNSCH可以不区分上下行，NNCCH也可以不区分上下行。该情况下，上述NNSCH也可以称为侧行链路NNSCH，NNCCH也可以称为侧行链路NNCCH。可理解，NNSCH不仅可以用于传输与神经网络处理相关的数据，还可以用于传输神经网络的模型数据等。NNCCH不仅可以用于传输与神经网络处理相关的控制信息，还可以用于传输NNSCH的相关参数或者其他神经网络处理相关的信令等，本申请实施例对此不作限定。
由于NNCCH需要可靠传输，因此其可以承载在专用的物理神经网络处理控制信道(physical neural network control channel，PNNCCH)上，或者，可以承载在传统的数据传输物理信道上，如物理下行共享信道(physical downlink shared channel，PDSCH)或物理上行共享信道(physical uplink shared channel，PUSCH)。NNSCH可以承载在物理神经网络处理共享信道(physical neural network shared channel，PNNSCH)上，从而提供准确度灵活可控 的传输。一般的，专用业务信道(dedicated traffic channel，DTCH)可以承载在下行共享信道(downlink shared channel，DL-SCH)和上行共享信道(uplink shared channel，UL-SCH)上。如DL-SCH和UL-SCH可以分别通过PDSCH和PUSCH承载。
可选的，本申请实施例中的参考信号可以包括神经网络处理参考信号，该神经网络处理参考信号可以用于神经网络处理过程中的信道估计和准确度估计。
可理解，图8b是本申请实施例提供的与神经网络处理对应的层次示意图。如图8b所示，终端设备和接入网设备(如图8b中的gNB)中可以包括与神经网络处理业务对应的层，该与神经网络处理业务对应的层位于物理层(physical，PHY)之上。该与神经网络处理业务对应的层可以用于对第一数据进行处理，或者，对第一传输符号进行处理，从而输入至神经网络等。本申请实施例对于该与神经网络处理业务对应的层的具体名称不作限定。示例性的，第一通信装置从与神经网络处理业务对应的层获得的第一数据可以直接通过物理层传输。第二通信装置获得第一传输符号之后，可以通过物理层处理后直接传输给与神经网络处理业务对应的层。
在另一种可能的实现方式中，如图8c所示，本申请上文所示的各个实施例中，可以定义专用的神经网络处理逻辑信道、神经网络处理传输信道和神经网络处理控制信道。示例性的，神经网络处理逻辑信道可以包括NNCCH或NNTCH等。神经网络处理传输信道可以包括NNSCH(包括上下行)，神经网络处理逻辑信道可以承载在DL-SCH、UL-SCH或NNSCH上。神经网络处理物理信道可以包括PNNSCH，神经网络传输信道可以承载在PNNSCH上等。图8c所示的NNCCH可以是神经网络控制信道，承载在DL-SCH和UL-SCH上。
例如，神经网络处理逻辑信道用于承载神经网络处理数据。例如，该逻辑信道可以用于承载神经网络处理的数据(如第一传输符号)。又如，该逻辑信道可以用于承载返回神经网络的处理结果等。可选的，神经网络处理传输信道可以用于承载神经网络处理逻辑信道，神经网络处理物理信道可以用于承载神经网络处理传输信道。本申请提供的信道，根据准确度需求，可以提供相应的差错控制水平和差错水平检测，即使有传输差错时，仍可向上层提供数据传输的服务。可选的，神经网络处理逻辑信道可承载在提供无差错服务的数据共享信道(如PDSCH或PUSCH)上，或者，可承载在神经网络处理共享信道(如NNSCH)上。可选的，数据传输也可以承载在神经网络处理传输信道上。神经网络处理控制信道可以用于传输神经网络处理相关的配置和参数等，如该神经网络处理控制信道可以承载在传统的数据共享信道(如DL-SCH或UL-SCH)上。
为进一步理解本申请实施例提供的数据处理方法，以下结合具体的实施例说明。
图9a是本申请实施例提供的一种数据处理方法的场景示意图。如图9a所示，第一通信装置对数据的处理过程可以包括：嵌入和预处理等；第二通信装置对数据的处理过程包括：后处理和推理等。或者，图9a还可以理解为：第一通信装置包括嵌入模块和预处理模块等，第二通信装置包括后处理模块和推理模块等。可理解，这里所示的各个模块可以是功能模块，该功能模块可以采用硬件的形式实现，或者，也可以采用软件功能模块的形式实现等，本申请实施例对此不作限定。由于第一通信装置可以不进行嵌入操作，因此图9a所示的嵌入用虚线表示。
图9a可以理解为是一种基于直接调制的数据处理方法。如图9a所示，第一通信装置首先将输入(如第一数据)嵌入到连续(如连续的实数或复数等)空间，根据神经网络处理数据的第一准确度进行预处理(即可以理解为是根据第一准确度确定第一参数，从而进行预处 理)，将输入变换到第一传输符号，以实现维度适配。第二通信装置对经过信道传输的数据(如第一传输符号)进行后处理，获得神经网络的输入。然后神经网络(如图9a所示的推理)对该输入进行处理，获得处理结果以及与该处理结果对应的第二准确度。第二通信装置还可以根据处理结果对应的第二准确度调整相关参数(如调整第一参数)或者指示重传；或者，第二通信装置反馈处理结果等。该第二通信装置对经过信道传输的数据进行后处理包括：该第二通信装置根据第一参数对经过信道传输的第一传输符号进行逆变换，如进行线性变换或非线性变换等。例如，第一通信装置根据第一变换系数将第一数据变换到第一传输符号，则第二通信装置可以根据该第一变换系数将其接收到的经过信道传输的第一传输符号逆变换到数据。
图9b是本申请实施例提供的另一种数据处理方法的场景示意图。如图9b所示，第一通信装置对数据的处理过程可以包括：嵌入、信源信道编码和正交振幅调制(quadrature amplitude modulation，QAM)等。第二通信装置对数据的处理过程可以包括：QAM解调、信源信道解码和推理等。或者，图9b还可以理解为：第一通信装置包括嵌入模块、信源信道编码模块和QAM调制模块等；第二通信装置包括QAM解调模块、信源信道解码模块和推理模块等。可理解，这里所示的QAM调制仅为示例，本申请实施例还适用于其他调制方式，如二进制相移键控(binary phase shift keying，BPSK)或正交相移键控(quadrature phase shift keying，QPSK)等。
可理解，图9b是以有信源信道编码为例示出的，如信源信道编码可以实现对嵌入数据的变换(如实数到实数的变换)和量化，并映射到调制符号。然而，如果没有信源信道编码，则本申请实施例提供的方法可以包括：通过编码将嵌入数据映射到调制符号等。
图9b可以理解为是基于传统数据传输的部分或全部模块，但是，相关参数或重传等是根据本申请实施例提供的数据处理方法确定的(如基于第一准确度或第二准确度确定的)。如图9b所示，第一通信装置首先将输入嵌入到连续空间，根据神经网络处理数据的第一准确度进行编码调制，以获得第一传输符号。第二通信装置对经过信道传输的数据(如第一传输符号)进行解调解码，获得神经网络的输入。然后神经网络对输入进行处理，获得处理结果以及与该处理结果对应的第二准确度。第二通信装置还可以根据处理结果对应的第二准确度调整相关参数(如调整第一参数)或者指示重传；或者，第二通信装置反馈处理结果等。
图9c是本申请实施例提供的又一种数据处理方法的场景示意图。图9c可以理解为基于直接调制的数据处理方法，或者是复用部分传统(如预处理和后处理可以是传统的调制解调等)的数据处理方法。但是，相关参数或重传等是根据本申请实施例提供的数据处理方法确定的(如基于特征空间的距离确定的)。可理解，关于图9c的具体说明可以参考图9a，这里不再详述。例如，图9c所示的后处理可以包括信道估计和差错估计，该差错估计可以用于表示神经网络输入数据的第一准确度，即该差错估计可以用于对第一准确度进行估计。或者，差错估计还可以用于表示处理结果的第二准确度，从而第二通信装置可以确定是否需要重传及调整参数等。例如，第一通信装置采用直接调制时(如该直接调制是在连续空间的变换)，可以将输入(如图9a所示的第一通信装置的输入)嵌入到连续空间(如图9a所示的嵌入操作)，然后将嵌入后的输出输入至预处理，从而得到第一传输符号。由此第二通信装置根据接收到的经过信道传输的第一传输符号可以估计信道或者估计传输对语义距离的影响，进而根据推理的第二准确度等，调整传输配置(如调整第一参数，以及可选的，向第一通信装置反馈第一参数的调整等)。
示例性的，本申请实施例进行如下定义：嵌入后的输入(即预处理的输入)为特征空间， 即需要进行神经网络推理的特征所在空间。第一传输符号处于信道空间，即信道所在空间。神经网络推理的输出为推理空间，即处理结果所在的空间。则信道空间的距离(或损失)可以为经过信道前后的传输符号之间的距离，特征空间的距离(或损失)为预处理之前和后处理之后特征之间的距离，推理空间的距离(或损失)为神经网络的输入经过传输和不经过传输进行推理的输出之间的距离。由此，信道空间的距离、特征空间的距离和推理距离之间的关系如下所示：如信道空间的距离经过传输预处理和后处理后可以推导出特征空间的距离，特征空间的距离经过神经网络处理可以得到推理空间的距离。
本申请实施例中，可以将传输与推理解耦，可以根据不同的任务定义相应的推理距离(或损失)；从推理距离计算出特征空间的距离，即是对传输的需求；进而可以计算出相应的信道距离。传输参数表以特征空间的距离作为性能指标，定义相应信道质量下的传输参数。也就是说，数据的传输可以对应多种神经网络处理，只要这些神经网络处理对特征空间的距离要求一样，就可以复用相同的传输配置(如第一参数或第一准确度等)。由此，本申请实施例中，可以以特征空间的距离作为重传或参数调整的指标，而特征空间的距离可以决定神经网络推理的距离。同时，神经网络推理的距离要求可以决定特征空间的距离。示例性的，当特征空间的距离以MSE为例时，MSE、编码码率、调整阶数和信道信息之间的关系可以如表7至表9所示。
可理解，本申请实施例所示的预处理的具体说明可以参考上文如图2所示的步骤202的相关描述，后处理的具体说明可以参考上文如图2所示的步骤204的相关描述。关于相关参数(如第一变换系数或第一编码码率等)的确定方法、重传方法或信道的具体说明可以参考上文，这里不再赘述。
图10是本申请实施例提供的一种仿真结果示意图。图10是以分类任务为例，横坐标是信噪比(dB)，纵坐标是神经网络推理，即分类的准确度(accuracy，即分类正确的占比)。虚线部分表示的是直接调制的数据处理方法的性能，实线部分表示的是联合图像专家组(joint photographic experts group，JPEG)(即面向连续色调静止图像的一种压缩方式)压缩加编码调制的性能上限，变换系数(compression ratio，CR)是12，与横轴平行的实线部分表示的是数据直接进行神经网络处理时的性能。从图10可以看出，推理准确度随着信噪比平滑变化，由此可以通过调整参数(如第一参数或第二参数等)达到所需的准确度，或者降低准确度以节省传输资源。另外，在低信噪比时，基于直接调制的数据处理方法(如基于第一变换系数进行的数据处理方法)可以达到更高的准确度；高信噪比时，基于编码调制的数据处理方法的准确度更高。因此可以在不同的信噪比条件下，采用不同的数据处理方法，可以达到更优的性能。
本申请提供的数据处理方法，根据神经网络处理数据的第一准确度设置处理数据的相关参数，从而对数据的处理只需要符合神经网络处理数据的第一准确度即可。改善了由于需要保证数据传输的正确性而不断重传数据的情况，有效提高了无线资源的利用效率。而且由于不需要重传数据直至无差错，因此还降低了端到端的时延。可选的，本申请提供的方法，不需要进行比特级处理，易于评估传输误差对神经网络处理准确度的影响。
以下将介绍本申请实施例提供的通信装置。
本申请根据上述方法实施例对通信装置进行功能模块的划分，例如，可以对应各个功能划分各个功能模块，也可以将两个或两个以上的功能集成在一个处理模块中。上述集成的模块既可以采用硬件的形式实现，也可以采用软件功能模块的形式实现。需要说明的是，本申 请中对模块的划分是示意性的，仅仅为一种逻辑功能划分，实际实现时可以有另外的划分方式。下面将结合图11至图13详细描述本申请实施例的通信装置。
图11是本申请实施例提供的一种通信装置的结构示意图，如图11所示，该通信装置包括处理单元1101和收发单元1102。
在本申请的一些实施例中，该通信装置可以是上文示出的第一通信装置或芯片，该芯片可以应用于(或设置于)第一通信装置中等。即该通信装置可以用于执行上文方法实施例中由第一通信装置执行的步骤或功能等。
处理单元1101，用于获得第一数据，以及根据第一参数对该第一数据进行处理，获得第一传输符号；
收发单元1102，用于向第二通信装置发送第一传输符号。
在一种可能的实现方式中，处理单元1101，具体用于根据第一变换系数对第一数据进行变换，获得第一传输符号；或者，根据第一调制阶数和第一编码码率对第一数据进行编码和调制，获得第一传输符号。
在一种可能的实现方式中，收发单元1102，还用于向第二通信装置发送第一请求消息，该第一请求消息用于请求神经网络处理，该第一请求消息包括第一指示信息，该第一指示信息用于指示第一准确度；
收发单元1102，还用于接收来自第二通信装置的第一响应消息，该第一响应消息包括第二指示信息，该第二指示信息用于指示第一参数。
在一种可能的实现方式中，收发单元1102，还用于接收来自第二通信装置的第二请求消息，该第二请求消息用于请求数据，该第二请求消息包括第三指示信息，该第三指示信息用于指示第一参数。
在一种可能的实现方式中，收发单元1102，还用于在第一数据的处理结果的第二准确度满足预设条件的情况下，接收来自第二通信装置的反馈消息，该反馈消息包括用于指示所述处理结果的信息。
在一种可能的实现方式中，收发单元1102，还用于在第一数据的处理结果的第二准确度不满足预设条件的情况下，接收来自第二通信装置的重传指示信息，该重传指示信息用于指示重传第一数据；
收发单元1102，还用于根据重传指示信息重传第一数据。
本申请实施例中，关于第一参数、第一准确度、第一信道信息、第二准确度、第一指示信息和第二指示信息等的说明可以参考上文方法实施例中的介绍，这里不再一一详述。
可理解，本申请实施例示出的收发单元和处理单元的具体说明仅为示例，对于收发单元和处理单元的具体功能或执行的步骤等，可以参考上述方法实施例，这里不再详述。示例性的，如该处理单元1101还可以用于执行图2所示的步骤201和步骤202，该收发单元1102还可以用于执行图2所示的步骤203中的发送步骤。又如，收发单元1102还可以用于执行图4所示的步骤401中的发送步骤，步骤402中的接收步骤以及步骤403中的发送步骤。又如，收发单元1102还可以用于执行图5所示的步骤501中的接收步骤，以及步骤502中的发送步骤。又如，收发单元1102还可以用于执行图6所示的步骤601中的发送步骤。可理解，收发单元和接收单元还可以用于执行图9a至图9c所示的方法等，这里不再详述。
复用图11，在本申请的另一些实施例中，该通信装置可以是上文示出的第二通信装置或芯片，该芯片可以应用于(或设置于)第二通信装置。即该通信装置可以用于执行上文方法实施例中由第二通信装置执行的步骤或功能等。
收发单元1102，用于接收来自第一通信装置的第一传输符号；
处理单元1101，用于根据第一参数对第一传输符号进行处理，获得神经网络的输入；
处理单元1101，还用于根据所述神经网络对所述输入进行处理，获得处理结果。
在一种可能的实现方式中，处理单元1101，具体用于根据第一变换系数对第一传输符号进行逆变换，获得神经网络的输入；或者，根据第一调制阶数和第一编码码率对第一传输符号进行解调和解码，获得神经网络的输入。
在一种可能的实现方式中，收发单元1102，还用于接收来自第一通信装置的第一请求消息，该第一请求消息用于请求神经网络处理，该第一请求消息包括第一指示信息，该第一指示信息用于指示第一准确度；
收发单元1102，还用于向第一通信装置发送第一响应消息，该第一响应消息包括第二指示信息，该第二指示信息用于指示第一参数。
在一种可能的实现方式中，收发单元1102，还用于向第一通信装置发送第二请求消息，该第二请求消息用于请求数据，该第二请求消息包括第三指示信息，该第三指示信息用于指示第一参数。
在一种可能的实现方式中，收发单元1102，还用于在输入的处理结果的第二准确度满足预设条件的情况下，向第一通信装置发送反馈消息，该反馈消息包括用于指示该处理结果的信息。
在一种可能的实现方式中，收发单元1102，还用于在输入的处理结果的第二准确度不满足预设条件的情况下，向第一通信装置发送重传指示信息，该重传指示信息用于指示重传第一数据，第一传输符号根据该第一数据得到。
本申请实施例中，关于第一参数、第一准确度、第一信道信息、第二准确度、第一指示信息和第二指示信息等的说明可以参考上文方法实施例中的介绍，这里不再一一详述。
可理解，本申请实施例示出的收发单元和处理单元的具体说明仅为示例，对于收发单元和处理单元的具体功能或执行的步骤等，可以参考上述方法实施例，这里不再详述。示例性的，如该处理单元1101还可以用于执行图2所示的步骤204和步骤205，该收发单元1102还可以用于执行图2所示的步骤203中的接收步骤。又如，处理单元1101还可以用于执行图4所示的步骤404至步骤406(如处理单元1101可以用于控制收发单元1102输出处理结果，或重传指示等)。又如，收发单元1102还可以用于执行图5所示的步骤501中的发送步骤，以及步骤502中的接收步骤，处理单元1101还可以用于执行图5所示的步骤503至步骤505。又如，收发单元1102还可以用于执行图6所示的步骤601中的接收步骤，处理单元1101还可以用于执行图6所示的步骤602。可理解，收发单元和接收单元还可以用于执行图9a至图9c所示的方法等，这里不再详述。
以上介绍了本申请实施例的第一通信装置和第二通信装置，以下介绍所述第一通信装置和第二通信装置可能的产品形态。应理解，但凡具备上述图11所述的第一通信装置的功能的任何形态的产品，或者，但凡具备上述图11所述的第二通信装置的功能的任何形态的产品，都落入本申请实施例的保护范围。还应理解，以下介绍仅为举例，不限制本申请实施例的第一通信装置和第二通信装置的产品形态仅限于此。
在一种可能的实现方式中，图11所示的通信装置中，处理单元1101可以是一个或多个处理器，收发单元1102可以是收发器，或者收发单元1102还可以是发送单元和接收单元，发送单元可以是发送器，接收单元可以是接收器，该发送单元和接收单元集成于一个器件，例如收发器。本申请实施例中，处理器和收发器可以被耦合等，对于处理器和收发器的连接方式，本申请实施例不作限定。
如图12所示，该通信装置120包括一个或多个处理器1220和收发器1210。
示例性的，当该通信装置用于执行上述第一通信装置执行的步骤或方法或功能时，处理器1220，用于获得第一数据之后，根据第一参数对该第一数据进行处理，获得第一传输符号；收发器1210，用于向第二通信装置发送该第一传输符号。
示例性的，当该通信装置用于执行上述第二通信装置执行的步骤或方法或功能时，收发器1210，用于接收来自第一通信装置的第一传输符号；处理器1220，用于根据第一参数对该第一传输符号进行处理，获得神经网络的输入，并根据神经网络对该输入进行处理，获得处理结果。
本申请实施例中，关于第一参数、第一准确度、第一信道信息、第二准确度、第一指示信息和第二指示信息等的说明可以参考上文方法实施例中的介绍，这里不再一一详述。
可理解，对于处理器和收发器的具体说明还可以参考图11所示的处理单元和收发单元的介绍，这里不再赘述。
在图12所示的通信装置的各个实现方式中，收发器可以包括接收机和发射机，该接收机用于执行接收的功能(或操作)，该发射机用于执行发射的功能(或操作)。以及收发器用于通过传输介质和其他设备/装置进行通信。
可选的，通信装置120还可以包括一个或多个存储器1230，用于存储程序指令和/或数据。存储器1230和处理器1220耦合。本申请实施例中的耦合是装置、单元或模块之间的间接耦合或通信连接，可以是电性，机械或其它的形式，用于装置、单元或模块之间的信息交互。处理器1220可能和存储器1230协同操作。处理器1220可以执行存储器1230中存储的程序指令。可选的，上述一个或多个存储器中的至少一个可以包括于处理器中。示例性的，该存储器1230中可以存储有第一参数、第一准确度和第一信道信息之间的关系等。
本申请实施例中不限定上述收发器1210、处理器1220以及存储器1230之间的具体连接介质。本申请实施例在图12中以存储器1230、处理器1220以及收发器1210之间通过总线1240连接，总线在图12中以粗线表示，其它部件之间的连接方式，仅是进行示意性说明，并不引以为限。所述总线可以分为地址总线、数据总线、控制总线等。为便于表示，图12中仅用一条粗线表示，但并不表示仅有一根总线或一种类型的总线。
在本申请实施例中，处理器可以是通用处理器、数字信号处理器、专用集成电路、现场可编程门阵列或者其他可编程逻辑器件、分立门或者晶体管逻辑器件、分立硬件组件等，可以实现或者执行本申请实施例中的公开的各方法、步骤及逻辑框图。通用处理器可以是微处理器或者任何常规的处理器等。结合本申请实施例所公开的方法的步骤可以直接体现为硬件处理器执行完成，或者用处理器中的硬件及软件模块组合执行完成等。
本申请实施例中，存储器可包括但不限于硬盘(hard disk drive，HDD)或固态硬盘(solid-state drive，SSD)等非易失性存储器，随机存储记忆体(Random Access Memory，RAM)、可擦除可编程只读存储器(Erasable Programmable ROM，EPROM)、只读存储器(Read-Only Memory，ROM)或便携式只读存储器(Compact Disc Read-Only Memory，CD-ROM)等等。存储器是能够用于携带或存储具有指令或数据结构形式的程序代码，并能够由计算机(如本申请示出的通信装置等)读和/或写的任何存储介质，但不限于此。本申请实施例中的存储器还可以是电路或者其它任意能够实现存储功能的装置，用于存储程序指令和/或数据。
处理器1220主要用于对通信协议以及通信数据进行处理，以及对整个通信装置进行控制，执行软件程序，处理软件程序的数据。存储器1230主要用于存储软件程序和数据。收发器1210可以包括控制电路和天线，控制电路主要用于基带信号与射频信号的转换以及对射频 信号的处理。天线主要用于收发电磁波形式的射频信号。输入输出装置，例如触摸屏、显示屏，键盘等主要用于接收用户输入的数据以及对用户输出数据。
当通信装置开机后，处理器1220可以读取存储器1230中的软件程序，解释并执行软件程序的指令，处理软件程序的数据。当需要通过无线发送数据时，处理器1220对待发送的数据进行基带处理(如本申请上文所示的根据第一参数对第一数据进行处理)后，输出基带信号至射频电路，射频电路将基带信号进行射频处理后将射频信号通过天线以电磁波的形式向外发送。当有数据发送到通信装置时，射频电路通过天线接收到射频信号，将射频信号转换为基带信号，并将基带信号输出至处理器1220，处理器1220将基带信号转换为数据并对该数据进行处理。
在另一种实现中，所述的射频电路和天线可以独立于进行基带处理的处理器而设置，例如在分布式场景中，射频电路和天线可以与独立于通信装置，呈拉远式的布置。
可理解，本申请实施例示出的通信装置还可以具有比图12更多的元器件等，本申请实施例对此不作限定。以上所示的处理器和收发器所执行的方法仅为示例，对于该处理器和收发器具体所执行的步骤可参照上文介绍的方法。
在另一种可能的实现方式中，图11所示的通信装置中，处理单元1101可以是一个或多个逻辑电路，收发单元1102可以是输入输出接口，又或者称为通信接口，或者接口电路，或接口等等。或者收发单元1102还可以是发送单元和接收单元，发送单元可以是输出接口，接收单元可以是输入接口，该发送单元和接收单元集成于一个单元，例如输入输出接口。如图13所示，图13所示的通信装置包括逻辑电路1301和接口1302。即上述处理单元1101可以用逻辑电路1301实现，收发单元1102可以用接口1302实现。其中，该逻辑电路1301可以为芯片、处理电路、集成电路或片上系统(system on chip，SoC)芯片等，接口1302可以为通信接口、输入输出接口、管脚等。示例性的，图13是以上述通信装置为芯片为例出的，该芯片包括逻辑电路1301和接口1302。
本申请实施例中，逻辑电路和接口还可以相互耦合。对于逻辑电路和接口的具体连接方式，本申请实施例不作限定。可选的，图13所示的通信装置还可以包括存储器1303，该存储器可以用于存储第一参数、第一准确度和第一信道信息之间的关系。例如，该存储器可以用于存储如表1至表9，或者，还可以用于存储上文所示的各个公式等。由于图13所示的存储器也可能不与处理器集成于一起，而是位于芯片之外，因此图13所示的存储器用虚线表示。
示例性的，当该通信装置用于执行上述第一通信装置执行的步骤或方法或功能时，逻辑电路1301，用于获得第一数据之后，根据第一参数对该第一数据进行处理，获得第一传输符号；接口1302，用于输出该第一传输符号。示例性的，当该通信装置用于执行上述第二通信装置执行的步骤或方法或功能时，接口1302，用于输入第一传输符号；逻辑电路1301，用于根据第一参数对该第一传输符号进行处理，获得神经网络的输入，并根据神经网络对该输入进行处理，获得处理结果。
可理解，本申请实施例示出的通信装置可以采用硬件的形式实现本申请实施例提供的方法，也可以采用软件的形式实现本申请实施例提供的方法等，本申请实施例对此不作限定。
本申请实施例中，关于第一参数、第一准确度、第一信道信息、第二准确度、第一指示信息和第二指示信息等的说明可以参考上文方法实施例中的介绍，这里不再一一详述。
对于图13所示的各个实施例的具体实现方式，还可以参考上述各个实施例，这里不再详述。
本申请实施例还提供了一种通信系统，该通信系统包括第一通信装置和第二通信装置， 该第一通信装置和该第二通信装置可以用于执行前述任一实施例中的方法。
此外，本申请还提供一种计算机程序，该计算机程序用于实现本申请提供的方法中由第一通信装置执行的操作和/或处理。
本申请还提供一种计算机程序，该计算机程序用于实现本申请提供的方法中由第二通信装置执行的操作和/或处理。
本申请还提供一种计算机可读存储介质，该计算机可读存储介质中存储有计算机程序或计算机可执行指令，当计算机程序或计算机可执行指令在计算机上运行时，使得计算机执行本申请提供的方法中由第一通信装置执行的操作和/或处理。
本申请还提供一种计算机可读存储介质，该计算机可读存储介质中存储有计算机程序或计算机可执行指令，当计算机程序或计算机可执行指令在计算机上运行时，使得计算机执行本申请提供的方法中由第二通信装置执行的操作和/或处理。
本申请还提供一种计算机程序产品，该计算机程序产品包括计算机可执行指令或计算机程序，当该计算机可执行指令或计算机程序在计算机上运行时，使得本申请提供的方法中由第一通信装置执行的操作和/或处理被执行。
本申请还提供一种计算机程序产品，该计算机程序产品包括计算机可执行指令或计算机程序，当该计算机可执行指令或计算机程序在计算机上运行时，使得本申请提供的方法中由第二通信装置执行的操作和/或处理被执行。
在本申请所提供的几个实施例中，应该理解到，所揭露的系统、装置和方法，可以通过其它的方式实现。例如，以上所描述的装置实施例仅是示意性的，例如，所述单元的划分，仅仅为一种逻辑功能划分，实际实现时可以有另外的划分方式，例如多个单元或组件可以结合或者可以集成到另一个系统，或一些特征可以忽略，或不执行。另外，所显示或讨论的相互之间的耦合或直接耦合或通信连接可以是通过一些接口、装置或单元的间接耦合或通信连接，也可以是电的，机械的或其它的形式连接。
所述作为分离部件说明的单元可以是或者也可以不是物理上分开的，作为单元显示的部件可以是或者也可以不是物理单元，即可以位于一个地方，或者也可以分布到多个网络单元上。可以根据实际的需要选择其中的部分或者全部单元来实现本申请实施例提供的方案的技术效果。
另外，在本申请各个实施例中的各功能单元可以集成在一个处理单元中，也可以是各个单元单独物理存在，也可以是两个或两个以上单元集成在一个单元中。上述集成的单元既可以采用硬件的形式实现，也可以采用软件功能单元的形式实现。
所述集成的单元如果以软件功能单元的形式实现并作为独立的产品销售或使用时，可以存储在一个计算机可读取存储介质中。基于这样的理解，该技术方案的全部或部分可以以软件产品的形式体现出来，该计算机软件产品存储在一个可读存储介质中，包括若干指令用以使得一台计算机设备(可以是个人计算机，服务器，或者网络设备等)执行本申请各个实施例所述方法的全部或部分步骤。而前述的可读存储介质包括：U盘、移动硬盘、只读存储器(read-only memory，ROM)、随机存取存储器(random access memory，RAM)、磁碟或者光盘等各种可以存储程序代码的介质。
以上所述，仅为本申请的具体实施方式，但本申请的保护范围并不局限于此，任何熟悉本技术领域的技术人员在本申请揭露的技术范围内，可轻易想到变化或替换，都应涵盖在本申请的保护范围之内。因此，本申请的保护范围应以所述权利要求的保护范围为准。

Claims (44)

  1. 一种数据处理方法，其特征在于，所述方法包括：
    第一通信装置获得第一数据；
    所述第一通信装置根据第一参数对所述第一数据进行处理，获得第一传输符号，所述第一参数根据第一准确度以及第一信道信息确定，所述第一信道信息为所述第一通信装置与第二通信装置之间的信道信息，所述第一准确度用于表示所述第二通信装置中神经网络处理数据的准确度；
    所述第一通信装置向所述第二通信装置发送所述第一传输符号。
  2. 根据权利要求1所述的方法，其特征在于，
    所述第一参数包括第一变换系数，所述第一变换系数用于表示所述第一数据与所述第一传输符号之间的维度比，所述第一通信装置根据第一参数对所述第一数据进行处理包括：
    所述第一通信装置根据所述第一变换系数对所述第一数据进行变换，获得所述第一传输符号；或者，
    所述第一参数包括第一调制阶数和第一编码码率，所述第一通信装置根据第一参数对所述第一数据进行处理包括：
    所述第一通信装置根据所述第一调制阶数和所述第一编码码率对所述第一数据进行编码和调制，获得所述第一传输符号。
  3. 根据权利要求1或2所述的方法，其特征在于，所述方法还包括：
    所述第一通信装置向所述第二通信装置发送第一请求消息，所述第一请求消息用于请求所述神经网络进行数据处理，所述第一请求消息包括第一指示信息，所述第一指示信息用于指示所述第一准确度；
    所述第一通信装置接收来自所述第二通信装置的第一响应消息，所述第一响应消息包括第二指示信息，所述第二指示信息用于指示所述第一参数。
  4. 根据权利要求1或2所述的方法，其特征在于，所述方法还包括：
    所述第一通信装置接收来自所述第二通信装置的第二请求消息，所述第二请求消息用于请求数据，所述第二请求消息包括第三指示信息，所述第三指示信息用于指示所述第一参数。
  5. 根据权利要求1-4任一项所述的方法，其特征在于，所述方法还包括：
    所述第一通信装置接收来自所述第二通信装置的反馈消息，所述反馈消息包括用于指示所述处理结果的信息；或者，
    所述第一通信装置接收来自所述第二通信装置的重传指示信息，所述重传指示信息用于指示重传所述第一数据。
  6. 根据权利要求3-5任一项所述的方法，其特征在于，以下一项或多项信息承载于神经网络处理控制信道中：
    所述第一指示信息、所述第二指示信息、所述第三指示信息或所述重传指示信息。
  7. 根据权利要求1-6任一项所述的方法，其特征在于，所述第一传输符号承载于神经网络处理共享信道。
  8. 根据权利要求5所述的方法，其特征在于，所述重传指示信息还用于指示第二参数，所述第二参数为所述第一参数进行更新后的参数。
  9. 根据权利要求1-8任一项所述的方法，其特征在于，所述第一准确度用以下任一项或多项表示：所述神经网络的处理结果的置信度；所述神经网络的处理结果的分类精度；所述神 经网络的输入数据与所述第一数据之间的均方误差MSE；所述神经网络的输入数据与所述第一数据之间的平均绝对误差MAE。
  10. 一种数据处理方法，其特征在于，所述方法包括：
    第二通信装置接收来自第一通信装置的第一传输符号；
    所述第二通信装置根据第一参数对所述第一传输符号进行处理，获得神经网络的输入，所述第一参数根据第一准确度以及第一信道信息确定，所述第一信道信息为所述第一通信装置与所述第二通信装置之间的信道信息，所述第一准确度用于表示所述神经网络处理数据的准确度；
    所述第二通信装置根据所述神经网络对所述输入进行处理，获得处理结果。
  11. 根据权利要求10所述的方法，其特征在于，
    所述第一参数包括第一变换系数，所述第一变换系数用于表示所述第一数据与所述第一传输符号之间的维度比，所述第二通信装置根据第一参数对所述第一传输符号进行处理，获得神经网络的输入包括：
    所述第二通信装置根据所述第一变换系数对所述第一传输符号进行逆变换，获得所述神经网络的输入；或者，
    所述第一参数包括第一调制阶数和第一编码码率，所述第二通信装置根据第一参数对所述第一传输符号进行处理，获得神经网络的输入包括：
    所述第二通信装置根据所述第一调制阶数和所述第一编码码率对所述第一传输符号进行解调和解码，获得所述神经网络的输入。
  12. 根据权利要求10或11所述的方法，其特征在于，所述方法还包括：
    所述第二通信装置接收来自所述第一通信装置的第一请求消息，所述第一请求消息用于请求所述神经网络处理，所述第一请求消息包括第一指示信息，所述第一指示信息用于指示所述第一准确度；
    所述第二通信装置向所述第一通信装置发送第一响应消息，所述第一响应消息包括第二指示信息，所述第二指示信息用于指示所述第一参数。
  13. 根据权利要求10或11所述的方法，其特征在于，所述方法还包括：
    所述第二通信装置向所述第一通信装置发送第二请求消息，所述第二请求消息用于请求数据，所述第二请求消息包括第三指示信息，所述第三指示信息用于指示所述第一参数。
  14. 根据权利要求10-13任一项所述的方法，其特征在于，所述方法还包括：
    在所述输入的处理结果的第二准确度满足预设条件的情况下，所述第二通信装置向所述第一通信装置发送反馈消息，所述反馈消息包括用于指示所述处理结果的信息；或者，
    在所述输入的处理结果的第二准确度不满足预设条件的情况下，所述第二通信装置向所述第一通信装置发送重传指示信息，所述重传指示信息用于指示重传第一数据，所述第一传输符号根据所述第一数据得到。
  15. 根据权利要求10-14任一项所述的方法，其特征在于，以下一项或多项信息承载于神经网络处理控制信道中：
    所述第一指示信息、所述第二指示信息、所述第三指示信息或所述重传指示信息。
  16. 根据权利要求10-15任一项所述的方法，其特征在于，所述第一传输符号承载于神经网络处理共享信道。
  17. 根据权利要求14所述的方法，其特征在于，所述重传指示信息还用于指示第二参数，所述第二参数为所述第一参数进行更新后的参数。
  18. 根据权利要求10-17任一项所述的方法，其特征在于，所述第一准确度用以下任一项或多项表示：所述神经网络的处理结果的置信度；所述神经网络的处理结果的分类精度；所述神经网络的输入数据与所述第一数据之间的均方误差MSE；所述神经网络的输入数据与所述第一数据之间的平均绝对误差MAE。
  19. 一种第一通信装置，其特征在于，所述装置包括：
    处理单元，用于获得第一数据，以及根据第一参数对所述第一数据进行处理，获得第一传输符号，所述第一参数根据第一准确度以及第一信道信息确定，所述第一信道信息为所述第一通信装置与第二通信装置之间的信道信息，所述第一准确度用于表示所述第二通信装置中神经网络处理数据的准确度；
    收发单元，用于向所述第二通信装置发送所述第一传输符号。
  20. 根据权利要求19所述的装置，其特征在于，所述第一参数包括第一变换系数，所述第一变换系数用于表示所述第一数据与所述第一传输符号之间的维度比，所述处理单元，具体用于根据所述第一变换系数对所述第一数据进行变换，获得所述第一传输符号；或者，
    所述第一参数包括第一调制阶数和第一编码码率，所述处理单元，具体用于根据所述第一调制阶数和所述第一编码码率对所述第一数据进行编码和调制，获得所述第一传输符号。
  21. 根据权利要求19或20所述的装置，其特征在于，
    所述收发单元，还用于向所述第二通信装置发送第一请求消息，所述第一请求消息用于请求所述神经网络进行数据处理，所述第一请求消息包括第一指示信息，所述第一指示信息用于指示所述第一准确度；以及还用于接收来自所述第二通信装置的第一响应消息，所述第一响应消息包括第二指示信息，所述第二指示信息用于指示所述第一参数。
  22. 根据权利要求19或20所述的装置，其特征在于，
    所述收发单元，还用于接收来自所述第二通信装置的第二请求消息，所述第二请求消息用于请求数据，所述第二请求消息包括第三指示信息，所述第三指示信息用于指示所述第一参数。
  23. 根据权利要求19-22任一项所述的装置，其特征在于，
    所述收发单元，还用于接收来自所述第二通信装置的反馈消息，所述反馈消息包括用于指示所述处理结果的信息；或者，
    所述收发单元，还用于接收来自所述第二通信装置的重传指示信息，所述重传指示信息用于指示重传所述第一数据。
  24. 根据权利要求21-23任一项所述的装置，其特征在于，以下一项或多项信息承载于神经网络处理控制信道中：
    所述第一指示信息、所述第二指示信息、所述第三指示信息或所述重传指示信息。
  25. 根据权利要求19-24任一项所述的装置，其特征在于，所述第一传输符号承载于神经网络处理共享信道。
  26. 根据权利要求23所述的装置，其特征在于，所述重传指示信息还用于指示第二参数，所述第二参数为所述第一参数进行更新后的参数。
  27. 根据权利要求19-26任一项所述的装置，其特征在于，所述第一准确度用以下任一项或多项表示：所述神经网络的处理结果的置信度；所述神经网络的处理结果的分类精度；所述神经网络的输入数据与所述第一数据之间的均方误差MSE；所述神经网络的输入数据与所述第一数据之间的平均绝对误差MAE。
  28. 一种第二通信装置，其特征在于，所述装置包括：
    收发单元，用于接收来自第一通信装置的第一传输符号；
    处理单元，用于根据第一参数对所述第一传输符号进行处理，获得神经网络的输入，所述第一参数根据第一准确度以及第一信道信息确定，所述第一信道信息为所述第一通信装置与所述第二通信装置之间的信道信息，所述第一准确度用于表示所述神经网络处理数据的准确度；
    所述处理单元，还用于根据所述神经网络对所述输入进行处理，获得处理结果。
  29. 根据权利要求28所述的装置，其特征在于，所述第一参数包括第一变换系数，所述第一变换系数用于表示所述第一数据与所述第一传输符号之间的维度比，所述处理单元，具体用于根据所述第一变换系数对所述第一传输符号进行逆变换，获得所述神经网络的输入；或者，
    所述第一参数包括第一调制阶数和第一编码码率，所述处理单元，具体用于根据所述第一调制阶数和所述第一编码码率对所述第一传输符号进行解调和解码，获得所述神经网络的输入。
  30. 根据权利要求28或29所述的装置，其特征在于，
    所述收发单元，还用于接收来自所述第一通信装置的第一请求消息，所述第一请求消息用于请求所述神经网络处理，所述第一请求消息包括第一指示信息，所述第一指示信息用于指示所述第一准确度；以及向所述第一通信装置发送第一响应消息，所述第一响应消息包括第二指示信息，所述第二指示信息用于指示所述第一参数。
  31. 根据权利要求28或29所述的装置，其特征在于，
    所述收发单元，还用于向所述第一通信装置发送第二请求消息，所述第二请求消息用于请求数据，所述第二请求消息包括第三指示信息，所述第三指示信息用于指示所述第一参数。
  32. 根据权利要求28-31任一项所述的装置，其特征在于，
    所述收发单元，还用于在所述输入的处理结果的第二准确度满足预设条件的情况下，向所述第一通信装置发送反馈消息，所述反馈消息包括用于指示所述处理结果的信息；或者，
    所述收发单元，还用于在所述输入的处理结果的第二准确度不满足预设条件的情况下，向所述第一通信装置发送重传指示信息，所述重传指示信息用于指示重传第一数据，所述第一传输符号根据所述第一数据得到。
  33. 根据权利要求28-32任一项所述的装置，其特征在于，以下一项或多项信息承载于神经网络处理控制信道中：
    所述第一指示信息、所述第二指示信息、所述第三指示信息或所述重传指示信息。
  34. 根据权利要求28-33任一项所述的装置，其特征在于，所述第一传输符号承载于神经网络处理共享信道。
  35. 根据权利要求32所述的装置，其特征在于，所述重传指示信息还用于指示第二参数，所述第二参数为所述第一参数进行更新后的参数。
  36. 根据权利要求19-35任一项所述的装置，其特征在于，所述第一准确度用以下任一项或多项表示：所述神经网络的处理结果的置信度；所述神经网络的处理结果的分类精度；所述神经网络的输入数据与所述第一数据之间的均方误差MSE；所述神经网络的输入数据与所述第一数据之间的平均绝对误差MAE。
  37. 一种通信装置，其特征在于，包括处理器，所述处理器用于执行计算机指令，以使权利要求1-9任一项所述的方法被执行；或者，以使权利要求10-18任一项所述的方法被执行。
  38. 一种通信装置，其特征在于，包括处理器和存储器；
    所述处理器用于存储计算机执行指令；
    所述处理器用于执行所述计算机执行指令，以使权利要求1-9任一项所述的方法被执行；或者，
    所述处理器用于执行所述计算机执行指令，以使权利要求10-18任一项所述的方法被执行。
  39. 一种通信装置，其特征在于，包括处理器和收发器；
    所述收发器，用于接收信号或发送信号；
    所述处理器，用于执行权利要求1-9任一项所述的方法；或者，用于执行权利要求10-18任一项所述的方法。
  40. 一种通信装置，其特征在于，包括逻辑电路和接口，所述逻辑电路和接口耦合；
    所述接口用于输入第一数据，所述逻辑电路用于按照权利要求1-9任一项所述的方法处理所述第一数据，获得第一传输符号，所述接口还用于输出所述第一传输符号；或者，
    所述接口用于输入第一传输符号，所述逻辑电路用于按照权利要求10-18任一项所述的方法处理所述第一传输符号，获得处理结果。
  41. 一种计算机可读存储介质，其特征在于，所述计算机可读存储介质用于存储计算机程序，当所述计算机程序被执行时，权利要求1-9任一项所述的方法被执行；或者，当所述计算机程序被执行时，权利要求10-18任一项所述的方法被执行。
  42. 一种包含指令的计算机程序产品，其特征在于，当所述指令在计算机上运行时，使得权利要求1-9任一项所述的方法被执行；或者，权利要求10-18任一项所述的方法被执行。
  43. 一种计算机程序，其特征在于，所述计算机程序被运行时，权利要求1-9任一项所述的方法被执行，或者，权利要求10-18任一项所述的方法被执行。
  44. 一种通信系统，其特征在于，所述系统包括第一通信装置和第二通信装置，所述第一通信装置用于执行如权利要求1-9任一项所述的方法，所述第二通信装置用于执行如权利要求10-18任一项所述的方法。
PCT/CN2022/109296 2021-08-04 2022-07-30 数据处理方法及装置 WO2023011381A1 (zh)

Priority Applications (2)

Application Number Priority Date Filing Date Title
EP22852094.6A EP4366253A1 (en) 2021-08-04 2022-07-30 Data processing method and apparatus
US18/431,054 US20240214329A1 (en) 2021-08-04 2024-02-02 Data processing method and apparatus

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202110892508.7A CN115842601A (zh) 2021-08-04 2021-08-04 数据处理方法及装置
CN202110892508.7 2021-08-04

Related Child Applications (1)

Application Number Title Priority Date Filing Date
US18/431,054 Continuation US20240214329A1 (en) 2021-08-04 2024-02-02 Data processing method and apparatus

Publications (1)

Publication Number Publication Date
WO2023011381A1 true WO2023011381A1 (zh) 2023-02-09

Family

ID=85154374

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/109296 WO2023011381A1 (zh) 2021-08-04 2022-07-30 数据处理方法及装置

Country Status (4)

Country Link
US (1) US20240214329A1 (zh)
EP (1) EP4366253A1 (zh)
CN (1) CN115842601A (zh)
WO (1) WO2023011381A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN118019068A (zh) * 2024-04-10 2024-05-10 中国电子科技集团公司第十研究所 一种基于神经网络的链路在线自适应速率控制方法

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20160323067A1 (en) * 2015-04-29 2016-11-03 At&T Intellectual Property I, L.P. System and method for harq in cloud ran with large front haul latency
CN109714086A (zh) * 2019-01-23 2019-05-03 上海大学 基于深度学习的优化mimo检测方法
CN111464465A (zh) * 2020-03-11 2020-07-28 重庆邮电大学 一种基于集成神经网络模型的信道估计方法
CN111713035A (zh) * 2020-04-07 2020-09-25 东莞理工学院 基于人工智能的mimo多天线信号传输与检测技术

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20160323067A1 (en) * 2015-04-29 2016-11-03 At&T Intellectual Property I, L.P. System and method for harq in cloud ran with large front haul latency
CN109714086A (zh) * 2019-01-23 2019-05-03 上海大学 基于深度学习的优化mimo检测方法
CN111464465A (zh) * 2020-03-11 2020-07-28 重庆邮电大学 一种基于集成神经网络模型的信道估计方法
CN111713035A (zh) * 2020-04-07 2020-09-25 东莞理工学院 基于人工智能的mimo多天线信号传输与检测技术

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN118019068A (zh) * 2024-04-10 2024-05-10 中国电子科技集团公司第十研究所 一种基于神经网络的链路在线自适应速率控制方法

Also Published As

Publication number Publication date
CN115842601A (zh) 2023-03-24
EP4366253A1 (en) 2024-05-08
US20240214329A1 (en) 2024-06-27

Similar Documents

Publication Publication Date Title
WO2022012257A1 (zh) 通信的方法及通信装置
CN104412665B (zh) 用于动态可配置空中接口的系统和方法
US20240214329A1 (en) Data processing method and apparatus
WO2018201831A1 (zh) 通信方法和装置
WO2020200014A1 (zh) 通信方法及装置
CN112205045B (zh) 非地面网络中的混合自动重传请求
CN112187419B (zh) 一种针对卫星通信的新型harq传输方法及系统
US20230143476A1 (en) Methods and systems for reducing fronthaul bandwidth in a wireless communication system
WO2018201984A1 (zh) 数据的传输方法和设备
CN112152761B (zh) 一种通信方法、装置及存储介质
CN109565360B (zh) 信息发送方法、接收方法、装置及存储介质
CN117813913A (zh) 使用神经网络进行信源编码的方法和系统
CN115336202A (zh) 一种极化码的编译码方法及装置
EP4228353A1 (en) Method and apparatus for resource determination
CN115943695A (zh) Nr v2x中重新选择侧链路传输资源的方法和设备
CN112187401B (zh) 多时间单元传输方法及相关装置
WO2023066193A1 (zh) 通信方法及装置、系统、存储介质、程序产品
WO2019238014A1 (zh) 反馈信息的传输方法和装置
WO2022110086A1 (zh) 一种通信方法、装置及计算机可读存储介质
WO2020103633A1 (zh) 一种控制信息的发送和接收方法及终端设备
WO2024092415A1 (zh) 一种通信方法及装置
WO2024036526A1 (zh) 一种模型调度方法和装置
WO2023185616A1 (zh) 一种psfch传输方法、装置、芯片及模组设备
CN112423313B (zh) 传输块大小确定方法及装置
WO2024012319A1 (zh) 用于无线通信的电子设备和方法、计算机可读存储介质

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: 22852094

Country of ref document: EP

Kind code of ref document: A1

WWE Wipo information: entry into national phase

Ref document number: 2022852094

Country of ref document: EP

ENP Entry into the national phase

Ref document number: 2022852094

Country of ref document: EP

Effective date: 20240202

NENP Non-entry into the national phase

Ref country code: DE