WO2022247739A1 - 一种数据传输方法和相关装置 - Google Patents

一种数据传输方法和相关装置 Download PDF

Info

Publication number
WO2022247739A1
WO2022247739A1 PCT/CN2022/094051 CN2022094051W WO2022247739A1 WO 2022247739 A1 WO2022247739 A1 WO 2022247739A1 CN 2022094051 W CN2022094051 W CN 2022094051W WO 2022247739 A1 WO2022247739 A1 WO 2022247739A1
Authority
WO
WIPO (PCT)
Prior art keywords
communication device
transmission resource
data
machine learning
learning model
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Ceased
Application number
PCT/CN2022/094051
Other languages
English (en)
French (fr)
Inventor
王坚
徐晨
张公正
乔云飞
李榕
王俊
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Huawei Technologies Co Ltd
Original Assignee
Huawei Technologies Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Huawei Technologies Co Ltd filed Critical Huawei Technologies Co Ltd
Priority to EP22810466.7A priority Critical patent/EP4340283A4/en
Publication of WO2022247739A1 publication Critical patent/WO2022247739A1/zh
Priority to US18/514,066 priority patent/US20240089742A1/en
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W16/00—Network planning, e.g. coverage or traffic planning tools; Network deployment, e.g. resource partitioning or cells structures
    • H04W16/18—Network planning tools
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L5/00—Arrangements affording multiple use of the transmission path
    • H04L5/003—Arrangements for allocating sub-channels of the transmission path
    • H04L5/0053—Allocation of signalling, i.e. of overhead other than pilot signals
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/0499—Feedforward networks
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/09—Supervised learning
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W28/00—Network traffic management; Network resource management
    • H04W28/02—Traffic management, e.g. flow control or congestion control
    • H04W28/06—Optimizing the usage of the radio link, e.g. header compression, information sizing, discarding information
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W72/00—Local resource management
    • H04W72/50—Allocation or scheduling criteria for wireless resources

Definitions

  • the embodiments of the present application relate to the communication field, and in particular, to a data transmission method and a related device.
  • a machine learning model is a common tool in the AI field, and the machine learning model includes an input layer, an output layer, and at least one intermediate layer. Specifically in the field of communication, the input of the sending end can be used as the input layer, the output of the receiving end can be used as the output layer, and some machine learning models deployed on the sending end and the receiving end can be used as the middle layer.
  • the end-to-end machine learning model forms an end-to-end machine learning model system.
  • the results of forward inference as intermediate data and gradients of backpropagation are directly transmitted between layers.
  • the transmission of the result of forward reasoning and the transmission of the gradient of backpropagation need to go through the communication channel, which will affect the transmission, and then It also affects the training of the end-to-end machine learning model system. Since the channel state may change at any time, the corresponding machine learning model system will also change, and it may be necessary to initiate the training of the machine learning models at both ends of the transceiver at any time.
  • training a machine learning model requires the sender and the receiver to transmit intermediate data interactively, which means that the sender and receiver need to allocate transmission resources for the transmission of intermediate data.
  • the machine learning model system needs to undergo multiple interactions of intermediate data to complete the training, so multiple transmission resource allocations and scheduling are required, and transmission resource allocation and scheduling occupy a large signaling overhead.
  • the embodiment of the present application provides a data transmission method and related devices, which are used to realize the training of the machine learning model deployed at the sending and receiving ends in the end-to-end communication system with low signaling overhead, and complete the data transmission at the sending and receiving ends of the communication system .
  • the first aspect of the embodiment of the present application provides a data transmission method, which is applied to a first communication device in a communication system, and the communication system further includes a second communication device; wherein, the first communication device is deployed with a first machine Learning model, the second communication device is deployed with a second machine learning model, the first machine learning model and the second machine learning model are used to realize the communication between the first communication device and the second communication device, the method includes:
  • the first communication device acquires the first information, and the first information includes indication information of the first transmission resource and the second transmission resource; wherein, the first transmission resource is used to transmit the forward reasoning result, and the second transmission resource is used to transmit the reverse gradient ;
  • the first transmission resource is used for the first communication device to send the first data to the second communication device, and the first data is the first output of the first machine learning model, that is, the forward reasoning result of the first machine learning model;
  • the second transmission resource is used by the first communication device to receive the first feedback data from the second communication device, the first feedback data is used to indicate the first gradient, the first gradient is the aforementioned reverse gradient, and the first gradient is used to update the first gradient.
  • a machine learning model the first communication device sends the first data to the second communication device on the first transmission resource.
  • the control signaling is used to indicate transmission resources, and the existing control signaling can only indicate transmission resources in one direction.
  • the sending end device (first communication device) or the receiving end device (second communication device) in order to realize the training of the machine learning model system, it is necessary to obtain multiple control signaling to determine the forward reasoning results for transmission and the transfer resources in the two directions of the reverse gradient.
  • the transmission resource information in two directions can be determined at one time. Sending the first information as control signaling can reduce the amount of control signaling in the communication system, thereby reducing control signaling overhead in the communication system.
  • the first data is transmitted through the channel to obtain the second data
  • the second data is input into the second machine learning model to obtain the second output.
  • the first gradient is the second communication device according to the first gradient
  • the second output is calculated.
  • the first transmission resource is used to transmit the first output of the first machine learning model, that is, the forward reasoning result; the second transmission resource is used to transmit the first gradient, that is, the reverse gradient.
  • the calculation of the first gradient is derived from the first output, that is, the transmission resource indicated in the first information, and the forward inference result and the backward gradient of the transmission resource used for transmission are interrelated , through the interrelated forward inference results and reverse gradients, one-time training of the machine learning model system can be realized. That is to say, the first information in the embodiment of the present application may be used to indicate all transmission resources required for one training of the machine learning model system.
  • the first information includes indication information of a transmission resource unit, and the first transmission resource and the second transmission resource are included in the transmission resource unit.
  • the transmission resource unit includes: a first transmission resource for transmitting a forward inference result, and a second transmission resource for transmitting a reverse gradient.
  • the first transmission resource and the second transmission resource contained in the transmission resource unit are used to realize the training of the machine learning model system.
  • the first information is training control information TCI, where TCI is indication information of a transmission resource unit.
  • TCI is the indication information of the transmission resource unit.
  • the transmission resource of the forward inference result and the reverse gradient can be notified to the sending end device and/or the second communication device, so that Realize the training of the machine learning model system.
  • the first information may indicate that multiple transmission resources are used to transmit multiple forward inference results, and may also indicate that multiple transmission resources are used to transmit multiple backward gradients.
  • the first information may also include indication information of the third transmission resource and the fourth transmission resource;
  • the second communication device sends the third data, and the third data is the third output of the first machine learning model;
  • the fourth transmission resource is used for the first communication device to receive the second feedback data from the second communication device, and the second feedback data is used for A second gradient is indicated, and the second gradient is used to update the first machine learning model;
  • the first communication device sends third data to the second communication device on the third transmission resource.
  • the first information may indicate multiple transmission resources (the first transmission resource and the third transmission resource) used to transmit the forward inference result, and may also indicate multiple transmission resources used to transmit the reverse gradient resources (the second transmission resource and the fourth transmission resource).
  • Sending the first information as a control signaling can indicate multiple forward and reverse transmission resources, and multiple trainings of the machine learning network can be realized.
  • the method of the embodiment of the present application reduces the number of control signaling that needs to be transmitted in the communication system, thereby reducing It reduces the control signaling overhead in the communication system.
  • the third data is transmitted through the channel to obtain the fourth data
  • the fourth data is input into the second machine learning model to obtain the fourth output
  • the second gradient is the second gradient of the second communication device according to the first Four outputs are calculated.
  • the fourth transmission resource is used to transmit the second gradient
  • the calculation source of the second gradient is the third output
  • the third transmission resource is used to transmit the third output. Therefore, the third transmission resource and the fourth transmission resource can be used to transmit interrelated forward reasoning results and reverse gradients, which can realize one training of the machine learning model.
  • the first transmission resource and the second transmission resource can also implement a training of the machine learning model
  • the forward inference result (first output) and reverse gradient (the first output) transmitted by the first transmission resource and the second transmission resource gradient) is different from the forward reasoning result (third output) and reverse gradient (second gradient) transmitted by the third transmission resource and the fourth transmission resource, that is to say, the first transmission resource and the fourth transmission resource
  • the transmission resource indicated by the second transmission resource is used to implement one training of the machine learning model system
  • the transmission resources indicated by the third transmission resource and the fourth transmission resource are used to implement another training of the machine learning model system. Therefore, the embodiment of the present application can use one piece of first information to indicate all the transmission resources required for multiple trainings of the machine learning model system, thereby greatly reducing the overhead of control signaling.
  • the first transmission resource is the forward transmission resource of the first round
  • the second transmission resource is the reverse transmission resource of the first round
  • the third transmission resource is the forward transmission resource of the second round resource
  • the fourth transmission resource is the reverse transmission resource of the second round.
  • the first transmission resource and the second transmission resource are located in fixed transmission resource positions of at least one of time domain, frequency domain, space domain, code domain, and power domain.
  • the transmission resources of the forward reasoning result and the reverse gradient are configured on fixed transmission resources. If the fixed transmission resource is represented by an identifier, the two directions can be represented by a small amount of data. The transmission resource reduces the data amount of the first information, thereby reducing control signaling overhead.
  • the first transmission resource and the second transmission resource are located on resource positions of at least one of time domain, frequency domain, air domain, code domain, and power domain used to transmit service data.
  • the forward reasoning results and the backward gradient transmission resources are configured in the resource blocks used to transmit business data, and the machine learning model system can be trained during the process of transmitting business data to improve the training efficiency. flexibility.
  • the transmission resources are flexibly set in the data blocks used to transmit business data. Since the data required for training can be transmitted without waiting for the completion of business data transmission, the time interval between transmission resources required for training can be reduced. In this way, the delay of waiting for the transmission resources required for training is reduced, and the training efficiency of the machine learning model system is improved.
  • the acquisition of the first information by the first communication device includes: the first communication device receives the first information from the second communication device; or, the first communication device receives the first information from the third communication device A message, the third communication device is used to control the communication between the first communication device and the second communication device.
  • the method further includes: the first communication device transmits the first information to the second communication device.
  • the method further includes: inputting training data into the first machine learning model to obtain a first output, that is, a forward reasoning result of the first machine learning model; On the resource, send the aforementioned first output to the second communication device; on the second transmission resource, the first communication device receives the first feedback data from the second communication device, the first feedback data is used to indicate the first gradient, the first The gradient is the aforementioned reverse gradient; the first communication device updates the first machine learning model according to the first gradient.
  • the method before the first communication device obtains the first information, the method further includes: the first communication device receives a training instruction from the second communication device, and the training instruction is used to indicate that the first machine learning Model training.
  • the first communication device can start the training process of the first machine learning model according to the training instruction after receiving the training instruction from the second communication device, for example, it can start to calculate the first output; Information, the first output can be sent to the second communication device on the first transmission resource immediately, compared with starting to calculate the first output after receiving the first information, the training efficiency of the first machine learning model is improved.
  • the second aspect of the embodiment of the present application provides a data transmission method, the method is applied to a second communication device in a communication system, and the communication system further includes a first communication device; wherein, the first communication device is deployed with a first machine learning A model, the second communication device is deployed with a second machine learning model, the first machine learning model and the second machine learning model are used to implement communication between the first communication device and the second communication device, the method includes:
  • the second communication device acquires the first information, and the first information includes indication information of the first transmission resource and the second transmission resource; wherein, the first transmission resource is used to transmit the forward reasoning result, and the second transmission resource is used to transmit the reverse gradient ;
  • the first transmission resource is used by the second communication device to receive the second data from the first communication device, the second data is the data obtained after the first data is transmitted through the channel, and the first data is the first machine learning model The first output;
  • the second transmission resource is used by the second communication device to send the first feedback data to the first communication device, the first feedback data is used to indicate the first gradient, and the first gradient is used to update the first machine learning model;
  • the second The communication device sends the first feedback data to the first communication device on the second transmission resource.
  • the second data is used to input the second machine learning model, and a third gradient is calculated according to the output, and the third gradient is used to update the second machine learning model.
  • the second output can be obtained by inputting the second data into the second machine learning model, and the first gradient is calculated by the second communication device according to the second output.
  • the first information includes indication information of a transmission resource unit, and the first transmission resource and the second transmission resource are included in the transmission resource unit.
  • the first information is training control information TCI, where TCI is indication information of a transmission resource unit.
  • the first information may indicate that multiple transmission resources are used to transmit multiple forward inference results, and may also indicate that multiple transmission resources are used to transmit multiple backward gradients.
  • the first information may also include indication information of the third transmission resource and the fourth transmission resource; wherein, the third transmission resource is used for the second communication device to receive information from The fourth data of the first communication device, the fourth data is the data obtained after the third data is transmitted through the channel, and the third data is the third output of the first machine learning model; the fourth transmission resource is used by the second communication device to send data to the first communication device.
  • a communication device sends second feedback data, the second feedback data is used to indicate the second gradient, and the second gradient is used to update the first machine learning model; the second communication device sends the first communication device on the fourth transmission resource to the first communication device Two feedback data.
  • the first transmission resource is the forward transmission resource of the first round
  • the second transmission resource is the reverse transmission resource of the first round
  • the third transmission resource is the forward transmission resource of the second round resource
  • the fourth transmission resource is the reverse transmission resource of the second round.
  • the fourth data is used to input the second machine learning model, and a fourth gradient is calculated according to the output, and the fourth gradient is used to update the second machine learning model.
  • the fourth output can be obtained by inputting the fourth data into the second machine learning model, the second gradient is calculated by the second communication device according to the fourth output, and the fourth gradient is also the second The communication device calculates it according to the fourth output.
  • the first transmission resource and the second transmission resource are located in fixed transmission resource positions of at least one of time domain, frequency domain, space domain, code domain, and power domain.
  • the first transmission resource and the second transmission resource are located on resource positions of at least one of time domain, frequency domain, air domain, code domain, and power domain used to transmit service data.
  • the acquiring the first information by the second communication device includes: the second communication device receives the first information from the first communication device; or, the second communication device receives the first information from the third communication device A message, the third communication device is used to control the communication between the first communication device and the second communication device.
  • the method further includes: the second communication device sends the first information to the first communication device.
  • the method further includes: the second communication device receives the second data from the first communication device on the first transmission resource; the second communication device inputs the second data into the second machine learning model to obtain the second output, and the second communication device calculates the first gradient according to the second output; the second communication device sends the first feedback data to the first communication device on the second transmission resource, and the first feedback data is used to bear The first gradient is used by the first communication device to update the first machine learning model.
  • the method before the second communication device acquires the first information, the method further includes: transmitting a training indication to the first communication device, where the training indication is used to indicate training of the first machine learning model.
  • the third aspect of the embodiment of the present application provides a data transmission method.
  • the method is applied to a third communication device in a communication system.
  • the communication system further includes a first communication device and a second communication device.
  • the third communication device is used to control the third communication device.
  • the first communication device is deployed with a first machine learning model
  • the second communication device is deployed with a second machine learning model
  • the first machine learning model and the second machine learning model are used to realize the communication between the first communication device and the second communication device.
  • the third communication device obtains the first information, and the first information simultaneously carries indication information of the first transmission resource and the second transmission resource; wherein, the first transmission resource is used for the first communication device to send the first data to the second communication device, The first data is the first output of the first machine learning model; the second transmission resource is used for the second communication device to send the first feedback data to the first communication device, the first feedback data is used to indicate the first gradient, and the first gradient is used for updating the first machine learning model;
  • the third communication device sends the first information to the first communication device and/or the second communication device.
  • the third communication device as the central control device for the communication between the first communication device and the second communication device, can transmit the first information to the first communication device and/or the second communication device, so that the previous Informing the first communication device and/or the second communication device of the transmission resources of the inference result and the reverse gradient, so as to realize the training of the machine learning model system. Since the central control device is used to control the communication between the first communication device and the second communication device, there will be other control signaling interactions between the central control device and the first communication device and/or the second communication device. Content carried by other control signaling can be carried in the first message, so that the amount of control signaling in the communication system can be reduced, thereby reducing the overhead of control signaling in the communication system.
  • the third communication device as the central control device between the first communication device and the second communication device, can know the usage of all transmission resources between the first communication device and the second communication device, and can make a decision based on the usage. More reasonable allocation of transmission resources (for example, the interval between transmission resources can be reduced as much as possible to reduce the delay of data transmission in training, or the transmission resources can be flexibly set in the middle of allocated resource blocks, etc.). Regarding the beneficial effect of the first information, refer to the first aspect of the embodiment of the present application, and details are not repeated here.
  • the first data is transmitted through the channel to obtain the second data
  • the second data is input into the second machine learning model to obtain the second output.
  • the first gradient is the second communication device according to the first gradient
  • the second output is calculated.
  • the first information includes indication information of a transmission resource unit, and the first transmission resource and the second transmission resource are included in the transmission resource unit.
  • the first information is training control information TCI, where TCI is indication information of a transmission resource unit.
  • the first information may indicate that multiple transmission resources are used to transmit multiple forward inference results, and may also indicate that multiple transmission resources are used to transmit multiple backward gradients.
  • the first information may also include indication information of the third transmission resource and the fourth transmission resource;
  • the second communication device sends the third data, the third data is the third output of the first machine learning model;
  • the fourth transmission resource is used for the second communication device to send the second feedback data to the first communication device, and the second feedback data is used for indicating The second gradient, where the second gradient is used to update the first machine learning model;
  • the first communication device sends third data to the second communication device on the third transmission resource.
  • the third data is transmitted through the channel to obtain the fourth data
  • the fourth data is input into the second machine learning model to obtain the fourth output
  • the second gradient is the second gradient of the second communication device according to the first Four outputs are calculated.
  • the first transmission resource and the second transmission resource are located in fixed transmission resource positions of at least one of time domain, frequency domain, space domain, code domain, and power domain.
  • the first transmission resource and the second transmission resource are located on resource positions of at least one of time domain, frequency domain, air domain, code domain, and power domain used to transmit service data.
  • the method before the third communication device obtains the first information, the method further includes: the third communication device receives a training instruction from the second communication device, and the training instruction is used to indicate that the first machine learning model and the training of the second machine learning model.
  • the fourth aspect of the embodiment of the present application provides a communication device, which is characterized in that the communication device is a first communication device in a communication system, and the communication system further includes a second communication device, and the first communication device is deployed with a first machine learning model , the second communication device is deployed with a second machine learning model, the first machine learning model and the second machine learning model are used to realize the communication between the first communication device and the second communication device, the first communication device includes: a processing unit, transceiver unit;
  • the processing unit is configured to: acquire first information, where the first information includes indication information of a first transmission resource and a second transmission resource; wherein, the first transmission resource is used to send first data to the second communication device, and the first data is the first transmission resource.
  • a first output of a machine learning model; the second transmission resource is used to receive first feedback data from the second communication device, the first feedback data is used to indicate a first gradient, and the first gradient is used to update the first machine learning model;
  • the transceiver unit is configured to: send the first data to the second communication device on the first transmission resource.
  • the first communication device is used to execute the method of the aforementioned first aspect.
  • the fifth aspect of the embodiment of the present application provides a communication device, which is characterized in that the communication device is a second communication device in a communication system, and the communication system further includes a first communication device, and the first communication device is deployed with a first machine learning model , the second communication device is deployed with a second machine learning model, the first machine learning model and the second machine learning model are used to realize communication between the first communication device and the second communication device, the second communication device includes: a processing unit, transceiver unit;
  • the processing unit is configured to: acquire first information, where the first information includes indication information of a first transmission resource and a second transmission resource; wherein, the first transmission resource is used to receive second data from the first communication device, and the second data is The data obtained after the first data is transmitted through the channel, the first data is the first output of the first machine learning model; the second transmission resource is used to send the first feedback data to the second communication device, and the first feedback data is used to indicate the first feedback data a gradient, the first gradient is used to update the first machine learning model;
  • the transceiver unit is configured to: send the first feedback data to the first communication device on the second transmission resource.
  • the second communication device is used to implement the method of the aforementioned second aspect.
  • the sixth aspect of the embodiment of the present application provides a communication device, which is characterized in that the communication device is a third communication device in a communication system, and the communication system further includes a first communication device and a second communication device, and the third communication device is used to controlling communications between a first communication device and a second communication device, the first communication device deploying a first machine learning model, the second communication device deploying a second machine learning model, the first machine learning model and the second machine learning
  • the model is used to realize the communication between the first communication device and the second communication device
  • the third communication device includes: a processing unit and a transceiver unit;
  • the processing unit is configured to: obtain first information, where the first information includes indication information of a first transmission resource and a second transmission resource; wherein, the first transmission resource is used for the first communication device to send the first data to the second communication device, and the first transmission resource is used for sending the first data to the second communication device.
  • One data is the first output of the first machine learning model; the second transmission resource is used for the second communication device to send the first feedback data to the first communication device, the first feedback data is used to indicate the first gradient, and the first gradient is used for updating the first machine learning model;
  • the transceiver unit is used for: sending the first information to the first communication device and/or the second communication device.
  • the third communication device is used to implement the method of the aforementioned third aspect.
  • the seventh aspect of the embodiment of the present application provides a communication device, which is characterized in that the communication device is a first communication device in a communication system, and the communication system further includes a second communication device, and the first communication device is deployed with the first machine learning model, the second communication device is deployed with a second machine learning model, the first machine learning model and the second machine learning model are used to realize communication between the first communication device and the second communication device, and the first communication device includes: a processor ,transceiver;
  • the transceiver is used to send and receive data or information
  • the processor is configured to execute the data transmission method of the first aspect.
  • the eighth aspect of the embodiment of the present application provides a communication device, which is characterized in that the communication device is a second communication device in a communication system, and the communication system further includes a first communication device, and the first communication device is deployed with a first machine learning model , the second communication device is deployed with a second machine learning model, the first machine learning model and the second machine learning model are used to realize the communication between the first communication device and the second communication device, the second communication device includes: a processor, transceiver;
  • the transceiver is used to send and receive data or information
  • the processor is configured to execute the data transmission method of the second aspect.
  • a ninth aspect of the embodiment of the present application provides a communication device, wherein the communication device is a third communication device in a communication system, and the communication system further includes a first communication device and a second communication device, and the third communication device is used to controlling communications between a first communication device and a second communication device, the first communication device deploying a first machine learning model, the second communication device deploying a second machine learning model, the first machine learning model and the second machine learning
  • the model is used to implement communication between the first communication device and the second communication device
  • the third communication device includes: a processor and a transceiver;
  • the transceiver is used to send and receive data or information
  • the processor is configured to execute the data transmission method of the second aspect.
  • the tenth aspect of the embodiment of the present application provides a communication device, including:
  • the processor is used to execute the program instructions stored in the memory, so as to implement the method in any one of the aforementioned first to third aspects;
  • the communication interface is used to communicate with other devices.
  • the eleventh aspect of the embodiment of the present application provides a communication device, including: a processor, where the processor is coupled to a memory;
  • a processor configured to execute the program in the memory, so that the processor executes the method of any one of the aforementioned first aspect to the third aspect.
  • the communications device further includes the foregoing memory.
  • the memory and the processor are integrated together.
  • the memory is located outside the communication device.
  • the twelfth aspect of the embodiment of the present application provides a computer-readable storage medium, the computer-readable storage medium stores a program, and when the computer executes the program, any one of the aforementioned first to third aspects is executed Methods.
  • the thirteenth aspect of the embodiments of the present application provides a computer program product.
  • the computer program product When the computer program product is executed on a computer, the computer executes the method of any one of the foregoing first to third aspects.
  • a fourteenth aspect of the embodiment of the present application provides a communication system, the communication system includes the first communication device of the fourth aspect and the second communication device of the fifth aspect.
  • the communication system further includes the third communication device of the sixth aspect.
  • the communication system is used to implement the data transmission method in any one of the aforementioned first aspect to the third aspect.
  • Figure 1a is a schematic diagram of the application scenario of the data transmission method provided by the embodiment of the present application.
  • Figure 1b is a schematic diagram of the system architecture of the data transmission method provided by the embodiment of the present application.
  • Fig. 1c is a schematic diagram of the neural network system provided by the embodiment of the present application.
  • FIG. 2 is an interactive schematic diagram of the data transmission method provided by the embodiment of the present application.
  • FIG. 3a is a schematic diagram of a single-resource transmission resource unit TU provided by an embodiment of the present application.
  • FIG. 3b is a schematic diagram of a multi-resource TU provided by an embodiment of the present application.
  • Figure 4a is a schematic diagram of a fixed TU type provided by the embodiment of the present application.
  • Figure 4b is a schematic diagram of the dynamic TU type provided by the embodiment of the present application.
  • FIG. 5 is another interactive schematic diagram of the data transmission method provided by the embodiment of the present application.
  • FIG. 6 is another interactive schematic diagram of the data transmission method provided by the embodiment of the present application.
  • Fig. 7a is a schematic diagram of TU in the simulated gradient return method provided by the embodiment of the present application.
  • Fig. 7b is another schematic diagram of TU in the simulated gradient return method provided by the embodiment of the present application.
  • FIG. 8 is another interactive schematic diagram of the data transmission method provided by the embodiment of the present application.
  • FIG. 9 is a schematic structural diagram of a first communication device provided by an embodiment of the present application.
  • FIG. 10 is a schematic structural diagram of a second communication device provided by an embodiment of the present application.
  • Fig. 11 is a schematic structural diagram of the central control device provided by the embodiment of the present application.
  • FIG. 12 is a schematic structural diagram of a communication device provided by an embodiment of the present application.
  • Fig. 13 is a schematic structural diagram of the device provided by the embodiment of the present application.
  • FIG. 1a is a schematic diagram of an application scenario of the data transmission method provided by the embodiment of the present application.
  • the communication system 100 shown in FIG. 1 a includes a communication device 110 at a sending end and a communication device 120 at a receiving end.
  • the first machine learning model 210 is deployed on the communication device 110 at the sending end
  • the second machine learning model 220 is deployed on the communication device 120 at the receiving end.
  • the first machine learning model 210 and the second machine learning model 220 are used to realize the 110 and the communication device 120 at the receiving end.
  • the communication device 110 at the sending end may be a terminal device.
  • the communication device 110 at the sending end may also be other devices capable of communication, such as a base station, which is not limited here.
  • the communication device 120 at the receiving end may be a base station.
  • the communication device 120 at the receiving end may also be other devices with communication capabilities, such as terminal equipment, etc., which is not limited here.
  • the communication system 100 in the embodiment of the present application may further include a central control device, and the central control device is used to control communication between the communication device 110 at the sending end and the communication device 120 at the receiving end.
  • the central control device may be a base station.
  • the central control device in addition to the base station, can also be other communication devices, such as edge devices, as long as it has the ability to control the communication between the communication device at the sending end and the communication device at the receiving end. Do limited.
  • the communication system in the embodiment of the present application may be a wireless communication system such as fifth generation mobile communication technology (5th generation mobile communication technology, 5G), satellite communication, and short distance, and the system architecture is shown in FIG. 1b.
  • the system architecture includes a network device 130, and the network device 130 provides communication services to the terminal device 140A and the terminal device 140B.
  • the wireless communication system can also perform point-to-point communication, such as communication between multiple terminal devices.
  • the network device 130 may also provide services to more or fewer terminal devices, and the number and types of terminal devices are determined according to actual needs, which are not specifically limited here.
  • the wireless communication systems mentioned in the embodiments of this application include but are not limited to: narrow band-internet of things (NB-IoT), long term evolution (long term evolution, LTE) and 5G
  • NB-IoT narrow band-internet of things
  • LTE long term evolution
  • 5G 5th Generationан ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ ⁇
  • the network device 130 is a device deployed in a radio access network to provide a wireless communication function for a terminal device.
  • the network device 130 may provide a wireless communication function for multiple terminal devices, such as the terminal device 140A and the terminal device 140B shown in FIG. 1 b .
  • the network device 130 may include various forms of macro base stations, micro base stations (also called small stations), relay stations, access points, and the like.
  • the name of the device having the function of network device 130 may be different, for example, in an LTE system, it is called an evolved node B (evolved nodeB, eNB or eNodeB), in In the third generation (3rd generation, 3G) system, it is called Node B (Node B) and so on.
  • eNB evolved node B
  • Node B Node B
  • the above-mentioned devices that provide wireless communication functions for terminal devices are collectively referred to as network devices or base stations (BS).
  • the network device 130 in this application can be an evolved base station (evolutional Node B, eNB or eNodeB) in LTE; or a base station in a 5G network, a broadband network gateway (broadband network gateway, BNG), an aggregation switch or a non-third Generation partnership project (3rd generation partnership project, 3GPP) access equipment, etc., which are not specifically limited in this embodiment of the present application.
  • eNB evolved Node B
  • eNodeB base station in LTE
  • BNG broadband network gateway
  • 3GPP non-third Generation partnership project
  • the base stations in this embodiment of the present application may include various forms of base stations, for example: macro base stations, micro base stations (also called small stations), relay stations, access points, next-generation base stations (gNodeB, gNB), transmission Point (transmitting and receiving point, TRP), transmitting point (transmitting point, TP), mobile switching center and device-to-device (Device-to-Device, D2D), vehicle-to-everything (V2X), machine
  • M2M machine-to-machine
  • Internet of Things Internet of Things
  • the terminal devices involved in the embodiments of the present application may include various handheld devices with wireless communication functions, vehicle-mounted devices, wearable devices, computing devices or other processing devices connected to wireless modems.
  • the terminal device can also be called a terminal, and the terminal device can also be a subscriber unit, a cellular phone, a smart phone, a wireless data card, a personal digital assistant (PDA) ) computer, tablet computer, wireless modem (modem), handheld device (handset), laptop computer (laptop computer), machine type communication (machine type communication, MTC) terminal, etc., are not limited here.
  • PDA personal digital assistant
  • terminal devices there may be more or less terminal devices or in the communication system, and the number and types of terminal devices are determined according to actual needs, which are not specifically limited here.
  • the terminal equipment mentioned in the embodiment of the present application may be a device with a wireless transceiver function, and specifically may refer to user equipment (user equipment, UE), access terminal, subscriber unit (subscriber unit), subscriber station, mobile station (mobile station), remote station, remote terminal, mobile device, user terminal, wireless communication device, user agent, or user device.
  • user equipment user equipment, UE
  • access terminal subscriber unit (subscriber unit), subscriber station, mobile station (mobile station), remote station, remote terminal, mobile device, user terminal, wireless communication device, user agent, or user device.
  • the terminal device may also be a satellite phone, a cellular phone, a smartphone, a wireless data card, a wireless modem, a machine type communication device, may be a cordless phone, a session initiation protocol (SIP) phone, a wireless local loop (wireless local loop) loop, WLL) station, personal digital assistant (personal digital assistant, PDA), handheld device with wireless communication function, computing device or other processing device connected to a wireless modem, vehicle-mounted device, communication device carried on high-altitude aircraft, wearable Equipment, unmanned aerial vehicles, robots, terminals in device-to-device (D2D) communication, terminals in vehicle to everything (V2X), virtual reality (virtual reality, VR) terminal equipment, Augmented reality (augmented reality, AR) terminal equipment, mixed reality (mixed reality, MR), wireless terminals in industrial control (industrial control), wireless terminals in self driving (self driving), remote medical (remote medical) Wireless terminals in smart grid, wireless terminals in transportation safety, wireless terminals in smart city, wireless terminals in smart
  • the communication device 120 at the receiving end may be a terminal device or a network device; if the communication device 110 at the sending end is a terminal device, the communication device 120 at the receiving end may be The network device may also be a terminal device, which is not limited here.
  • the communication system architecture of the embodiment of the present application is described above, and the data processing and transmission process in the machine learning model and how to train the machine learning model in the embodiment of the present application are described next.
  • the machine learning model system 200 in FIG. 1a may be a neural network system.
  • the neural network system will be taken as an example to illustrate the data processing and transmission process in the machine learning model system 200 .
  • FIG. 1c is a schematic diagram of a neural network system provided by an embodiment of the present application.
  • the first neural network 210A is deployed on the communication device 110 at the sending end
  • the second neural network 220A is deployed on the communication device 120 at the receiving end.
  • s in Fig. 1c is the input value of the neural network system, input s into the first neural network 210A, the first neural network 210A calculates the corresponding output value X, and sends X to the second neural network 220A through the communication channel.
  • the communication device 110 at the sending end inputs the training data into the first machine learning model 210 to obtain a forward reasoning result.
  • the transmitting end communication device 110 sends the forward reasoning result to the receiving end communication device 120, and the receiving end communication device inputs the forward reasoning result into the second machine learning model 220 to obtain an output value, and then calculate the first gradient based on the output value and the second gradient.
  • the first gradient is used to update the parameter ⁇ of the first neural network 210A
  • the second gradient is used to update the parameter ⁇ of the second neural network 220A.
  • the communication device 120 at the receiving end sends the first gradient to the communication device 110 at the sending end, and the communication device 110 at the sending end updates the parameter ⁇ of the first machine learning model 210 according to the first gradient.
  • the process of using the first gradient and the second gradient to update the first machine learning model and the second machine learning model is called the reverse transfer of the gradient, and the first gradient used to update the first machine learning model is called the reverse gradient below .
  • the communication device 110 at the sending end needs to send the forward reasoning result to the communication device 120 at the receiving end, and the communication device 120 at the receiving end needs to feed back the results based on the forward reasoning to the communication device 110 at the sending end.
  • the resulting reverse gradient is to say, during the training process of the machine learning model, the communication device 110 at the sending end needs to send the forward reasoning result to the communication device 120 at the receiving end, and the communication device 120 at the receiving end needs to feed back the results based on the forward reasoning to the communication device 110 at the sending end.
  • the resulting reverse gradient is to say, during the training process of the machine learning model, the communication device 110 at the sending end needs to send the forward reasoning result to the communication device 120 at the receiving end, and the communication device 120 at the receiving end needs to feed back the results based on the forward reasoning to the communication device 110 at the sending end.
  • the resulting reverse gradient is to say, during the training process of the machine learning model, the communication device 110 at the sending end needs to send the forward reasoning result to the communication device 120 at the receiving
  • the first communication device is used to represent the sending end communication device 110
  • the second communication device is used to represent the receiving end communication device 120 .
  • the data transmitted from the first communication device to the second communication device is the result of forward reasoning, so the direction from the first communication device to the second communication device is called forward;
  • the data transmitted by the first communication device is a reverse gradient, so the direction from the second communication device to the first communication device is called reverse.
  • the data transmission resources between the first communication device and the second communication device are indicated through control signaling.
  • both the first communication device and the second communication device need to know the forward and reverse Only in this way can the first communication device and the second communication device transmit data smoothly on the resource.
  • control signaling can be generated and transmitted by any one of the first communication device, the second communication device, or the central control device, so that both the first communication device and the second communication device can know the content of the control signaling .
  • the control signaling is generated by the first communication device, the first communication device will transmit the control signaling to the second communication device; other cases can be deduced by analogy, which will not be repeated one by one.
  • control signaling only includes data transmission resources in one direction of forward and reverse. Since the training of the machine learning model system requires data transmission in two directions, the forward and reverse transmission Each control signaling needs to be sent once, and the training of the machine learning model system needs multiple data transmissions to complete, so the number of control signaling is large, and the control signaling overhead in the communication system is relatively large.
  • the embodiment of the present application provides a data transmission method and related equipment, which are used to realize the training of the machine learning model deployed at the sending and receiving ends in the end-to-end communication system with low signaling overhead, and complete the sending and receiving of the communication system. Data transmission at both ends.
  • the data transmission method provided by the embodiment of the present application will be described.
  • Figure 2 is an interactive schematic diagram of the data transmission method provided by the embodiment of the present application, the method includes:
  • the first communication device acquires first information.
  • the first communication device needs to determine forward transmission resources for transmitting forward inference results and backward transmission resources for transmitting backward gradients. Specifically, the first communication device may determine the forward transmission resource and the reverse transmission resource according to the first information.
  • the first communication device acquires first information, where the first information includes indication information of the first transmission resource and the second transmission resource.
  • the first transmission resource is the aforementioned forward transmission resource
  • the second transmission resource is the aforementioned reverse transmission resource.
  • the first communication device may acquire the first information by receiving training control information (training control information, TCI). Similar to downlink control information (DCI) and uplink control information (UCI), TCI is also used to indicate transmission resources. The difference is that DCI and UCI can only indicate transmission resources in one direction, while TCI can indicate transmission resources in both forward and reverse directions.
  • training control information training control information
  • DCI downlink control information
  • UCI uplink control information
  • a training unit (training unit, TU) is defined.
  • TU includes forward transmission resources and reverse transmission resources, which are used to transmit forward reasoning results and reverse gradients required for machine learning model system training.
  • TUs can be classified in various ways, such as resource types based on bearing, the size ratio of forward and reverse transmission resources in TUs, resource locations indicated by TUs, etc. Next, we will introduce different types of TUs. .
  • the TU can be divided into a single-resource TU and a multi-resource TU.
  • a single-resource TU contains transmission resources for a single-round training of the machine learning model system.
  • FIG. 3a is a schematic diagram of a single-resource TU provided by the embodiment of the present application.
  • a single-resource TU contains forward and reverse transmission resources required for one round of training.
  • F in the figure represents forward transmission resources, and G represents reverse transmission resources.
  • the first transmission resource and the second transmission resource are taken as an example for description.
  • the first transmission resource included in the TU is the forward transmission resource in one round of training
  • the second transmission resource included in the TU is the reverse transmission resource in the round of training.
  • the first transmission resource and the second transmission resource may not be transmission resources of the same training round, which is not limited here.
  • one TCI can be used to indicate the transmission resources required for one round of machine learning model training. If the TCI is A single control signaling can use one control signaling to indicate the transmission resources required for one round of machine learning model training, which reduces the number of control signaling in the communication system.
  • Multi-resource TU Contains the transmission resources required for multi-round training. As shown in Figure 3b, one multi-resource TU contains forward and reverse transmission resources required for multi-round end-to-end training. In consideration of reducing forward and reverse transmission interference, forward and reverse transmission resources do not appear in the same time period.
  • the TU may include first to fourth transmission resources, the first transmission resource is the forward transmission resource of the first round, the second transmission resource is the reverse transmission resource of the first round; the third transmission resource is the second round forward transmission resource, and the fourth transmission resource is the reverse transmission resource of the second round.
  • the aforementioned first round and second round do not limit the sequence of the two rounds, but are only used to distinguish different rounds.
  • the aforementioned two rounds are only examples of the multi-round training included in the TU, and do not limit the number of training rounds included in the TU.
  • the TU may include transmission resources required for more rounds of training. There is no limit here,
  • TU types Based on the size ratio of forward and reverse transmission resources in TU, different TU types can be determined.
  • the amount of forward and backward transfer data may vary, and the size ratio of forward and reverse transfer resources may vary. Therefore, in the embodiment of the present application, multiple TU types (types) are defined, and the proportions of forward and reverse transmission resources in different types of TUs are different. For example, as shown in Figure A in Figure 3a, the ratio of forward and reverse transmission resources in TU type1 is 1:1; as shown in Figure B in Figure 3a, the ratio of forward and reverse transmission resources in type2 is 2 :1.
  • multiple TU types can be defined, corresponding to different ratios of forward and reverse transmission resources.
  • TU type1 can be used in multi-round training, that is, 1:1 forward and reverse transmission resources for the communication required for training.
  • the resources required for reverse transmission during training will gradually be less than those required for forward transmission.
  • the number of training rounds is greater than 0 and less than 50
  • the amount of data transmitted forward is similar to the amount of data transmitted backward.
  • Select TU type1 when the number of training rounds is greater than or equal to 50 and less than 100, the amount of data transmitted in the reverse direction is close to half of the amount of data transmitted in the forward direction, so TU type2 is selected.
  • the TU can be divided into a fixed location TU and a dynamic location TU.
  • Fixed location TU As shown in Figure 4a, the TU is configured on a fixed resource location.
  • Dynamic position TU As shown in Figure 4b, the position of the TU is not fixed, and the TU is configured on a dynamically variable resource position.
  • a part of resources may be dynamically selected for transmission required for end-to-end machine learning model training.
  • the transmission resource of user equipment (user equipment, UE) 1 and the transmission resource of UE2 are occupied.
  • the first transmission resource and the second transmission resource may be configured at fixed time-frequency transmission resource positions.
  • the first transmission resource and the second transmission resource in addition to being configured at a fixed time-frequency position, it is also possible not to combine the two dimensions of the time domain and the frequency domain for allocation, but to separately consider configuring the transmission resource at a fixed time-domain transmission resource position , or separately consider configuring the transmission resource at a fixed frequency domain transmission resource position, which is not limited here.
  • the first transmission resource and the second transmission resource in addition to the time domain or the frequency domain, can also be configured in fixed transmission resource positions in other dimensions, such as the air domain, code domain, power domain and other dimensions, where No limit.
  • the TU is configured on a fixed transmission resource. If the fixed transmission resource is represented by an identifier, the transmission resources in two directions can be represented by a small amount of data, and the data volume of the TCI can be reduced. , thereby reducing control signaling overhead.
  • the TU may be configured at a time-domain resource location for transmitting service data.
  • the first transmission resource and the second transmission resource can also be configured in resource positions of other dimensions used to transmit service data, such as frequency domain, air domain, code domain, power domain, etc. Dimensions are not limited here.
  • the TU is configured in the resource block used to transmit business data, and the training of the machine learning model system can be performed during the process of transmitting business data, which improves the flexibility of training and reduces the need for waiting for training.
  • the delay of transmission resource allocation improves the training efficiency of the machine learning model system.
  • a training unit TU is also called a transmission resource unit.
  • the content of the data transmitted by the forward and reverse transmission resources in the TU is not limited.
  • forward reasoning results and reverse gradients it can also be used to transmit other data, which is not limited here.
  • the application scenario of TCI is not limited, and it can also be applied to other scenarios except for the training of the machine learning model system, as long as in this scenario, the interactive transmission of data between two communication devices is sufficient , is not limited here.
  • the TCI may specifically include: information about the location of the TU in the frequency domain and information about the location of the TU in the time domain.
  • the TCI may further include: a forward transmission modulation and coding scheme and a reverse transmission modulation and coding scheme.
  • the TCI may include: the frequency domain position information of the forward transmission resource in this round, the time domain position information of the forward transmission resource in this round, the frequency domain position information of the reverse transmission resource in this round, and the position information of this round.
  • the time domain location information where the resource is located is transmitted in round reverse.
  • a round refers to a training round of the machine learning model system, and one round of training includes the transmission of forward inference results and the transmission of reverse gradients.
  • the preceding fields can be repeated, that is, "the frequency domain location information of the 1st round forward transmission resource, the time domain location information of the 1st round forward transmission resource , the frequency domain position information of the resource in the reverse transmission in the first round, the time domain position information of the resource in the reverse transmission in the first round; the frequency domain position information of the forward transmission resource in the second round, the time domain of the forward transmission resource in the second round Location information, the frequency domain location information of the second round of reverse transmission resources, the time domain location information of the second round of reverse transmission resources; ..., the frequency domain location information of the k-th round of forward transmission resources, the k-th round of forward transmission Time-domain location information where the resource is located, the frequency-domain location information where the resource is located is reversely transmitted at the k-th round, and the time-domain location information where the resource is located is reversely transmitted at the k-th round.”
  • This form is used to indicate the location of resources required for multi-round training in
  • each repeated field includes the frequency domain position information of the forward and reverse transmission resources and the time domain position information of the forward and reverse transmission resources in the same round.
  • the order in which each content appears in the repeated field, and the content included in the repeated field as long as the repeated field can be used to indicate the transmission resources to be used for a single round of training, there is no limit here .
  • the TCI of the multi-resource TU may also include an identifier, which is used to indicate the training round corresponding to each transmission resource, so as to distinguish different training rounds.
  • the aforementioned way of distinguishing different training rounds is based on the positional relationship of the repeated fields corresponding to different rounds in the TCI field. No additional identification is required to indicate the training rounds corresponding to the transmission resources, which reduces the amount of TCI data and saves signaling overhead.
  • the transmission resources of different training rounds may also be distinguished through the preceding and following positions of the identifier or the repeated fields in the TCI field.
  • the first communication device in addition to receiving the TCI, may also acquire the first information in other ways. For example, the first communication device determines the transmission resource in the first information, so as to obtain the first information, etc., which is not limited here.
  • the TCI may come from the second communication device, or from other communication devices, and the transmission of the TCI will be described in detail in the following embodiments, which is not limited here.
  • the second communication device acquires first information.
  • the second communication device in order to train the machine learning network model system, the second communication device also needs to determine forward transmission resources and reverse transmission resources. Specifically, the second communication device may determine the forward transmission resource and the reverse transmission resource according to the first information.
  • the second communication device may determine the transmission resource in the first information to obtain the first information; it may also obtain the first information by receiving the TCI, which is not limited here.
  • the TCI may come from the first communication device, or from other communication devices, and the transmission of the TCI will be described in detail in the following embodiments, which is not limited here.
  • step 201 For the description of TU and TCI, refer to step 201, which will not be repeated here.
  • step 201 can be executed first and then step 202 can be executed, or step 202 can be executed first and then step 201 is executed. Do limited.
  • the first communication device sends the first data to the second communication device on the first transmission resource.
  • the first transmission resource in the TU is used for the first communication device to send the first data to the second communication device, and the first data is the first output of the first machine learning model, that is, the forward reasoning result of the first machine learning model.
  • the first communication device sends the first data to the second communication device on the first transmission resource.
  • the gradient calculated based on the first data may also be used to update the second machine learning model, which will be described in the embodiments shown in FIG. 5 to FIG. 8 and will not be repeated here.
  • the second communication device sends the first feedback data to the first communication device on the second transmission resource.
  • the second transmission resource is used for the second communication device to send the first feedback data to the first communication device, the first feedback data is used to indicate the first gradient, the first gradient is the aforementioned reverse gradient, and the first gradient is used to update the first machine learning model.
  • the second communication device sends the first feedback data to the first communication device on the second transmission resource.
  • the TU in step 201 and/or 202 is a multi-resource TU
  • the TU may also include a third transmission resource and a fourth transmission resource, wherein the third transmission resource
  • the resource is used for the first communication device to send third data to the second communication device, and the third data is the third output of the first machine learning model;
  • the fourth transmission resource is used for the second communication device to send second feedback to the first communication device data, the second feedback data is used to indicate the second gradient, and the second gradient is used to update the first machine learning model.
  • step 205 and step 206 may also be performed.
  • the first communications device sends third data to the second communications device on the third transmission resource.
  • the third transmission resource in the TU is used for the first communication device to send third data to the second communication device, and the third data is the third output of the first machine learning model, that is, the forward reasoning result of the first machine learning model.
  • the first communication device sends third data to the second communication device on the third transmission resource.
  • the gradient calculated based on the third data may also be used to update the second machine learning model.
  • the second communication device sends the second feedback data to the first communication device on the fourth transmission resource.
  • the fourth transmission resource is used for the second communication device to send second feedback data to the first communication device, the second feedback data is used to indicate the second gradient, the second gradient is the aforementioned reverse gradient, and the second gradient is used to update the first machine learning model
  • the second communication device sends the second feedback data to the first communication device on the fourth transmission resource.
  • a unified TU type can be used in multiple rounds, such as single-resource TU or multi-resource TU or type1 TU.
  • you can change the TU type during the multi-round training process for example, use a single-resource TU first and change it to a multi-resource TU from a certain round, or use a type1 TU first and change it to a type2 TU from a certain round etc., there is no limitation here.
  • steps 205 and 206 are optional steps. When the TU does not include the third transmission resource and the fourth transmission resource, steps 205 and 206 may not be performed, which is not limited here.
  • the TU can be determined by any one of the first communication device, the second communication device, and the central control device, and the TCI is also sent by the corresponding device, which will be described separately next.
  • the TU is determined by the first communication device.
  • Figure 5 is a flow chart of the data transmission method provided by the embodiment of the present application, as shown in Figure 2, the method includes:
  • the first communication device initializes a first machine learning model
  • the second communication device initializes a second machine learning model
  • a local random initialization method may be used to initialize the machine learning model.
  • machine learning models can also be initialized in other ways, such as downloading corresponding machine learning model parameters from a specified model server, which is not limited here.
  • step 501 is an optional step, and in the training process of the machine learning model system, step 501 is not executed before each round of training.
  • the first communication device transmits the test data to the second communication device through the machine learning model system, and correspondingly, the second communication device receives the test data.
  • a first machine learning model is deployed on the first communication device.
  • the first output can be obtained by inputting the test data to be transmitted to the second communication device into the first machine learning model.
  • the first output is also referred to as first data.
  • the first communication device transmits the first data to the second communication device. Since the communication channel will affect the data, the second communication device will receive the second data. The second data is obtained after the first data is transmitted through the channel. data.
  • the second machine learning model is deployed on the second communication device, and the second data is input into the second machine learning model to obtain a second output, which is the reasoning result of the aforementioned business data input into the first machine learning model.
  • the machine learning model system composed of the first machine learning model and the second machine learning model can not only transmit business data as described above, but also perform data processing, such as semantic analysis, image For classification, etc.
  • the second output obtained at the second machine learning model may correspond to semantic analysis results, image classification results, etc., which are not limited here.
  • test data transmitted in step 502 may also be service data, transmission signaling, etc., as long as the performance evaluation in step 503 can be realized, which is not limited here.
  • the second communication device performs performance evaluation, and determines that a machine learning model system needs to be trained.
  • the second communication device may perform performance evaluation, evaluating the performance of data transmission based on the current machine learning model system. Specifically, the second communication device may evaluate the bit error rate of data transmission, and when the bit error rate is higher than a certain threshold (such as 0.1 commonly used in eMBB, 0.00001 commonly used in uRLLC), it is determined that the machine learning model system needs to be trained.
  • a certain threshold such as 0.1 commonly used in eMBB, 0.00001 commonly used in uRLLC
  • the second communication device may also determine whether it is necessary to train the machine learning model system through other performances, such as evaluating performances such as throughput and delay, which are not limited here.
  • evaluating performances such as throughput and delay, which are not limited here.
  • the delay in uRLLC is greater than or equal to 1 ms, it may be determined that the machine learning model system needs to be trained.
  • step 502 and the performance evaluation can be repeated until the performance cannot meet the preset conditions, and it is determined that the machine learning model system needs to be trained
  • the model is learned (ie step 503 occurs).
  • the second communication device transmits a training instruction to the first communication device, and correspondingly, the first communication device receives the training instruction.
  • the second communication device may transmit a training instruction to the first communication device, where the training instruction is used to indicate the training of the first machine learning model.
  • the first communications apparatus determines a TU.
  • the first communication device may determine a transmission resource unit TU, where the TU includes the first transmission resource and the second transmission resource.
  • the first transmission is a forward transmission resource, which is used for the first communication device to transmit the forward reasoning result to the second communication device;
  • the second transmission resource is a reverse transmission resource, which is used for the second communication device to communicate with the first communication device.
  • the device transmits a reverse gradient.
  • the first communication device transmits the TCI to the second communication device, and correspondingly, the second communication device receives the TCI.
  • the first communication device can determine the TCI and transmit the TCI to the second communication device.
  • the TCI is the indication information of the TU, and is used to indicate the location information of all transmission resources included in the TU.
  • the TCI is also referred to as first information, which is not limited here.
  • the TCI is not necessarily sent by the first communication device.
  • the first communication device may also notify other devices (such as relay devices) of the TU, and the other devices transmit the TCI to the second communication device, which is not limited here.
  • the first communication device and the second communication device train the machine learning model system according to the transmission resource indicated in the TCI.
  • a single-resource TU is taken as an example to illustrate the learning process of the machine learning model system.
  • the TU includes a first transmission resource and a second transmission resource; wherein, the first transmission resource is used to indicate the transmission resource of the forward inference result; the second transmission resource is used to indicate the transmission resource of the reverse gradient.
  • the TCI includes location information of the first transmission resource and the second transmission resource.
  • the first communication device inputs the training data into the first machine learning model to obtain a first output.
  • the first output is also referred to as first data.
  • the first communication device transmits the first data to the second communication device on the first transmission resource. Since the communication channel will affect the data, the second communication device will receive the second data on the first transmission resource, and the second communication device will receive the second data on the first transmission resource.
  • the data is data obtained after the first data is transmitted through the channel.
  • the second communication device inputs the second data into the second machine learning model to obtain a second output, and the second output is an inference result of the aforementioned training data input into the first machine learning model.
  • the machine learning model system composed of the first machine learning model and the second machine learning model can not only transmit business data, but also perform data processing, such as semantic analysis, image classification, etc., in
  • the second output obtained by the second machine learning model may correspond to semantic analysis results, image classification results, etc., which are not limited here.
  • the second communication device calculates the first gradient based on the second output and the loss function of the machine learning model system, and the first gradient is used to update the first machine learning model.
  • the second communication device may also calculate a third gradient based on the second output and the loss function of the machine learning model system, and the third gradient is used to update the second machine learning model.
  • the second communication device may transmit the first feedback data to the first communication device on the second transmission resource, where the first feedback data is used to indicate the bearer of the first gradient.
  • the first communication device may update the first machine learning model based on the first gradient.
  • the second communication device may update the second machine learning model based on the third gradient.
  • the above describes the role of the forward and reverse transmission resources in the single-resource TU in the training process of the machine learning model system.
  • the role of the forward and reverse transmission resources in each round is the same as that of the first transmission above.
  • the role of the resource is similar to that of the second transmission resource, and will not be repeated here.
  • the second communication device performs performance evaluation, and transmits a training end indication to the first communication device, and correspondingly, the first communication device receives the training end indication.
  • the second communication device may evaluate the performance of the trained machine learning model system. Specifically, the second communication device may evaluate the bit error rate of the transmission based on the transmission of the forward inference result in the last round of training, and determine that the machine learning model system has met the requirements when the bit error rate is lower than a certain threshold. Using requirements, the training of the machine learning model system can be ended.
  • the second communication device may also determine whether to end the training of the machine learning model system through other performances, such as evaluating performances such as throughput and delay, which are not limited here.
  • the second communication device may transmit a training end indication to the first communication device, where the training end indication is used to notify the training of the first machine learning model.
  • Training the machine learning model system triggers an action of transmitting a training end indication to the first communication device in step 208 .
  • the training process includes: determining the TU, sending the TCI indicating the TU, training the machine learning model system according to the TCI, and evaluating by the second communication device.
  • Existing control signaling can only indicate transmission resources in one direction.
  • two control signalings need to be obtained to determine the two directions for transmitting the forward reasoning result and the reverse gradient transfer resources.
  • the transmission resource information in the two directions can be determined at one time by using the TCI to indicate the transmission resources in the forward direction and the reverse direction.
  • Sending TCI as control signaling doubles the amount of transmission resource information carried by the control signaling, and reduces the amount of control signaling to be transmitted in the communication system by half, thus reducing the control signaling overhead in the communication system.
  • the forward transmission resources and reverse transmission resources included in the TU may not be used for the training of the machine learning system, as long as the transmitted data appears in pairs, two directions of forward and reverse are required If the transmission resources are used for transmission, the TCI provided in the embodiment of the present application may be used to indicate transmission resources in two directions, which is not limited here.
  • FIG. 5 illustrates the solution of determining the TU by the first communication device.
  • the solution of determining the TU by the second communication device is described through the embodiment shown in FIG. 6 .
  • the transmission of forward inference results and reverse gradients can be processed as data transmission through encoding, modulation, etc., or can be transmitted directly on the air interface in an analog manner. In the embodiment shown in FIG. 6, it will be introduced how the analog transmission is realized.
  • the TU is determined by the second communication device.
  • FIG. 6 is a schematic flow chart of the data transmission method provided by the embodiment of the present application. As shown in FIG. 6, the method includes:
  • the frame structure refers to the structure of the TU, and in the process of configuring the frame structure, it can be determined whether the TU adopts a fixed position TU or a dynamic position TU.
  • the frame structure configuration can also determine other content related to the TU, such as whether it is a single-resource TU or a multi-resource TU, etc., which is not limited here.
  • step 601 may be implemented through parameter configuration when the communication system is deployed.
  • the frame structure configuration may be given by a standard, or may be determined in other ways besides the standard, such as determined by a central control device, etc., which is not limited here.
  • step 601 may also be included, which is not limited here.
  • step 601 is an optional step. During the training process of the machine learning model system, step 601 is not required to be performed before each round of training.
  • step 601 may be performed when the communication system is established or updated, which is not limited here.
  • the first communication device and the second communication device perform channel reciprocity measurement.
  • the training of the machine learning model can be realized by means of simulated transmission.
  • the analog transmission method is that the forward reasoning results and the reverse gradient transmission are not digitized, but directly transmitted in the form of analog signals. Since there is both forward data transmission and reverse data transmission between the first communication device and the second communication device, the forward and reverse directions are opposite directions, so channel reciprocity measurement is required . Ensuring that forward and reverse data transfer between two communicating devices can be achieved over a reciprocal channel.
  • the forward reasoning result represents the output of the first machine learning model.
  • forward and reverse transmission resources with reciprocity can be determined. For example, as shown in Figure 7a, it is determined that the channel has reciprocity in a long time T and a wide frequency band W. Then the system parameter numerology of the communication system (such as symbol length and subcarrier spacing configuration) can be determined, so that the time domain length of the TU is less than or equal to T, and the frequency domain width is less than or equal to W.
  • the system parameter numerology of the communication system such as symbol length and subcarrier spacing configuration
  • the TU in the method of analog transmission, can be a single-resource TU as shown in Figure A or B, or a multi-resource TU as shown in Figure C, which is not limited here.
  • step 602 is also performed before step 501 , which is not limited here.
  • step 602 is an optional step. During the training process of the machine learning model system, step 602 is not required to be performed before each round of training.
  • step 602 may be performed when the communication system is established or updated, which is not limited here.
  • the first communication device initializes the first machine learning model
  • the second communication device initializes the second machine learning model
  • step 603 is an optional step. During the training process of the machine learning model system, step 603 is not required to be performed before each round of training.
  • step 602 may be performed when the communication system is established or updated, which is not limited here.
  • steps 601 to 603 may be executed, or none may be executed, or a part thereof may be executed, for example, step 602 is executed, and steps 601 and 603 are not executed; or steps 601 and 602 are executed, but not Execute step 603 and so on, which are not limited here.
  • the first communication device transmits the test data to the second communication device through the machine learning model system, and correspondingly, the second communication device receives the test data.
  • the second communication device performs performance evaluation, and determines that a machine learning model system needs to be trained.
  • steps 603 to 605 refer to steps 501 to 503 in the embodiment shown in FIG. 5 , which will not be repeated here.
  • the second communication device transmits a training instruction to the first communication device, and correspondingly, the first communication device receives the training instruction.
  • the second communication device may transmit a training instruction to the first communication device, where the training instruction is used to indicate the training of the first machine learning model.
  • the second communications apparatus determines a TU.
  • step 605 the second communication device determines that the machine learning model system needs to be trained, and then determines the transmission resource unit TU.
  • the transmission resource unit TU For a detailed description of the TU, refer to step 505 in the embodiment shown in FIG. 5 , which will not be repeated here.
  • step 607 may be performed before or after step 606, as long as it is performed after step 605, which is not limited here.
  • the second communication device transmits the TCI to the first communication device, and correspondingly, the first communication device receives the TCI.
  • the second communication device can determine the TCI and transmit the TCI to the first communication device.
  • the TCI is indication information of a TU, and is used to indicate location information of all transmission resources included in the TU.
  • the TCI is not necessarily sent by the second communication device.
  • the second communication device may also notify other devices (such as relay devices) of the TU, and the other devices transmit the TCI to the first communication device, which is not limited here.
  • step 606 is an optional step.
  • the second communication device transmits the TCI to the first communication device, and the first communication device may learn based on the TCI that it needs to train the first machine learning model. Therefore, the second communication device may also not transmit the training instruction to the first communication device.
  • the advantage of the second communication device transmitting the training instruction (step 606) to the first communication device is that the first communication device receives
  • the training process of the first machine learning model can be started according to the training instruction, for example, the first output can be calculated; when the TCI is received, the first output can be transmitted immediately on the forward transmission resource indicated by the TCI
  • the training efficiency of the first machine learning model and the entire machine learning model system is improved.
  • the first communication device and the second communication device train the machine learning model system according to the transmission resource indicated in the TCI.
  • the second communication device performs performance evaluation, and transmits a training end indication to the first communication device, and correspondingly, the first communication device receives the training end indication.
  • Step 609 and step 610 refer to step 507 and step 508 in the embodiment shown in FIG. 5 , and details are not repeated here.
  • the second communication device may be a terminal device or a network device; if the first communication device is a terminal device, the second communication device may be a network device or It may be a terminal device, which is not limited here.
  • FIG. 5 and FIG. 6 illustrate the solution of determining the TU by the first communication device or the second communication device.
  • the solution of determining the TU by the central control device is described through the embodiment shown in FIG. 8 .
  • the TU is determined by the central control device.
  • the TU may be determined by the central control device.
  • the first communication device and the second communication device may be terminal devices, and the central control device may be network devices.
  • both the first communication device and the second communication device may be terminal devices.
  • the central control device may be a base station.
  • the central control device in addition to the base station, can also be other communication devices, such as edge devices, as long as it has the ability to control the communication between the first communication device and the second communication device. Do limited.
  • one of the first communication device and the second communication device may be a terminal device and the other may be a network device (such as a base station), and the central control device is used to control the communication between the network device and the terminal device .
  • FIG. 8 is a schematic flow chart of the data transmission method provided by the embodiment of the present application. As shown in FIG. 8, the method includes:
  • the frame structure refers to the structure of the TU, and in the process of configuring the frame structure, it can be determined whether the TU adopts a fixed position TU or a dynamic position TU.
  • the frame structure configuration can also determine other content related to the TU, such as whether it is a single-resource TU or a multi-resource TU, etc., which is not limited here.
  • step 801 may be implemented through parameter configuration when the communication system is deployed.
  • the frame structure configuration may be given by a standard, or may be determined in other ways besides the standard, such as determined by a central control device, etc., which is not limited here.
  • the first communication device and the second communication device perform channel reciprocity measurement.
  • the first communication device initializes the first machine learning model
  • the second communication device initializes the second machine learning model
  • steps 801 to 803 is an optional step, and in the training process of the machine learning model system, steps 801, 802 or 803 are not required to be performed before each round of training.
  • any one of steps 801 to 803 may be performed when the communication system is established or updated, which is not limited here.
  • steps 801 to 803 may be executed, or none may be executed, or a part thereof may be executed, for example, step 802 is executed, and steps 801 and 803 are not executed; or steps 801 and 802 are executed, but not Execute step 803 and so on, which are not limited here.
  • the first communication device transmits the service data to the second communication device through the machine learning model system.
  • the second communication device performs performance evaluation, and determines that a machine learning model system needs to be trained.
  • steps 802 to 805 refer to the description of steps 602 to 605 in the embodiment shown in FIG. 6 , and details are not repeated here. Wherein, step 802 and step 803 are optional steps.
  • the second communication device transmits the training instruction to the central control device, and correspondingly, the central control device receives the training instruction.
  • the second communication device may transmit a training instruction to the central control device, where the training instruction is used to indicate the training of the machine learning model system.
  • the second communication device transmits a training instruction to the first communication device, and correspondingly, the first communication device receives the training instruction.
  • the second communication device may transmit a training instruction to the first communication device, where the training instruction is used to indicate the training of the first machine learning model.
  • the central control device determines the TU.
  • the central control device can determine the transmission resource unit TU after receiving the training instruction.
  • the transmission resource unit TU For the detailed description of the TU, refer to step 505 in the embodiment shown in FIG. 5 , which will not be repeated here.
  • the central control device transmits the TCI to the first communication device and the second communication device, and correspondingly, the first communication device and the second communication device receive the TCI.
  • the central control device can determine the TCI, and transmit the TCI to the first communication device and the second communication device.
  • the TCI is indication information of a TU, and is used to indicate location information of all transmission resources included in the TU.
  • the TCI is not necessarily sent by the central control device.
  • the central control device may also notify other devices (such as relay devices) of the TU, and the other devices transmit the TCI to the first communication device and the second communication device, which is not limited here.
  • step 807 is an optional step.
  • the central control device transmits the TCI to the first communication device, and the first communication device may learn based on the TCI that it needs to train the first machine learning model. Therefore, the second communication device may also not transmit the training instruction to the first communication device.
  • the beneficial effect of sending the training instruction to the sending device refer to the description before step 609 in the embodiment shown in FIG. 6 , which will not be repeated here.
  • the first communication device and the second communication device train the machine learning model system according to the transmission resource indicated in the TCI.
  • step 810 refer to step 507 in the embodiment shown in FIG. 5 , which will not be repeated here.
  • the second communication device performs performance evaluation, and transmits a training end indication to the first communication device and the central control device, and correspondingly, the first communication device and the central control device receive the training end indication.
  • step 508 of the embodiment shown in FIG. 5 For the description of performance evaluation performed by the second communication device, refer to step 508 of the embodiment shown in FIG. 5 .
  • the second communication device After the second communication device determines that the training of the machine learning model system can be ended, it can transmit a training end instruction to the first communication device and the central control device, and the training end instruction is used to notify the training of the machine learning model system.
  • the present application also provides corresponding device embodiments.
  • the first communication device, the second communication device, and the third communication device may all include a hardware structure and/or a software module, and a hardware structure, a software module, or a hardware structure Add software modules to realize the above functions. Whether one of the above-mentioned functions is executed in the form of a hardware structure, a software module, or a hardware structure plus a software module depends on the specific application and design constraints of the technical solution.
  • the structure of the first communication device provided by the embodiment of the present application is described below.
  • the first communication device (the communication device at the sending end) is included in the communication system, and the communication system further includes a second communication device (the communication device at the receiving end), and the first communication device is deployed with a first machine learning model, The second communication device is deployed with a second machine learning model, and the first machine learning model and the second machine learning model are used to implement communication between the first communication device and the second communication device.
  • the first communication device 900 includes: a processing unit 901, a transceiver unit 902;
  • the processing unit 901 is configured to: acquire first information, and the first information simultaneously carries indication information of a first transmission resource and a second transmission resource; wherein, the first transmission resource is used to transmit the first data to the second communication device, and the first The data is the first output of the first machine learning model; the second transmission resource is used to receive the first feedback data from the second communication device, the first feedback data is used to indicate the first gradient, and the first gradient is used to update the first machine learning model;
  • the transceiving unit 902 is configured to: transmit the first data to the second communication device on the first transmission resource.
  • the first gradient is calculated by the second communication device according to the second output
  • the second output is the output obtained by inputting the second data into the second machine learning model
  • the second data is the first data after the Data obtained after channel transmission.
  • the first information includes indication information of a transmission resource unit, and the first transmission resource and the second transmission resource are included in the transmission resource unit.
  • the first information carries indication information of the third transmission resource and the fourth transmission resource at the same time; wherein, the third transmission resource is used to transmit the third data to the second communication device, and the third data is the third output of the first machine learning model; the fourth transmission resource is used to receive the second feedback data from the second communication device, the second feedback data is used to indicate the second gradient, and the second gradient is used to update the first machine learning model Model;
  • the transceiving unit 902 is configured to: transmit third data to the second communication device on the third transmission resource.
  • the first transmission resource and the second transmission resource are located in fixed transmission resource positions in at least one of the time domain, frequency domain, space domain, code domain, and power domain; or, used for transmission In the resource block of at least one of the time domain, frequency domain, air domain, code domain, and power domain of service data.
  • the transceiver unit 902 is configured to: receive the first information from the second communication device; or, receive the first information from the third communication device, and the third communication device is used to control the first communication communication between the device and the second communication device;
  • the processing unit 901 is specifically configured to: acquire the first information from the transceiver unit 902 .
  • the transceiver unit 902 is further configured to: obtain the first information from the processing unit 901; and transmit the first information to the second communication device.
  • the transceiving unit 902 is further configured to: receive a training instruction from the second communication device, where the training instruction is used to indicate the training of the first machine learning model.
  • the first communication device provided by the embodiment of the present application is introduced above, and the structure of the second communication device provided by the embodiment of the present application is described next with reference to FIG. 10 .
  • the second communication device (receiving end communication device) is included in the communication system, and the communication system further includes a first communication device (sending end communication device), and the first communication device is deployed with a first machine learning model, The second communication device is deployed with a second machine learning model, and the first machine learning model and the second machine learning model are used to implement communication between the first communication device and the second communication device.
  • the second communication device 1000 includes: a processing unit 1000 and a transceiver unit 1002;
  • the processing unit 1001 is configured to: obtain first information, and the first information simultaneously carries indication information of a first transmission resource and a second transmission resource; wherein, the first transmission resource is used to receive second data from the first communication device, and the first transmission resource is used to receive second data from the first communication device, and the first The second data is the data obtained after the first data is transmitted through the channel, and the first data is the first output of the first machine learning model; the second transmission resource is used to transmit the first feedback data to the first communication device, and the first feedback data is used for For indicating the first gradient, the first gradient is used to update the first machine learning model;
  • the transceiving unit 1002 is configured to: transmit the first feedback data to the first communication device on the second transmission resource.
  • the first gradient is calculated by the second communication device according to the second output
  • the second output is an output obtained by inputting the second data into the second machine learning model.
  • the first information includes indication information of a transmission resource unit, and the first transmission resource and the second transmission resource are included in the transmission resource unit.
  • the first information carries indication information of the third transmission resource and the fourth transmission resource at the same time; wherein, the third transmission resource is used to receive fourth data from the first communication device, and the fourth The data is the data obtained after the third data is transmitted through the channel, and the third data is the third output of the first machine learning model; the fourth transmission resource is used to transmit the second feedback data to the first communication device, and the second feedback data is used for The second gradient is indicated, and the second gradient is used to update the first machine learning model; the transceiving unit 1002 is configured to: transmit the second feedback data to the first communication device on the fourth transmission resource.
  • the first transmission resource and the second transmission resource are located in fixed transmission resource positions in at least one of the time domain, frequency domain, space domain, code domain, and power domain; or, used for transmission In the resource block of at least one of the time domain, frequency domain, air domain, code domain, and power domain of service data.
  • the transceiver unit 1002 is configured to: receive first information from a first communication device; or receive first information from a third communication device, and the third communication device is used to control the first communication communication between the device and the second communication device;
  • the processing unit 1001 is specifically configured to: acquire first information from the transceiver unit 1002 .
  • the transceiver unit 1002 is further configured to: obtain first information from the processing unit 1001; and transmit the first information to the first communication device.
  • the processing unit 1001 is further configured to: generate a training instruction, where the training instruction is used to indicate the training of the first machine learning model;
  • the transceiving unit 1002 is further configured to: send the training instruction to the first communication device.
  • the processing unit 1001 is further configured to: generate a training instruction, where the training instruction is used to indicate the training of the machine learning model network;
  • the transceiving unit 1002 is further configured to: send the training instruction to the third communication device.
  • the first communication device and the second communication device provided by the embodiment of the present application are introduced above, and the structure of the central control device provided by the embodiment of the present application is described next with reference to FIG. 11 .
  • the central control device is also referred to as a third communication device.
  • the third communication device is included in a communication system, and the communication system also includes a first communication device (transmitter communication device) and a second communication device (receiver communication device). communication device), the third communication device is used to control communication between the first communication device and the second communication device, the first communication device is deployed with a first machine learning model, and the second communication device is deployed with a second machine learning model, The first machine learning model and the second machine learning model are used to implement communication between the first communication device and the second communication device.
  • the third communication device 1100 includes: a processor 1101, a transceiver unit 1102;
  • the processing unit 1101 is configured to: obtain first information, and the first information simultaneously carries indication information of a first transmission resource and a second transmission resource; wherein, the first transmission resource is used for the first communication device to transmit the first transmission resource to the second communication device.
  • data the first data is the first output of the first machine learning model;
  • the second transmission resource is used for the second communication device to transmit the first feedback data to the first communication device, the first feedback data is used to indicate the first gradient, the first Gradients are used to update the first machine learning model;
  • the transceiver unit 1102 is configured to: transmit the first information to the first communication device and/or the second communication device.
  • the first information includes indication information of a transmission resource unit, and the first transmission resource and the second transmission resource are included in the transmission resource unit.
  • the first information carries indication information of the third transmission resource and the fourth transmission resource at the same time; wherein, the third transmission resource is used for the first communication device to transmit the third data to the second communication device , the third data is the third output of the first machine learning model; the fourth transmission resource is used for the second communication device to transmit the second feedback data to the first communication device, the second feedback data is used to indicate the second gradient, the second gradient Used to update the first machine learning model.
  • the first transmission resource and the second transmission resource are located in fixed transmission resource positions in at least one of the time domain, frequency domain, space domain, code domain, and power domain; or, used for transmission In the resource block of at least one of the time domain, frequency domain, air domain, code domain, and power domain of service data.
  • the transceiving unit 1102 is configured to receive a training instruction from the second communication device, where the training instruction is used to indicate the training of the machine learning model system.
  • the transceiving unit 1102 is further configured to: receive a training instruction from the second communication device, where the training instruction is used to indicate the training of the machine learning model network.
  • the communication device in this embodiment of the application may also have the following structure:
  • the embodiment of the present application also provides a communication device 1200 for realizing the functions of the terminal and the network device in the above method, that is, the functions of the first communication device, the second communication device or the third communication device.
  • the communication device may be a first communication device, a second communication device or a third communication device, or may be a device in the first communication device, a second communication device or a third communication device, or be capable of communicating with the first communication device,
  • the second communication device or the third communication device matches the device used.
  • the communication device 1200 may be a system on a chip.
  • the system-on-a-chip may be composed of chips, or may include chips and other discrete devices.
  • the communication device 1200 includes at least one processor 1210, configured to implement the functions of the first communication device, the second communication device, or the third communication device in the method provided in the embodiment of the present application.
  • the communication device 1200 may also include a communication interface 1220 .
  • the communication interface 1220 may be a transceiver, a circuit, a bus, a module or other types of communication interfaces, and is used for communicating with other devices through a transmission medium.
  • the communication interface 1220 is used for devices in the communication device 1200 to communicate with other devices.
  • the processor 1210 can execute the functions executed by the processing unit 910 in the first communication device 900 ; the communication interface 1220 can be used to execute the functions executed by the transceiver unit 920 in the communication device 900 .
  • the processor 1210 When the communication device 1200 is used to perform the operation performed by the first communication device, the processor 1210 is used to obtain first information, and the first information simultaneously carries indication information of the first transmission resource and the second transmission resource; wherein, the first The transmission resource is used to transmit the first data to the second communication device, the first data is the first output of the first machine learning model; the second transmission resource is used to receive the first feedback data from the second communication device, the first feedback data It is used to indicate the first gradient, and the first gradient is used to update the first machine learning model; the communication interface 1220 is used to transmit the first data to the second communication device on the first transmission resource.
  • the processor 1210 is used to: acquire first information, and the first information simultaneously carries indication information of the first transmission resource and the second transmission resource; wherein, the first information A transmission resource is used to receive second data from the first communication device, the second data is the data obtained after the first data is transmitted through the channel, and the first data is the first output of the first machine learning model; the second transmission resource is used for For transmitting the first feedback data to the first communication device, the first feedback data is used to indicate the first gradient, and the first gradient is used to update the first machine learning model; the communication interface 1220 is used to send the first feedback data to the first The communication device transmits first feedback data.
  • the processor 1210 is used to: obtain first information, and the first information simultaneously carries indication information of the first transmission resource and the second transmission resource; wherein, the first A transmission resource is used for the first communication device to transmit the first data to the second communication device, the first data is the first output of the first machine learning model; the second transmission resource is used for the second communication device to transmit the first data to the first communication device A feedback data, the first feedback data is used to indicate the first gradient, and the first gradient is used to update the first machine learning model; the communication interface 1220 is used to transmit the first information to the first communication device and/or the second communication device.
  • the communication interface 1220 is also used to perform other receiving or sending steps or operations performed by the first communication device, the second communication device or the third communication device in the above method embodiments.
  • the processor 1210 may also be configured to execute other corresponding steps or operations performed by the first communication device, the second communication device, or the third communication device in the above method embodiment except sending and receiving, which will not be repeated here.
  • the communication device 1200 may also include at least one memory 1230 for storing program instructions and/or data.
  • the memory 1230 is coupled to the processor 1210 .
  • 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 1210 may execute program instructions stored in memory 1230 .
  • at least one of the at least one memory may be integrated with the processor.
  • the memory 1230 is located outside the communication device 1200 .
  • a specific connection medium among the communication interface 1220, the processor 1210, and the memory 1230 is not limited.
  • the memory 1230, the processor 1210, and the communication interface 1220 are connected through the bus 1240.
  • the bus is represented by a thick line in FIG. 12, and the connection mode 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 1210 can be one or more central processing units (Central Processing Unit, CPU), and in the case where the processor 1210 is a CPU, the CPU can be a single-core CPU or a multi-core CPU .
  • the processor 1210 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, or a discrete hardware component, and may implement or execute the The disclosed methods, steps and logical block diagrams.
  • 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 1230 may include but not limited to hard disk (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.
  • a memory is, but is not limited to, any other medium that can be used to carry or store desired program code in the form of instructions or data structures and that can be accessed by a computer.
  • 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 communication device in this embodiment of the application may also have the following structure:
  • the embodiment of the present application also provides a device 1300, which can be used to realize the functions of the first communication device, the second communication device or the third communication device in the above method.
  • the device 1300 can be a communication device or a communication device chip.
  • the communication device includes:
  • the input-output interface 1310 may be an input-output circuit, and may also be called a communication interface.
  • the logic circuit 1320 may be a signal processor, a chip, or other integrated circuits that can implement the method of the present application.
  • At least one input and output interface 1310 is used for input or output of signals or data.
  • the input and output interface 1310 is used for communicating with a second communication device.
  • the input and output interface 1310 is used to output the first feedback data.
  • the logic circuit 1320 is configured to execute some or all steps of any method provided in the embodiments of the present application.
  • the logic circuit may realize the functions realized by the processing unit 901 in the first communication device 900 , the processing unit 1001 in the second communication device 1000 or the processing unit 1101 in the third communication device 1100 .
  • the device is the first communication device or is used for the first communication device, it is used to execute the steps performed by the first communication device (transmitter communication device) in various possible implementation manners in the above method embodiments,
  • the logic circuit 1320 is used to obtain the first output.
  • the device When the device is the second communication device or is used for the second communication device, it is used to execute the steps performed by the second communication device (receiving end communication device) in various possible implementation methods in the above method embodiments, such as logic circuit 1320 Used to determine the first gradient.
  • the terminal chip When the above communication device is a chip applied to the first communication device, the terminal chip implements the functions of the first communication device in the above method embodiment.
  • the terminal chip receives information from other modules (such as radio frequency modules or antennas) in the terminal, and the information is sent to the first communication device by the second communication device or the third communication device; or, the terminal chip sends information to the second communication device Other modules (such as a radio frequency module or an antenna) send information, and the information is sent by the first communication device to the second communication device or the third communication device.
  • modules such as radio frequency modules or antennas
  • the chip of the second communication device implements the functions of the second communication device in the above method embodiment.
  • the second communication device chip receives information from other modules (such as radio frequency modules or antennas) in the second communication device, and the information is sent to the second communication device by the first communication device or the third communication device; or, the second communication device
  • the chip of the communication device sends information to other modules (such as a radio frequency module or an antenna) in the second communication device, and the information is sent by the second communication device to the first communication device or the third communication device.
  • the chip of the third communication device implements the function of the third communication device in the above method embodiment.
  • the third communication device chip receives information from other modules (such as radio frequency modules or antennas) in the third communication device, and the information is sent to the third communication device by the first communication device or the second communication device; or, the third communication device
  • the chip of the communication device sends information to other modules (such as a radio frequency module or an antenna) in the third communication device, and the information is sent by the third communication device to the first communication device or the second communication device.
  • the embodiment of the present application also provides a computer-readable storage medium, the computer-readable storage medium stores a computer program, and the computer program is executed by hardware (such as a processor, etc.) to Implement some or all steps of any method executed by any device in the embodiments of the present application.
  • the embodiment of the present application also provides a computer program product including program instructions, when the computer program product is run on a computer, the computer is made to execute any one of the above aspects. Some or all steps of a method.
  • the present application further provides a chip or a chip system, where the chip may include a processor.
  • the chip may also include memory (or storage module) and/or transceiver (or communication module), or, the chip is coupled with memory (or storage module) and/or transceiver (or communication module), wherein the transceiver ( or communication module) can be used to support the chip for wired and/or wireless communication, and the memory (or storage module) can be used to store a program, and the processor can call the program to implement any of the above method embodiments and method embodiments.
  • the system-on-a-chip may include the above-mentioned chips, and may also include the above-mentioned chips and other discrete devices, such as memory (or storage module) and/or transceiver (or communication module).
  • the present application further provides a communication system, which may include at least one of the above first communication device, second communication device, and third communication device.
  • the communication system may be used to implement the operations performed by the terminal or the network device in any of the foregoing method embodiments and any possible implementation manners of the method embodiments.
  • the communication system may have a structure as shown in FIG. 1a or FIG. 1b.
  • the disclosed system, device and method can be implemented in other ways.
  • the device embodiments described above are only illustrative.
  • the division of the units is only a logical function division. In actual implementation, there may be other division methods.
  • multiple units or components can be combined or May be integrated into another system, or some features may be ignored, or not implemented.
  • the mutual coupling or direct coupling or communication connection shown or discussed may be through some interfaces, and the indirect coupling or communication connection of devices or units may be in electrical, mechanical or other forms.
  • the units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place, or may be distributed to multiple network units. Part or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.
  • each functional unit in each embodiment of the present application may be integrated into one processing unit, each unit may exist separately physically, or two or more units may be integrated into one unit.
  • the 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 technical solution of the present application is essentially or part of the contribution to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium , including several instructions to make a computer device (which may be a personal computer, a server, or a network device, etc.) execute all or part of the steps of the methods described in the various embodiments of the present application.
  • the aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (read-only memory, ROM), random access memory (random access memory, RAM), magnetic disk or optical disc and other media that can store program codes. .

Landscapes

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

Abstract

本申请实施例公开了一种数据传输方法和相关装置，用于以低信令开销实现端到端通信系统中收发两端部署的机器学习模型的训练。本申请实施例方法应用于包括第一通信装置和第二通信装置的通信系统，第一通信装置部署有第一机器学习模型，第二通信装置部署有第二机器学习模型，该方法包括：获取同时承载有第一传输资源和第二传输资源的指示信息的第一信息；其中，第一传输资源用于第一通信装置向第二通信装置传输第一机器学习模型的第一输出；第二传输资源用于第一通信装置接收来自第二通信装置的，包括第一梯度的第一反馈数据，第一梯度用于更新第一机器学习模型；第一通信装置在第一传输资源上，向第二通信装置传输第一输出。

Description

一种数据传输方法和相关装置
本申请要求于2021年5月27日提交中国国家知识产权局、申请号为202110587562.0、发明名称为“一种数据传输方法和相关装置”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本申请实施例涉及通信领域，尤其涉及一种数据传输方法和相关装置。
背景技术
人工智能(artificial intelligence，AI)技术日益成熟，目前已经可以将AI技术应用于通信领域。机器学习模型是AI领域的常用工具，机器学习模型包括输入层、输出层和至少一个中间层。具体到通信领域，可以将发送端的输入量作为输入层，将接收端的输出量作为输出层，将发送端、接收端上部署的部分机器学习模型作为中间层，这种分别部署在发送端和接收端上的机器学习模型组成了一个端到端的机器学习模型系统。
对于一般的机器学习模型来说，在层与层之间，作为中间数据的前向推理的结果和反向传播的梯度是直接传输的。在通信领域中，端到端机器学习模型系统的部分层与层之间，前向推理的结果的传输，以及反向传播的梯度的传输，需要经过通信信道，信道会对传输产生影响，进而也就影响了端到端机器学习模型系统的训练。由于信道状态随时可能变化，对应的机器学习模型系统也就跟着变化，随时可能需要发起对收发两端机器学习模型的训练。
在通信领域中，训练机器学习模型需要发送端与接收端交互传输中间数据，也就需要为发送端和接收端各自分配传输资源用于传输中间数据。机器学习模型系统需要经过多次中间数据的交互才能完成训练，因此需要进行多次的传输资源分配和调度，传输资源的分配和调度占用了较大的信令开销。
发明内容
本申请实施例提供了一种数据传输方法和相关装置，用于以低信令开销实现端到端通信系统中收发两端部署的机器学习模型的训练，完成所述通信系统收发两端的数据传输。
本申请实施例第一方面提供了一种数据传输方法，该方法应用于通信系统中的第一通信装置，该通信系统中还包括第二通信装置；其中，第一通信装置部署有第一机器学习模型，第二通信装置部署有第二机器学习模型，第一机器学习模型和第二机器学习模型用于实现第一通信装置与第二通信装置之间的通信，该方法包括：
第一通信装置获取第一信息，第一信息包括第一传输资源和第二传输资源的指示信息；其中，第一传输资源用于传输前向推理结果，第二传输资源用于传输反向梯度；具体的，第一传输资源用于第一通信装置向第二通信装置发送第一数据，第一数据为第一机器学习模型的第一输出，即第一机器学习模型的前向推理结果；第二传输资源用于第一通信装置接收来自第二通信装置的第一反馈数据，第一反馈数据用于指示第一梯度，第一梯度即为 前述反向梯度，第一梯度用于更新第一机器学习模型；第一通信装置在第一传输资源上，向第二通信装置发送第一数据。
为了实现对机器学习模型系统的训练，需要确定前向推理结果和反向梯度的传输资源，这两个数据的传输方向是相反的。控制信令用于指示传输资源，现有的控制信令只能指示一个方向的传输资源。对于发送端设备(第一通信装置)或接收端设备(第二通信装置)来说，为了实现对机器学习模型系统的训练，需要获取多个控制信令，以确定用于传输前向推理结果和反向梯度的两个方向的传输资源。在本申请实施例中，通过同时指示第一传输资源和第二传输资源的第一信息，可以一次性确定两个方向的传输资源信息。将第一信息作为控制信令发出，可以减小通信系统中控制信令的数量，从而减小通信系统中的控制信令开销。
在一种可选的实施方式中，第一数据经过信道传输后得到第二数据，将第二数据输入第二机器学习模型，就可以得到第二输出，第一梯度为第二通信装置根据第二输出计算所得。
在本申请实施例中，第一传输资源用于传输第一机器学习模型的第一输出，即前向推理结果；第二传输资源用于传输第一梯度，即反向梯度。在本实施方式中，第一梯度的计算来源于第一输出，也就是说，第一信息中所指示的传输资源，该传输资源用于传输的前向推理结果和反向梯度是相互关联的，通过相互关联的前向推理结果和反向梯度，就可以实现对机器学习模型系统的一次训练。也就是说，本申请实施例中的第一信息，可以用于指示机器学习模型系统的一次训练所需的所有传输资源。
在一种可选的实施方式中，第一信息包括传输资源单元的指示信息，第一传输资源和第二传输资源包含于传输资源单元。
在本申请实施例中，传输资源单元包括：用于传输前向推理结果的第一传输资源，以及用于传输反向梯度的第二传输资源。传输资源单元所包含的第一传输资源和第二传输资源，用于实现对机器学习模型系统的训练。
在一种可选的实施方式中，第一信息为训练控制信息TCI，TCI即为传输资源单元的指示信息。
在本申请实施例中，TCI为传输资源单元的指示信息，通过TCI的传输，即可将前向推理结果和反向梯度的传输资源告知给发送端设备和/或第二通信装置设备，从而实现对机器学习模型系统的训练。
在一种可选的实施方式中，第一信息可以指示多个传输资源用于传输多个前向推理结果，还可以指示多个传输资源用于传输多个反向梯度。具体的，除了第一传输资源和第二传输资源的指示信息，第一信息还可以包括第三传输资源和第四传输资源的指示信息；其中，第三传输资源用于第一通信装置向第二通信装置发送第三数据，第三数据为第一机器学习模型的第三输出；第四传输资源用于第一通信装置接收来自第二通信装置的第二反馈数据，第二反馈数据用于指示第二梯度，第二梯度用于更新第一机器学习模型；第一通信装置在第三传输资源上，向第二通信装置发送第三数据。
在本申请实施例中，第一信息可以指示多个用于传输前向推理结果的传输资源(第一 传输资源和第三传输资源)，还可以同时指示多个用于传输反向梯度的传输资源(第二传输资源和第四传输资源)。将第一信息作为控制信令发出，控制信令就能指示多个前向和反向传输资源，可以实现对机器学习网络的多次训练。相比于现有技术中需要通过多个控制信令来指示多次训练所需的传输资源，通过本申请实施例的方法，减少了通信系统中需要传输的控制信令的数量，从而减小了通信系统中的控制信令开销。
在一种可选的实施方式中，第三数据经过信道传输后得到第四数据，将第四数据输入第二机器学习模型，就可以得到第四输出，第二梯度为第二通信装置根据第四输出计算所得。
在本申请实施例中，第四传输资源用于传输第二梯度，第二梯度的计算来源为第三输出，而第三传输资源用于传输第三输出。因此第三传输资源和第四传输资源可以用于传输相互关联的前向推理结果和反向梯度，可以实现对机器学习模型的一次训练。前述可知第一传输资源和第二传输资源也可以实现对机器学习模型的一次训练，而第一传输资源和第二传输资源所传输的前向推理结果(第一输出)和反向梯度(第一梯度)，与第三传输资源和第四传输资源所传输的前向推理结果(第三输出)和反向梯度(第二梯度)，是不同的，也就是说，第一传输资源和第二传输资源所指示的传输资源，用于实现对机器学习模型系统的一次训练，而第三传输资源和第四传输资源所指示的传输资源，用于实现对机器学习模型系统的另一次训练。因此，本申请实施例可以通过一个第一信息，指示机器学习模型系统的多次训练所需的所有传输资源，从而大大减小控制信令的开销。
在一种可选的实施方式中，第一传输资源为第一回合的前向传输资源，第二传输资源为第一回合的反向传输资源；第三传输资源为第二回合的前向传输资源，第四传输资源为第二回合的反向传输资源。
在一种可选的实施方式中，第一传输资源和第二传输资源位于：时域、频域、空域、码域、功率域中至少一项的固定传输资源位置上。
在本申请实施例中，将前向推理结果和反向梯度的传输资源配置在固定的传输资源上，若用标识来表示该固定传输资源，就可以通过较少的数据量表示两个方向的传输资源，减小第一信息的数据量，从而减小控制信令开销。
在一种可选的实施方式中，第一传输资源和第二传输资源位于：用于传输业务数据的时域、频域、空域、码域、功率域中至少一项的资源位置上。
在本申请实施例中，将前向推理结果和反向梯度的传输资源配置在用于传输业务数据的资源块中，可以在传输业务数据的过程中进行机器学习模型系统的训练，提升训练的灵活性。将传输资源灵活的设置在用于传输业务数据的数据块中，由于不需要等待业务数据传输完毕就能传输训练所需的数据，因此可以减小训练所需的传输资源之间的时间间隔，从而减小等待训练所需传输资源的时延，提升机器学习模型系统的训练效率。
在一种可选的实施方式中，第一通信装置获取第一信息，包括：第一通信装置接收来自第二通信装置的第一信息；或者，第一通信装置接收来自第三通信装置的第一信息，第三通信装置用于控制第一通信装置与第二通信装置之间的通信。
在一种可选的实施方式中，在第一通信装置获取第一信息之后，该方法还包括：第一 通信装置向第二通信装置传输第一信息。
在一种可选的实施方式中，该方法还包括：将训练数据输入第一机器学习模型，得到第一输出，即第一机器学习模型的前向推理结果；第一通信装置在第一传输资源上，向第二通信装置发送前述第一输出；第一通信装置在第二传输资源上，接收来自第二通信装置的第一反馈数据，第一反馈数据用于指示第一梯度，第一梯度即为前述反向梯度；第一通信装置根据第一梯度更新第一机器学习模型。
在一种可选的实施方式中，在第一通信装置获取第一信息之前，该方法还包括：第一通信装置接收来自第二通信装置的训练指示，训练指示用于指示对第一机器学习模型的训练。
在本申请实施例中，第一通信装置接收来自第二通信装置的训练指示，就可以根据该训练指示开启对第一机器学习模型的训练过程，例如可以开始计算第一输出；接收到第一信息，就可以马上在第一传输资源上将第一输出发送给第二通信装置，相较于在接收到第一信息之后再开始计算第一输出，提升了第一机器学习模型的训练效率。
本申请实施例第二方面提供了一种数据传输方法，该方法应用于通信系统中的第二通信装置，该通信系统还包括第一通信装置；其中，第一通信装置部署有第一机器学习模型，第二通信装置部署有第二机器学习模型，第一机器学习模型和第二机器学习模型用于实现第一通信装置与第二通信装置之间的通信，该方法包括：
第二通信装置获取第一信息，第一信息包括第一传输资源和第二传输资源的指示信息；其中，第一传输资源用于传输前向推理结果，第二传输资源用于传输反向梯度；具体的，第一传输资源用于第二通信装置接收来自第一通信装置的第二数据，第二数据为第一数据经过信道传输后所得的数据，第一数据为第一机器学习模型的第一输出；第二传输资源用于第二通信装置向第一通信装置发送第一反馈数据，第一反馈数据用于指示第一梯度，第一梯度用于更新第一机器学习模型；第二通信装置在第二传输资源上，向第一通信装置发送第一反馈数据。
本申请实施例第二方面的有益效果参见第一方面，此处不再赘述。
在一种可选的实施方式中，第二数据用于输入第二机器学习模型，并根据输出计算得到第三梯度，第三梯度用于更新第二机器学习模型。
在一种可选的实施方式中，将第二数据输入第二机器学习模型，就可以得到第二输出，第一梯度为第二通信装置根据第二输出计算所得。
在一种可选的实施方式中，第一信息包括传输资源单元的指示信息，第一传输资源和第二传输资源包含于传输资源单元。
在一种可选的实施方式中，第一信息为训练控制信息TCI，TCI即为传输资源单元的指示信息。
在一种可选的实施方式中，第一信息可以指示多个传输资源用于传输多个前向推理结果，还可以指示多个传输资源用于传输多个反向梯度。具体的，除了第一传输资源和第二传输资源的指示信息，第一信息还可以包括第三传输资源和第四传输资源的指示信息；其中，第三传输资源用于第二通信装置接收来自第一通信装置的第四数据，第四数据为第三 数据经过信道传输后所得的数据，第三数据为第一机器学习模型的第三输出；第四传输资源用于第二通信装置向第一通信装置发送第二反馈数据，第二反馈数据用于指示第二梯度，第二梯度用于更新第一机器学习模型；第二通信装置在第四传输资源上，向第一通信装置发送第二反馈数据。
在一种可选的实施方式中，第一传输资源为第一回合的前向传输资源，第二传输资源为第一回合的反向传输资源；第三传输资源为第二回合的前向传输资源，第四传输资源为第二回合的反向传输资源。
在一种可选的实施方式中，第四数据用于输入第二机器学习模型，并根据输出计算得到第四梯度，第四梯度用于更新第二机器学习模型。
在一种可选的实施方式中，将第四数据输入第二机器学习模型，就可以得到第四输出，第二梯度为第二通信装置根据第四输出计算所得，第四梯度也为第二通信装置根据第四输出计算所得。
在一种可选的实施方式中，第一传输资源和第二传输资源位于：时域、频域、空域、码域、功率域中至少一项的固定传输资源位置上。
在一种可选的实施方式中，第一传输资源和第二传输资源位于：用于传输业务数据的时域、频域、空域、码域、功率域中至少一项的资源位置上。
在一种可选的实施方式中，第二通信装置获取第一信息，包括：第二通信装置接收来自第一通信装置的第一信息；或者，第二通信装置接收来自第三通信装置的第一信息，第三通信装置用于控制第一通信装置与第二通信装置之间的通信。
在一种可选的实施方式中，在第二通信装置获取第一信息之后，该方法还包括：第二通信装置向第一通信装置发送第一信息。
在一种可选的实施方式中，该方法还包括：第二通信装置在第一传输资源上，接收来自第一通信装置的第二数据；第二通信装置将第二数据输入第二机器学习模型，得到第二输出，第二通信装置根据第二输出计算得到第一梯度；第二通信装置在第二传输资源上，向第一通信装置发送第一反馈数据，第一反馈数据用于承载第一梯度，第一梯度用于第一通信装置更新第一机器学习模型。
在一种可选的实施方式中，在第二通信装置获取第一信息之前，该方法还包括：向第一通信装置传输训练指示，训练指示用于指示对第一机器学习模型的训练。
本申请实施例第三方面提供了一种数据传输方法，该方法应用于通信系统中的第三通信装置，通信系统还包括第一通信装置和第二通信装置，第三通信装置用于控制第一通信装置和第二通信装置的之间的通信，第一通信装置部署有第一机器学习模型，第二通信装置部署有第二机器学习模型，第一机器学习模型和第二机器学习模型用于实现第一通信装置与第二通信装置之间的通信，方法包括：
第三通信装置获取第一信息，第一信息同时承载有第一传输资源和第二传输资源的指示信息；其中，第一传输资源用于第一通信装置向第二通信装置发送第一数据，第一数据为第一机器学习模型的第一输出；第二传输资源用于第二通信装置向第一通信装置发送第一反馈数据，第一反馈数据用于指示第一梯度，第一梯度用于更新第一机器学习模型；
第三通信装置向第一通信装置和/或第二通信装置发送第一信息。
在本申请实施例中，第三通信装置作为第一通信装置和第二通信装置之间通信的中心控制设备，可以向第一通信装置和/或第二通信装置传输第一信息，从而将前向推理结果和反向梯度的传输资源告知给第一通信装置和/或第二通信装置，实现对机器学习模型系统的训练。由于中心控制设备用于控制第一通信装置与第二通信装置之间的通信，因此中心控制设备与第一通信装置和/或第二通信装置之间会有其他控制信令的交互。可以在第一消息中承载其他控制信令所承载的内容，就能减少通信系统中控制信令的数量，从而减小通信系统中的控制信令开销。并且，第三通信装置作为第一通信装置与第二通信装置之间的中心控制设备，可以获知第一通信装置与第二通信装置之间的所有传输资源的使用情况，可以基于该使用情况作出更为合理的传输资源分配(例如可以尽量减小传输资源之间的间隔从而减小训练中数据传输的时延，或者将传输资源灵活地设置在已被分配出去的资源块中间等)。关于第一信息的有益效果，参见本申请实施例第一方面，此处不再赘述。
本申请实施例第三方面的有益效果，参见第一方面，此处不再赘述。
在一种可选的实施方式中，第一数据经过信道传输后得到第二数据，将第二数据输入第二机器学习模型，就可以得到第二输出，第一梯度为第二通信装置根据第二输出计算所得。
在一种可选的实施方式中，第一信息包括传输资源单元的指示信息，第一传输资源和第二传输资源包含于传输资源单元。
在一种可选的实施方式中，第一信息为训练控制信息TCI，TCI即为传输资源单元的指示信息。
在一种可选的实施方式中，第一信息可以指示多个传输资源用于传输多个前向推理结果，还可以指示多个传输资源用于传输多个反向梯度。具体的，除了第一传输资源和第二传输资源的指示信息，第一信息还可以包括第三传输资源和第四传输资源的指示信息；其中，第三传输资源用于第一通信装置向第二通信装置发送第三数据，第三数据为第一机器学习模型的第三输出；第四传输资源用于第二通信装置向第一通信装置发送第二反馈数据，第二反馈数据用于指示第二梯度，第二梯度用于更新第一机器学习模型；第一通信装置在第三传输资源上，向第二通信装置发送第三数据。
在一种可选的实施方式中，第三数据经过信道传输后得到第四数据，将第四数据输入第二机器学习模型，就可以得到第四输出，第二梯度为第二通信装置根据第四输出计算所得。
在一种可选的实施方式中，第一传输资源和第二传输资源位于：时域、频域、空域、码域、功率域中至少一项的固定传输资源位置上。
在一种可选的实施方式中，第一传输资源和第二传输资源位于：用于传输业务数据的时域、频域、空域、码域、功率域中至少一项的资源位置上。
在一种可选的实施方式中，在第三通信装置获取第一信息之前，该方法还包括：第三通信装置接收来自第二通信装置的训练指示，训练指示用于指示对第一机器学习模型和第二机器学习模型的训练。
本申请实施例第四方面提供了一种通信装置，其特征在于，通信装置为通信系统中的第一通信装置，通信系统还包括第二通信装置，第一通信装置部署有第一机器学习模型，第二通信装置部署有第二机器学习模型，第一机器学习模型和第二机器学习模型用于实现第一通信装置与第二通信装置之间的通信，第一通信装置包括：处理单元、收发单元；
处理单元用于：获取第一信息，第一信息包括第一传输资源和第二传输资源的指示信息；其中，第一传输资源用于向第二通信装置发送第一数据，第一数据为第一机器学习模型的第一输出；第二传输资源用于接收来自第二通信装置的第一反馈数据，第一反馈数据用于指示第一梯度，第一梯度用于更新第一机器学习模型；
收发单元用于：在第一传输资源上，向第二通信装置发送第一数据。
该第一通信装置用于执行前述第一方面的方法。
本申请实施例第五方面提供了一种通信装置，其特征在于，通信装置为通信系统中的第二通信装置，通信系统还包括第一通信装置，第一通信装置部署有第一机器学习模型，第二通信装置部署有第二机器学习模型，第一机器学习模型和第二机器学习模型用于实现第一通信装置与第二通信装置之间的通信，第二通信装置包括：处理单元、收发单元；
处理单元用于：获取第一信息，第一信息包括第一传输资源和第二传输资源的指示信息；其中，第一传输资源用于接收来自第一通信装置的第二数据，第二数据为第一数据经过信道传输后所得的数据，第一数据为第一机器学习模型的第一输出；第二传输资源用于向第二通信装置发送第一反馈数据，第一反馈数据用于指示第一梯度，第一梯度用于更新第一机器学习模型；
收发单元用于：在第二传输资源上，向第一通信装置发送第一反馈数据。
该第二通信装置用于实现前述第二方面的方法。
本申请实施例第六方面提供了一种通信装置，其特征在于，通信装置为通信系统中的第三通信装置，通信系统还包括第一通信装置和第二通信装置，第三通信装置用于控制第一通信装置和第二通信装置的之间的通信，第一通信装置部署有第一机器学习模型，第二通信装置部署有第二机器学习模型，第一机器学习模型和第二机器学习模型用于实现第一通信装置与第二通信装置之间的通信，第三通信装置包括：处理单元、收发单元；
处理单元用于：获取第一信息，第一信息包括第一传输资源和第二传输资源的指示信息；其中，第一传输资源用于第一通信装置向第二通信装置发送第一数据，第一数据为第一机器学习模型的第一输出；第二传输资源用于第二通信装置向第一通信装置发送第一反馈数据，第一反馈数据用于指示第一梯度，第一梯度用于更新第一机器学习模型；
收发单元用于：向第一通信装置和/或第二通信装置发送第一信息。
该第三通信装置用于实现前述第三方面的方法。
本申请实施例第七方面提供了一种通信装置，其特征在于，该通信装置为通信系统中的第一通信装置，通信系统还包括第二通信装置，第一通信装置部署有第一机器学习模型，第二通信装置部署有第二机器学习模型，第一机器学习模型和第二机器学习模型用于实现第一通信装置与第二通信装置之间的通信，第一通信装置包括：处理器、收发器；
收发器用于收发数据或信息；
处理器用于执行第一方面的数据传输方法。
本申请实施例第八方面提供了一种通信装置，其特征在于，通信装置为通信系统中的第二通信装置，通信系统还包括第一通信装置，第一通信装置部署有第一机器学习模型，第二通信装置部署有第二机器学习模型，第一机器学习模型和第二机器学习模型用于实现第一通信装置与第二通信装置之间的通信，第二通信装置包括：处理器、收发器；
收发器用于收发数据或信息；
处理器用于执行第二方面的数据传输方法。
本申请实施例第九方面提供了一种通信装置，其特征在于，通信装置为通信系统中的第三通信装置，通信系统还包括第一通信装置和第二通信装置，第三通信装置用于控制第一通信装置和第二通信装置的之间的通信，第一通信装置部署有第一机器学习模型，第二通信装置部署有第二机器学习模型，第一机器学习模型和第二机器学习模型用于实现第一通信装置与第二通信装置之间的通信，第三通信装置包括：处理器、收发器；
收发器用于收发数据或信息；
处理器用于执行第二方面的数据传输方法。
本申请实施例第十方面提供了一种通信设备，包括：
处理器、存储器、通信接口；
存储器用于存储程序指令；
处理器用于执行存储器中存储的程序指令，以实现前述第一方面至第三方面中任一方面的方法；
通信接口用于和其它设备进行通信。
本申请实施例第十一方面提供了一种通信设备，包括：处理器，该处理器与存储器耦合；
存储器，用于存储程序；
处理器，用于执行存储器中的程序，使得处理器执行前述第一方面至第三方面中任一方面的方法。
一种可能的实现中，通信设备还包括上述存储器。可选的，存储器和处理器集成在一起。另一种可能的实现中存储器位于所述通信设备之外。
本申请实施例第十二方面提供了一种计算机可读存储介质，该计算机可读存储介质中保存有程序，当该计算机执行该程序时，执行前述第一方面至第三方面中任一方面的方法。
本申请实施例第十三方面提供了一种计算机程序产品，当该计算机程序产品在计算机上执行时，计算机执行前述第一方面至第三方面中任一方面的方法。
本申请实施例第十四方面提供了一种通信系统，该通信系统包括前述第四方面的第一通信装置，和前述第五方面的第二通信装置。
在一种可能的实现中，该通信系统还包括前述第六方面的第三通信装置。
该通信系统用于实现前述第一方面至第三方面中任一方面的数据传输方法。
附图说明
图1a为本申请实施例提供的数据传输方法的应用场景示意图；
图1b为本申请实施例提供的数据传输方法的系统架构示意图；
图1c为本申请实施例提供的神经网络系统的示意图；
图2为本申请实施例提供的数据传输方法的一个交互示意图；
图3a为本申请实施例提供的单资源传输资源单元TU的示意图；
图3b为本申请实施例提供的多资源TU的示意图；
图4a为本申请实施例提供的固定TU type的示意图；
图4b为本申请实施例提供的动态TU type的示意图；
图5为本申请实施例提供的数据传输方法的另一交互示意图；
图6为本申请实施例提供的数据传输方法的另一交互示意图；
图7a为本申请实施例提供的模拟梯度回传方法中TU的一个示意图；
图7b为本申请实施例提供的模拟梯度回传方法中TU的另一示意图；
图8为本申请实施例提供的数据传输方法的另一交互示意图；
图9为本申请实施例提供的第一通信装置的一个结构示意图；
图10为本申请实施例提供的第二通信装置的一个结构示意图；
图11为本申请实施例提供的中心控制设备的一个结构示意图；
图12为本申请实施例提供的通信装置的一个结构示意图；
图13为本申请实施例提供的装置的一个结构示意图。
具体实施方式
下面结合附图，对本申请的实施例进行描述。本领域普通技术人员可知，随着技术的发展和新场景的出现，本申请实施例提供的技术方案对于类似的技术问题，同样适用。
本申请的说明书和权利要求书及上述附图中的术语“第一”、“第二”等是用于区别类似的对象，而不必用于描述特定的顺序或先后次序。应该理解这样使用的术语在适当情况下可以互换，这仅仅是描述本申请的实施例中对相同属性的对象在描述时所采用的区分方式。此外，术语“包括”和“具有”以及他们的任何变形，其目的在于覆盖不排他的包含，以便包含一系列单元的过程、方法、系统、产品或设备不必限于那些单元，而是可包括没有清楚地列出的或对于这些过程、方法、产品或设备固有的其它单元。
一、本申请实施例的应用场景。
1、通信系统架构。
请参阅图1a，图1a为本申请实施例提供的数据传输方法的应用场景示意图。如图1a所示的通信系统100中包括发送端通信装置110和接收端通信装置120。发送端通信装置110上部署有第一机器学习模型210，接收端通信装置120上部署有第二机器学习模型220，第一机器学习模型210和第二机器学习模型220用于实现发送端通信装置110与接收端通信装置120之间的通信。
具体的，发送端通信装置110可以是终端设备。在本申请实施例中，除了终端设备， 发送端通信装置110也可以是其他具备通信能力的装置，例如基站等，此处不做限定。
具体的，接收端通信装置120可以是基站。在本申请实施例中，除了基站，接收端通信装置120也可以是其他具备通信能力的装置，例如终端设备等，此处不做限定。
可选的，本申请实施例中的通信系统100中还可以包括中心控制设备，中心控制设备用于控制发送端通信装置110与接收端通信装置120之间的通信。
具体的，中心控制设备可以是基站。在本申请实施例中，除了基站，中心控制设备还可以是其他通信装置，例如边缘设备等，只要具备对发送端通信装置和接收端通信装置之间的通信的控制能力即可，此处不做限定。
具体的，本申请实施例中的通信系统可以是第五代移动通信技术(5th generation mobile communication technology，5G)、卫星通信及短距等无线通信系统，系统架构如图1b所示。该系统架构中包括网络设备130，网络设备130向终端设备140A和终端设备140B提供通信服务。无线通信系统也可以进行点对点通信，如多个终端设备之间互相通信。在本申请实施例中，网络设备130还可以向更多或更少的终端设备提供服务，终端设备的数量和种类根据实际需要确定，具体此处不做限定。
需要说明的是，本申请实施例中提及的无线通信系统包括但不限于：窄带物联网系统(narrow band-internet of things，NB-IoT)、长期演进系统(long term evolution，LTE)以及5G移动通信系统的三大应用场景增强移动宽带(enhanced mobile broadband，eMBB)，低时延高可靠通信(ultra-reliable&low-latency communication,URLLC)和海量物联网通信(massive machine type communications，mMTC)，以及未来可能出现的其他移动通信系统，此处不做限定。
在本申请实施例中，网络设备130是一种部署在无线接入网中为终端设备提供无线通信功能的装置。网络设备130可以为多个终端设备提供无线通信功能，例如图1b中所示的终端设备140A和终端设备140B。网络设备130可以包括各种形式的宏基站，微基站(也称为小站)，中继站，接入点等。在采用不同的无线接入技术的系统中，具备网络设备130功能的设备的名称可能会有所不同，例如，在LTE系统中，称为演进的节点B(evolved nodeB，eNB或者eNodeB),在第三代(3rd generation，3G)系统中，称为节点B(Node B)等。为方便描述，本申请实施例中，上述为终端设备提供无线通信功能的装置统称为网络设备或基站(base station，BS)。
本申请中的网络设备130可以是LTE中的演进型基站(evolutional Node B，eNB或eNodeB)；或者5G网络中的基站，宽带网络业务网关(broadband network gateway，BNG)，汇聚交换机或非第三代合作伙伴项目(3rd generation partnership project，3GPP)接入设备等，本申请实施例对此不作具体限定。可选的，本申请实施例中的基站可以包括各种形式的基站，例如：宏基站、微基站(也称为小站)、中继站、接入点、下一代基站(gNodeB，gNB)、传输点(transmitting and receiving point，TRP)、发射点(transmitting point，TP)、移动交换中心以及设备到设备(Device-to-Device，D2D)、车辆外联(vehicle-to-everything，V2X)、机器到机器(machine-to-machine，M2M)通信、物联网(Internet of Things)通信中承担基站功能的设备等，本申请实施例对此不作具体限 定。
本申请实施例中所涉及到的终端设备可以包括各种具有无线通信功能的手持设备、车载设备、可穿戴设备、计算设备或连接到无线调制解调器的其它处理设备。终端设备也可以称为终端(terminal)，终端设备还可以是用户单元(subscriber unit)、蜂窝电话(cellular phone)、智能手机(smart phone)、无线数据卡、个人数字助理(personal digital assistant，PDA)电脑、平板型电脑、无线调制解调器(modem)、手持设备(handset)、膝上型电脑(laptop computer)、机器类型通信(machine type communication，MTC)终端等，此处不做限定。
在本申请实施例中，还可以有更多或更少或终端设备在这个通信系统中，终端设备的数量和种类根据实际需要确定，具体此处不做限定。
本申请实施例中提及的终端设备，可以是一种具有无线收发功能的设备，具体可以指用户设备(user equipment，UE)、接入终端、用户单元(subscriber unit)、用户站、移动台(mobile station)、远方站、远程终端、移动设备、用户终端、无线通信设备、用户代理或用户装置。终端设备还可以是卫星电话、蜂窝电话、智能手机、无线数据卡、无线调制解调器、机器类型通信设备、可以是无绳电话、会话启动协议(session initiation protocol，SIP)电话、无线本地环路(wireless local loop，WLL)站、个人数字助理(personal digital assistant，PDA)、具有无线通信功能的手持设备、计算设备或连接到无线调制解调器的其它处理设备、车载设备、高空飞机上搭载的通信设备、可穿戴设备、无人机、机器人、设备到设备(device-to-device，D2D)通信中的终端、车辆外联(vehicle to everything，V2X)中的终端、虚拟现实(virtual reality，VR)终端设备、增强现实(augmented reality，AR)终端设备、混合现实(mixed reality，MR)、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)中的无线终端或者未来通信网络中的终端设备等，本申请不作限制。
在本申请实施例中，若发送端通信装置110是网络设备，接收端通信装置120可以是终端设备，也可以是网络设备；若发送端通信装置110是终端设备，接收端通信装置120可以是网络设备，也可以是终端设备，此处不做限定。
上面说明了本申请实施例的通信系统架构，接下来说明本申请实施例中，机器学习模型中的数据处理与传输过程，以及如何进行机器学习模型的训练。
2、机器学习模型系统中的数据处理与传输。
在本申请实施例中，图1a中的机器学习模型系统200可以为神经网络系统。接下来将以神经网络系统为例，说明机器学习模型系统200中的数据处理及传输过程。请参阅图1c，图1c为本申请实施例提供的神经网络系统的示意图。其中，第一神经网络210A部署在发送端通信装置110上，第二神经网络220A部署在接收端通信装置120上。图1c中的s为神经网络系统的输入值，将s输入第一神经网络210A中，第一神经网络210A计算得到对应的输出值X，并通过通信信道将X发送给第二神经网络220A。由于通信信道的影响，X 传输之后会变为Y。将Y输入第二神经网络220A中，第二神经网络220A计算得到对应的输出值r，r即为整个神经网络系统基于输入值s得到的输出值。若神经网络系统仅用于传输数据，则r是神经网络系统对s的还原值。除了传输数据，神经网络系统还可以用于进行图像处理、数据计算等，则r就是对应的推理结果。
由于通信信道随时间可能发生变化，导致Y随信道的变化也发生变化，因此需要基于信道变化调整第一神经网络210A的参数和第二神经网络220A的参数，确保调整参数后的神经网络系统，可以得到准确的输出值r。
上面以神经网络系统为例，说明了为什么要对机器学习模型系统进行训练，接下来说明机器学习模型系统训练的具体过程。
在机器学习模型系统训练的过程中，发送端通信装置110将训练数据输入第一机器学习模型210中，可以得到前向推理结果。发送端通信装置110将前向推理结果发送给接收端通信装置120，接收端通信装置将前向推理结果输入第二机器学习模型220中，可以得到输出值，再基于输出值计算得到第一梯度和第二梯度。其中，第一梯度用于更新第一神经网络210A的参数θ，第二梯度用于更新第二神经网络220A的参数ω。接收端通信装置120得到第一梯度后，向发送端通信装置110发送该第一梯度，发送端通信装置110再根据第一梯度更新第一机器学习模型210的参数θ。使用第一梯度和第二梯度更新第一机器学习模型和第二机器学习模型的过程被称为梯度的反向传递，以下将用于更新第一机器学习模型的第一梯度称为反向梯度。
也就是说，在机器学习模型的训练过程中，需要发送端通信装置110向接收端通信装置120发送前向推理结果，还需要接收端通信装置120向发送端通信装置110反馈基于该前向推理结果得到的反向梯度。
在本申请实施例中，由于接收端通信装置120也要执行发送数据的动作，为了防止混淆，后续统一用第一通信装置表示发送端通信装置110，用第二通信装置表示接收端通信装置120。
在本申请实施例中，从第一通信装置向第二通信装置传输的数据为前向推理结果，因此将第一通信装置向第二通信装置的方向称为前向；从第二通信装置向第一通信装置传输的数据为反向梯度，因此将第二通信装置向第一通信装置的方向称为反向。
在通信系统中，通过控制信令来指示第一通信装置与第二通信装置之间的数据传输资源，在数据传输的过程中，需要第一通信装置和第二通信装置都获知前向和反向的传输资源，才能使得第一通信装置和第二通信装置在该资源上顺利传输数据。
为了使第一通信装置和第二通信装置都获知前向推理结果和反向梯度的传输资源，需要向第一通信装置和/或第二通信装置传输控制信令，控制信令用于指示数据的传输资源。具体的，控制信令可以由第一通信装置、第二通信装置或中心控制设备中的任一个生成并传输出去，使得第一通信装置和第二通信装置都能获知该控制信令中的内容。例如，若控制信令由第一通信装置生成，则第一通信装置会将该控制信令传输给第二通信装置；其他情况以此类推，不再一一赘述。
在一种可能的实现中，控制信令中仅包含前向和反向中一个方向的数据传输资源，由 于机器学习模型系统的训练需要两个方向的数据传输，因此前向与反向的传输各需要发送一次控制信令，且机器学习模型系统的训练需要进行多次的数据传输才能完成，使得控制信令的数量较多，通信系统中的控制信令开销较大。
针对上述缺陷，本申请实施例提供了一种数据传输方法和相关设备，用于以低信令开销实现端到端通信系统中收发两端部署的机器学习模型的训练，完成所述通信系统收发两端的数据传输。接下来描述本申请实施例所提供的数据传输方法。
二、本申请实施例中的数据传输方法。
请参阅图2，图2为本申请实施例提供的数据传输方法的一个交互示意图，该方法包括：
201、第一通信装置获取第一信息。
在机器学习网络模型系统的训练过程中，第一通信装置需要确定用于传输前向推理结果的前向传输资源和用于传输反向梯度的反向传输资源。具体的，第一通信装置可以通过第一信息确定前向传输资源和反向传输资源。
第一通信装置获取第一信息，第一信息包括第一传输资源和第二传输资源的指示信息。其中，第一传输资源即为前述前向传输资源，第二传输资源即为前述反向传输资源。
可选的，在本申请实施例中，第一通信装置可以通过接收训练控制信息(training control information，TCI)获取第一信息。与下行控制信息(downlink control information，DCI)、上行控制信息(uplink control information，UCI)类似，TCI也用于指示传输资源。区别在于，DCI和UCI仅能指示一个方向上的传输资源，而TCI可以指示前向与反向两个方向的传输资源。
在本申请实施例中，为了更清楚地说明TCI所指示的信息，定义了训练单元(training unit，TU)。其中，TU包括前向传输资源和反向传输资源，用于传输机器学习模型系统训练所需的前向推理结果和反向梯度。
在不同维度上，可以对TU进行多种分类，例如基于承载的资源种类、TU中前向和反向传输资源的大小比例、TU所指示的资源位置等，接下来将分别介绍不同种类的TU。
1)单资源TU与多资源TU。
可选的，基于承载的资源种类，可以将TU分为单资源TU和多资源TU。
单资源TU，包含机器学习模型系统的单回合训练的传输资源。如图3a所示，图3a为本申请实施例提供的单资源TU的示意图。单资源TU中，包含一回合训练所需的前向和反向传输资源。图中的F表示前向传输资源，G表示反向传输资源。
在本申请实施例中，以第一传输资源和第二传输资源为例进行说明。TU中包括的第一传输资源为一回合训练中的前向传输资源，TU中包括的第二传输资源为该回合训练中的反向传输资源。
在本申请实施例中，第一传输资源和第二传输资源也可以不是同一训练回合的传输资源，此处不做限定。
在本申请实施例中，若单资源TU中包含同一回合的前向传输资源与反向传输资源，则可以在一个TCI中，指示一回合的机器学习模型训练所需的传输资源，若TCI为一个单独 的控制信令，则可以用一个控制信令指示一回合的机器学习模型训练所需的传输资源，减少了通信系统中控制信令的数量。
多资源TU：包含多回合训练所需传输资源。如图3b所示，1个多资源TU中包含多回合端到端训练所需的前向和反向传输资源。考虑到减少前向反向传输干扰，前向和反向传输资源不会出现在相同的时间段上。
例如，TU中可以包括第一至第四传输资源，第一传输资源为第一回合的前向传输资源，第二传输资源为第一回合的反向传输资源；第三传输资源为第二回合的前向传输资源，第四传输资源为第二回合的反向传输资源。前述第一回合和第二回合并不限定两回合之间的前后时序，只用于区分不同的回合。前述两回合仅是对TU所包括的多回合训练的示例，并不造成对TU所包括训练回合次数的限定，在本申请实施例中，TU可以包括更多回合的训练所需的传输资源，此处不做限定，
2)TU type。
基于TU中前向和反向传输资源的大小比例，可以确定不同的TU type。
在不同的训练回合中，前向和反向传输的数据量可能会有所变化，前向与反向传输资源的大小比例可能会有所不同。因此在本申请实施例中，定义多种TU类型(type)，不同类型的TU中前向和反向传输资源的比例不同。例如，如图3a中的A图所示，TU type1中前向和反向传输资源比例为1:1；如图3a中的B图所示，type2中前向和反向传输资源比例为2：1。
在多资源TU的情况下，可以定义多种TU type，对应不同的前向和反向传输资源比例。
可选的，在训练的前期(靠前的回合的训练过程中)，或者使用空口直接进行数据传输(模拟传输，在图6所示的实施例地步骤602中将会展开解释)的情况下，前向推理和反向梯度回传的数据量几乎相同，因此可以在多回合的训练中使用TU type1，即1:1的前向和反向传输资源进行训练所需的通信。随着训练的进行，大量梯度的取值逐渐趋近于0，反向梯度的数据量逐渐减小。因此训练中反向传输所需的资源将逐渐少于前向传输所需的资源，例如，当训练轮次数大于0小于50时，前向传输的数据量与反向传输的数据量差不多，可以选择TU type1；当训练轮次数大于等于50小于100时，反向传输的数据量接近前向传输数据量的一半，因此选择TU type2。
3)固定位置TU与动态位置TU。
基于TU所指示的资源位置，可以将TU分为固定位置TU与动态位置TU。
固定位置TU：如图4a所示，将TU配置在固定的资源位置上。
动态位置TU：如图4b所示，不固定TU的位置，将TU配置在动态可变的资源位置上。
可选的，可以在其他业务(如eMBB业务)所占用的传输资源中，动态地选择一部分资源进行端到端机器学习模型训练所需的传输。例如图4b中所示，就占用了用户设备(user equipment，UE)1的传输资源和UE2的传输资源。
可选的，如图4a所示，在固定位置TU的情况下，可以将第一传输资源和第二传输资源配置在固定的时频传输资源位置上。在本申请实施例中，除了配置在固定的时频位置上，也可以不将时域和频域两个维度结合起来分配，而是单独考虑将传输资源配置在固定的时 域传输资源位置上，或单独考虑将传输资源配置在固定的频域传输资源位置上，此处不作限定。在本申请实施例中，除了时域或频域，还可以将第一传输资源和第二传输资源配置在其他维度的固定传输资源位置上，例如空域、码域、功率域等维度，此处不做限定。
在固定位置TU的情况下，将TU配置在固定的传输资源上，若用标识来表示该固定传输资源，就可以通过较少的数据量表示两个方向的传输资源，减小TCI的数据量，从而减小控制信令开销。
例如，图4a左上角的图中共有16个资源块，指示16种可能的资源块位置最少需要4个比特，而将TU固定地配置在图中阴影所示的4个资源块中时，只需要2个比特就可以指示TU所在的资源位置，减少了TCI的数据量。
可选的，如图4b所示，在动态位置TU的情况下，可以将TU配置在用于传输业务数据的时域资源位置上。在本申请实施例中，除了时域，还可以将第一传输资源和第二传输资源配置在用于传输业务数据的其他维度的资源位置上，例如频域、空域、码域、功率域等维度，此处不做限定。
在动态位置TU的情况下，将TU配置在用于传输业务数据的资源块中，可以在传输业务数据的过程中进行机器学习模型系统的训练，提升训练的灵活性，减小等待训练所需传输资源分配的时延，提升机器学习模型系统的训练效率。
在本申请实施例中，训练单元TU也称为传输资源单元。不限定TU中的前向和反向传输资源所传输的数据内容，除了用于传输上述前向推理结果和反向梯度，也可以用于传输其他的数据，此处不做限定。
在本申请实施例中，不限定TCI的应用场景，除了用于机器学习模型系统的训练，还可以应用于其他场景，只要在该场景下，两个通信装置之间进行数据的交互传输即可，此处不做限定。
在一种可能的实现中，TCI中具体可以包括：TU所在频域位置信息和TU所在时域位置信息。
可选的，在一种可能的实现中，TCI中还可以包括：前向传输调制编码方案和反向传输调制编码方案。
在一种可能的实现中，TCI中可以包括：本回合前向传输资源所在频域位置信息，本回合前向传输资源所在时域位置信息，本回合反向传输资源所在频域位置信息和本回合反向传输资源所在时域位置信息。
机器学习模型系统需要经过多回合的训练，在本申请实施例中，回合表示机器学习模型系统的训练回合，一回合的训练包括前向推理结果的传输和反向梯度的传输。
可选的，当多个回合的资源都在同一个TU中时，可以重复前述字段，即“第1回合前向传输资源所在频域位置信息，第1回合前向传输资源所在时域位置信息，第1回合反向传输资源所在频域位置信息，第1回合反向传输资源所在时域位置信息；第2回合前向传输资源所在频域位置信息，第2回合前向传输资源所在时域位置信息，第2回合反向传输资源所在频域位置信息，第2回合反向传输资源所在时域位置信息；…，第k回合前向传输资源所在频域位置信息，第k回合前向传输资源所在时域位置信息，第k回合反向传输 资源所在频域位置信息，第k回合反向传输资源所在时域位置信息”。通过这个形式去指示同一多资源TU中多回合训练所需的资源的位置。
在本申请实施例中，当多个回合的资源都在同一个TU中时，不限定重复字段中所包含的内容。如前所述，每个重复字段中都包括同一回合中，前向、反向传输资源的频域位置信息，前向、反向传输资源的时域位置信息。实际上，并不限定重复字段中各内容出现的前后顺序，以及重复字段中所包括的内容，只要该重复字段可以用于指示单回合训练所要用到的传输资源即可，此处不做限定。
可选的，在多资源TU的TCI中，也可以包含标识，用于指示各传输资源对应的训练回合，以区分不同的训练回合。
前述通过不同回合对应的重复字段在TCI字段中的前后位置关系，来区分不同的训练回合的方式，不需要额外的标识来表示传输资源所对应的训练回合，减小了TCI的数据量，节省了信令开销。
可选的，在固定TU的情况下，也可以通过上述标识或TCI字段中的重复字段的前后位置关系来区分不同训练回合的传输资源。
在本申请实施例中，除了通过接收TCI，第一通信装置也可以通过其他方式获取第一信息。例如，第一通信装置确定第一信息中的传输资源，从而获取第一信息等，此处不做限定。
TCI可以来自第二通信装置，也可以来自其他的通信装置，在后面的实施例中将会详细说明TCI的传输，此处不做限定。
202、第二通信装置获取第一信息。
与第一通信装置相同，为了训练机器学习网络模型系统，第二通信装置也需要确定前向传输资源和反向传输资源。具体的，第二通信装置可以通过第一信息确定前向传输资源和反向传输资源。
可选的，在本申请实施例中，第二通信装置可以确定第一信息中的传输资源，从而获取第一信息；也可以通过接收TCI获取第一信息，此处不做限定。
TCI可以来自第一通信装置，也可以来自其他的通信装置，在后面的实施例中将会详细说明TCI的传输，此处不做限定。
对于TU和TCI的说明，参见步骤201，此处不再赘述。
值得注意的是，在本申请实施例中，并不限定步骤201与步骤202之间的时序关系，可以先执行步骤201再执行步骤202，也可以先执行步骤202再执行步骤201，此处不做限定。
203、第一通信装置在第一传输资源上，向第二通信装置发送第一数据。
TU中的第一传输资源用于第一通信装置向第二通信装置发送第一数据，第一数据为第一机器学习模型的第一输出，即第一机器学习模型的前向推理结果。
第一通信装置在第一传输资源上，向第二通信装置发送第一数据。
可选的，基于第一数据计算得到的梯度，也可以用于第二机器学习模型的更新，在图5至图8所示实施例将会展开说明，此处不再赘述。
204、第二通信装置在第二传输资源上，向第一通信装置发送第一反馈数据。
第二传输资源用于第二通信装置向第一通信装置发送第一反馈数据，第一反馈数据用于指示第一梯度，第一梯度即为前述反向梯度，第一梯度用于更新第一机器学习模型。
第二通信装置在第二传输资源上，向第一通信装置发送第一反馈数据。
可选的，若步骤201和/或202中的TU为多资源TU，除了第一传输资源和第二传输资源，TU中还可以包括第三传输资源和第四传输资源，其中，第三传输资源用于第一通信装置向第二通信装置发送第三数据，第三数据为第一机器学习模型的第三输出；第四传输资源用于第二通信装置向第一通信装置发送第二反馈数据，第二反馈数据用于指示第二梯度，第二梯度用于更新第一机器学习模型。
可选的，若TU中包括第三传输资源和第四传输资源，则还可能执行步骤205和步骤206。
205、第一通信装置在第三传输资源上，向第二通信装置发送第三数据。
TU中的第三传输资源用于第一通信装置向第二通信装置发送第三数据，第三数据为第一机器学习模型的第三输出，即第一机器学习模型的前向推理结果。
第一通信装置在第三传输资源上，向第二通信装置发送第三数据。
可选的，与第一数据类似，基于第三数据计算得到的梯度，也可以用于第二机器学习模型的更新。
206、第二通信装置在第四传输资源上，向第一通信装置发送第二反馈数据。
第四传输资源用于第二通信装置向第一通信装置发送第二反馈数据，第二反馈数据用于指示第二梯度，第二梯度即为前述反向梯度，第二梯度用于更新第一机器学习模型
第二通信装置在第四传输资源上，向第一通信装置发送第二反馈数据。
机器学习模型需要进行多回合的训练，可选的，可以在多回合中采用统一的TU类型，例如采用单资源TU或多资源TU或者采用type1的TU。或者也可以在多回合的训练过程中，改变TU类型，例如先采用单资源TU，从某一回合开始改为多资源TU，或者先采用type1的TU，从某一回合开始改为type2的TU等，此处不做限定。值得注意的是，步骤205和206是可选步骤，当TU中不包括第三传输资源和第四传输资源时，可以不执行步骤205和206，此处不做限定。
在本申请实施例中，TU可以由第一通信装置、第二通信装置和中心控制设备中的任一个来确定，TCI也就由对应的装置发出，接下来将分别描述。
1、由第一通信装置来确定TU。
请参阅图5，图5为本申请实施例提供的数据传输方法的流程图，如图2所示，该方法包括：
501、第一通信装置初始化第一机器学习模型，第二通信装置初始化第二机器学习模型。
具体的，可以采用本地随机初始化方法进行机器学习模型的初始化。除了本地初始化也可以通过其他方式进行机器学习模型的初始化，例如从指定的模型服务器下载相应的机器学习模型参数等，此处不做限定。
值得注意的是，步骤501为可选步骤，在机器学习模型系统的训练过程中，并不是每 轮训练之前都要执行步骤501。
502、第一通信装置通过机器学习模型系统向第二通信装置传输测试数据，相应的，第二通信装置接收测试数据。
第一通信装置上部署了第一机器学习模型。将需要传输给第二通信装置的测试数据输入第一机器学习模型，可以得到第一输出，在本申请实施例中，第一输出也称为第一数据。
第一通信装置将第一数据传输给第二通信装置，由于通信信道会对数据产生影响，因此第二通信装置将会接收到第二数据，第二数据为第一数据经过信道传输后得到的数据。
第二通信装置上部署了第二机器学习模型，将第二数据输入第二机器学习模型，可以得到第二输出，第二输出即为对前述输入第一机器学习模型的业务数据的推理结果。
在本申请实施例中，由第一机器学习模型和第二机器学习模型构成的机器学习模型系统，不仅可以如上所述的进行业务数据的传输，还可以进行数据处理，例如进行语义分析、图像分类等，在第二机器学习模型处得到的第二输出，对应的就可以是语义分析结果、图像分类结果等，此处不做限定。
可选的，步骤502中传输的测试数据也可以是业务数据、传输信令等，只要能实现步骤503中的性能评估即可，此处不做限定。
503、第二通信装置进行性能评估，确定需要训练机器学习模型系统。
基于步骤502中的业务数据的传输过程，第二通信装置可以进行性能评估，评估基于当前的机器学习模型系统，进行数据传输的性能。具体的，第二通信装置可以评估数据传输的误码率，当误码率高于某一阈值(例如eMBB中常用的0.1，uRLLC中常用的0.00001)时，确定需要训练机器学习模型系统。
在本申请实施例中，除了评估误码率，第二通信装置还可以通过其他性能确定是否需要训练机器学习模型系统，例如评估吞吐、时延等性能，此处不做限定。可选的，当uRLLC中时延大于或等于1ms，可以确定需要训练机器学习模型系统。
可选的，若第二通信装置进行性能评估后，确定性能满足预设条件，不需要训练机器学习模型系统，则可以重复步骤502和性能评估，直到性能无法满足预设条件，确定需要训练机器学习模型(即出现步骤503)。
504、第二通信装置向第一通信装置传输训练指示，相应的，第一通信装置接收训练指示。
第二通信装置确定了需要训练机器学习模型系统，就可以向第一通信装置传输训练指示，训练指示用于指示对第一机器学习模型的训练。
505、第一通信装置确定TU。
第一通信装置接收到训练指示，可以确定传输资源单元TU，TU包括第一传输资源和第二传输资源。其中，第一传输为前向的传输资源，用于第一通信装置向第二通信装置传输前向推理结果；第二传输资源为反向的传输资源，用于第二通信装置向第一通信装置传输反向梯度。
506、第一通信装置向第二通信装置传输TCI，相应的，第二通信装置接收TCI。
确定了TU，第一通信装置就可以确定TCI，向第二通信装置传输该TCI。TCI为TU的 指示信息，用于指示TU中所包括的所有传输资源的位置信息。
在本申请实施例中，TCI也称为第一信息，此处不做限定。
在本申请实施例中，TCI不一定由第一通信装置发出。在第一通信装置确定TU后，也可以将TU告知给其它设备(例如中继设备)，由其它设备来向第二通信装置传输TCI，此处不做限定。
507、第一通信装置和第二通信装置根据TCI中指示的传输资源，对机器学习模型系统进行训练。
接下来以单资源TU为例，说明机器学习模型系统的学习过程。
TU中包括第一传输资源和第二传输资源；其中，第一传输资源用于指示前向推理结果的传输资源；第二传输资源用于指示反向梯度的传输资源。TCI中包含了第一传输资源和第二传输资源的位置信息。
第一通信装置将训练数据输入第一机器学习模型，可以得到第一输出，在本申请实施例中，第一输出也称为第一数据。
第一通信装置在第一传输资源上向第二通信装置传输第一数据，由于通信信道会对数据产生影响，因此第二通信装置将会在第一传输资源上接收到第二数据，第二数据为第一数据经过信道传输后得到的数据。
第二通信装置将第二数据输入第二机器学习模型，可以得到第二输出，第二输出即为对前述输入第一机器学习模型的训练数据的推理结果。
在本申请实施例中，由第一机器学习模型和第二机器学习模型构成的机器学习模型系统，不仅可以进行业务数据的传输，还可以进行数据处理，例如进行语义分析、图像分类等，在第二机器学习模型处得到的第二输出，对应的就可以是语义分析结果、图像分类结果等，此处不做限定。
第二通信装置基于第二输出和机器学习模型系统的损失函数，计算得到第一梯度，第一梯度用于更新第一机器学习模型。第二通信装置还可以基于第二输出和机器学习模型系统的损失函数，计算得到第三梯度，第三梯度用于更新第二机器学习模型。
计算得到了第一梯度，第二通信装置就可以在第二传输资源上，向第一通信装置传输第一反馈数据，第一反馈数据用于指示第一梯度承载。
第一通信装置在第二传输资源上接收到第一反馈数据，就可以基于第一梯度更新第一机器学习模型。
计算得到了第三梯度，第二通信装置就可以基于第三梯度更新第二机器学习模型。
上面描述了单资源TU中前向和反向传输资源在机器学习模型系统训练过程中的作用，当TU为多资源TU时，各回合的前向与反向传输资源的作用与上述第一传输资源和第二传输资源的作用类似，此处不再赘述。
508、第二通信装置进行性能评估，向第一通信装置传输训练结束指示，相应的，第一通信装置接收训练结束指示。
基于步骤507中机器学习模型系统的训练过程，第二通信装置可以对训练后的机器学习模型系统进行性能评估。具体的，第二通信装置可以基于上一回合训练过程中前向推理 结果的传输，评估该次传输的误码率，当误码率低于某一阈值时，确定机器学习模型系统已经能满足使用需求，可以结束对机器学习模型系统的训练。
在本申请实施例中，除了评估误码率，第二通信装置还可以通过其他性能确定是否结束对机器学习模型系统的训练，例如评估吞吐，时延等性能，此处不做限定。
确定可以结束对机器学习模型系统的训练，第二通信装置就可以向第一通信装置传输训练结束指示，训练结束指示用于通知对第一机器学习模型的训练。
可选的，若第二通信装置进行性能评估后，确定性能不满足预设条件，需要继续训练机器学习模型系统，继续进行训练过程，直到第二通信装置确定性能满足预设条件，确定可以结束训练机器学习模型系统，触发步骤208中向第一通信装置传输训练结束指示的动作。其中，训练过程包括：确定TU、发送指示TU的TCI、根据TCI训练机器学习模型系统、第二通信装置进行评估。
现有的控制信令只能指示一个方向的传输资源。对于第一通信装置或第二通信装置来说，为了实现对机器学习模型系统的一次训练，需要获取两个控制信令，以确定用于传输前向推理结果和反向梯度的两个方向的传输资源。在本申请实施例中，通过TCI指示前向和反向两个方向的传输资源，可以一次性确定两个方向的传输资源信息。将TCI作为控制信令发出，控制信令承载的传输资源信息的信息量加倍，通信系统中需要传输的控制信令的数量减半，因此可以减小通信系统中的控制信令开销。
在本申请实施例中，TU中包括的前向传输资源和反向传输资源，也可以不用于机器学习系统的训练，只要传输的数据是成对出现的，需要前向和反向两个方向的传输资源来传输，就可以用本申请实施例所提供的TCI来指示两个方向的传输资源，此处不做限定。
图5所示实施例说明了由第一通信装置确定TU的方案，接下来通过图6所示的实施例说明，由第二通信装置确定TU的方案。
前向推理结果和反向梯度的传输，可以经过编码、调制等处理作为数据传输，也可以采用模拟方式直接在空口上传输。在图6所示的实施例中将会介绍模拟传输是如何实现的。
2、由第二通信装置来确定TU。
请参阅图6，图6为本申请实施例提供的数据传输方法的流程示意图，如图6所示，该方法包括：
601、确定帧结构配置。
在本申请实施例中，帧结构指的是TU的结构，在帧结构配置的过程中，可以确定TU采用固定位置TU还是采用动态位置TU。除了确定TU类型，帧结构配置还可以确定TU相关的其他内容，例如是单资源TU还是多资源TU等，此处不做限定。
可选的，步骤601可以在通信系统部署的时候就通过参数配置实现。
可选的,帧结构配置可以由标准给定，除了标准给定，也可以通过其他方式确定，例如由中心控制设备确定等，此处不做限定。
在图5所示实施例中的步骤501之前，也可以包括步骤601，此处不做限定。
值得注意的是，步骤601为可选步骤，在机器学习模型系统的训练过程中，并不是每轮训练之前都要执行步骤601。
可选的，步骤601可以在通信系统建立或更新的时候执行，此处不做限定。
602、第一通信装置和第二通信装置进行信道互易性测量。
可选的，机器学习模型的训练可以通过模拟传输的方式实现。模拟传输的方式具体为，前向推理结果和反向梯度的传输，都不进行数字化处理，而是直接以模拟信号的形式传输。由于在第一通信装置与第二通信装置之间，既有前向的数据传输，又有反向的数据传输，前向和反向是相反的两个方向，因此需要进行信道互易性测量。确保可以在具有互易性的信道上，实现数据在两个通信装置之间的前向与反向传输。
在本申请实施例中，前向推理结果表示第一机器学习模型的输出。
可选的，进行了信道互易性测量，则可以确定具有互易性的前向和反向的传输资源。例如，如图7a所示，确定在T长时间，W宽频段内，信道具有互易性。则可以确定通信系统的系统参数numerology(例如符号长度和子载波间隔配置)，使得TU的时域长度要小于或等于T，频域宽度要小于或等于W。
如图7a所示，在模拟传输的方法中，TU可以如A图或B图所示是单资源TU，也可以如C图所示是多资源TU，此处不做限定。
如图7b中的A图所示，若当前的通信环境较稳定，则具有互易性的时域长度较长，通信系统可以选择较窄的子载波间隔和较长的符号长度，如图7b中的B图或C图所示，当前的通信环境越不稳定，具有互易性的时域长度越短，通信系统使用的子载波间隔就会越宽、符号长度越短。
在图5所示的实施例中，若采用的是模拟传输的方式进行机器学习模型的训练，则在步骤501之前也要执行步骤602，此处不做限定。
值得注意的是，步骤602为可选步骤，在机器学习模型系统的训练过程中，并不是每轮训练之前都要执行步骤602。
可选的，步骤602可以在通信系统建立或更新的时候执行，此处不做限定。
603、第一通信装置初始化第一机器学习模型，第二通信装置初始化第二机器学习模型。
值得注意的是，步骤603为可选步骤，在机器学习模型系统的训练过程中，并不是每轮训练之前都要执行步骤603。
可选的，步骤602可以在通信系统建立或更新的时候执行，此处不做限定。
可选的，本实施例中，步骤601至603可以都执行，也可以都不执行，或者可以执行其中的一部分，例如执行步骤602，不执行步骤601和603；或者执行步骤601和602，不执行步骤603等，此处不做限定。
604、第一通信装置通过机器学习模型系统向第二通信装置传输测试数据，相应的，第二通信装置接收测试数据。
605、第二通信装置进行性能评估，确定需要训练机器学习模型系统。
步骤603至步骤605参见图5所示实施例的步骤501至步骤503，此处不再赘述。
606、第二通信装置向第一通信装置传输训练指示，相应的，第一通信装置接收训练指示。
可选的，第二通信装置确定了需要训练机器学习模型系统，就可以向第一通信装置传 输训练指示，训练指示用于指示对第一机器学习模型的训练。
607、第二通信装置确定TU。
在步骤605中第二通信装置确定了需要训练机器学习模型系统，就可以确定传输资源单元TU，对于TU的详细描述，参见图5所示实施例的步骤505，此处不再赘述。
在步骤606存在的情况下，步骤607可以在步骤606之前或之后执行，只要在步骤605之后执行即可，此处不做限定。
608、第二通信装置向第一通信装置传输TCI，相应的，第一通信装置接收TCI。
确定了TU，第二通信装置就可以确定TCI，向第一通信装置传输该TCI。TCI为TU的指示信息，用于指示TU中所包括的所有传输资源的位置信息。
在本申请实施例中，TCI不一定由第二通信装置发出。在第二通信装置确定TU后，也可以将TU告知给其它设备(如中继设备)，由其它设备来向第一通信装置传输TCI，此处不做限定。
在本申请实施例中，步骤606为可选步骤，在步骤608中第二通信装置向第一通信装置传输了TCI，则第一通信装置可以基于TCI获知需要对第一机器学习模型进行训练。因此第二通信装置也可以不用向第一通信装置传输训练指示。
在本申请实施例中，在第二通信装置向第一通信装置传输TCI(步骤608)之前，第二通信装置向第一通信装置传输训练指示(步骤606)的优势在于：第一通信装置接收到训练指示，就可以根据该训练指示开启对第一机器学习模型的训练过程，例如可以开始计算第一输出；接收到TCI，就可以马上在TCI指示的前向传输资源上将第一输出传输给第二通信装置，相较于在接收到TCI之后再开始计算第一输出，提升了第一机器学习模型以及整个机器学习模型系统的训练效率。
609、第一通信装置和第二通信装置根据TCI中指示的传输资源，对机器学习模型系统进行训练。
610、第二通信装置进行性能评估，向第一通信装置传输训练结束指示，相应的，第一通信装置接收训练结束指示。
步骤609和步骤610参见图5所示实施例中的步骤507和步骤508，此处不再赘述。
在本申请实施例中，若第一通信装置是网络设备，第二通信装置可以是终端设备，也可以是网络设备；若第一通信装置是终端设备，第二通信装置可以是网络设备，也可以是终端设备，此处不做限定。
图5和图6的实施例说明了由第一通信装置或第二通信装置确定TU的方案，接下来通过图8所示的实施例，说明由中心控制设备确定TU的方案。
3、由中心控制设备来确定TU。
在通信系统中包括中心控制设备，且通过中心控制设备来控制第一通信装置与第二通信装置之间的数据传输的情况下，可以由中心控制设备来确定TU。
可选的，第一通信装置和第二通信装置可以为终端设备，中心控制设备可以为网络设备。
可选的，本申请实施例可以应用于D2D场景中。示例地，在D2D场景中，第一通信装 置和第二通信装置都可以为终端设备。
具体的，中心控制设备可以是基站。在本申请实施例中，除了基站，中心控制设备还可以是其他通信装置，例如边缘设备等，只要具备对第一通信装置和第二通信装置之间的通信的控制能力即可，此处不做限定。
在一种可能的实现方式中，第一通信装置和第二通信装置中，可以一个是终端设备，一个是网络设备(例如基站)，中心控制设备用于控制网络设备与终端设备之间的通信。
请参阅图8，图8为本申请实施例提供的数据传输方法的流程示意图，如图8所示，该方法包括：
801、确定帧结构配置。
在本申请实施例中，帧结构指的是TU的结构，在帧结构配置的过程中，可以确定TU采用固定位置TU还是采用动态位置TU。除了确定TU类型，帧结构配置还可以确定TU相关的其他内容，例如是单资源TU还是多资源TU等，此处不做限定。
可选的，步骤801可以在通信系统部署的时候就通过参数配置实现。
可选的,帧结构配置可以由标准给定，除了标准给定，也可以通过其他方式确定，例如由中心控制设备确定等，此处不做限定。
802、第一通信装置和第二通信装置进行信道互易性测量。
803、第一通信装置初始化第一机器学习模型，第二通信装置初始化第二机器学习模型。
值得注意的是，步骤801至803中的任一步均为可选步骤，在机器学习模型系统的训练过程中，并不是每轮训练之前都要执行步骤801、802或803。
可选的，步骤801至803中的任一步，可以在通信系统建立或更新的时候执行，此处不做限定。
可选的，本实施例中，步骤801至803可以都执行，也可以都不执行，或者可以执行其中的一部分，例如执行步骤802，不执行步骤801和803；或者执行步骤801和802，不执行步骤803等，此处不做限定。
804、第一通信装置通过机器学习模型系统向第二通信装置传输业务数据。
805、第二通信装置进行性能评估，确定需要训练机器学习模型系统。
步骤802至步骤805参见图6所示实施例中步骤602至步骤605的说明，此处不再赘述。其中，步骤802和步骤803为可选步骤。
806、第二通信装置向中心控制设备传输训练指示，相应的，中心控制设备接收训练指示。
第二通信装置确定了需要训练机器学习模型系统，就可以向中心控制设备传输训练指示，训练指示用于指示对机器学习模型系统的训练。
807、第二通信装置向第一通信装置传输训练指示，相应的，第一通信装置接收训练指示。
可选的，第二通信装置确定了需要训练机器学习模型系统，就可以向第一通信装置传输训练指示，该训练指示用于指示对第一机器学习模型的训练。
808、中心控制设备确定TU。
中心控制设备接收到训练指示，就可以确定传输资源单元TU，对于TU的详细描述，参见图5所示实施例的步骤505，此处不再赘述。
809、中心控制设备向第一通信装置和第二通信装置传输TCI，相应的，第一通信装置和第二通信装置接收TCI。
确定了TU，中心控制设备就可以确定TCI，并向第一通信装置和第二通信装置传输该TCI。TCI为TU的指示信息，用于指示TU中所包括的所有传输资源的位置信息。
在本申请实施例中，TCI不一定由中心控制设备发出。在中心控制设备确定TU后，也可以将TU告知给其它设备(如中继设备)，由其它设备来向第一通信装置和第二通信装置传输TCI，此处不做限定。
在本申请实施例中，步骤807为可选步骤，在步骤809中，中心控制设备向第一通信装置传输了TCI，则第一通信装置可以基于TCI获知需要对第一机器学习模型进行训练。因此第二通信装置也可以不用向第一通信装置传输训练指示。向发送设备发送训练指示的有益效果，参见图6所示实施例中步骤609之前的说明，此处不再赘述。
810、第一通信装置和第二通信装置根据TCI中指示的传输资源，对机器学习模型系统进行训练。
步骤810参见图5所示实施例中的步骤507，此处不再赘述。
811、第二通信装置进行性能评估，向第一通信装置和中心控制设备传输训练结束指示，相应的，第一通信装置和中心控制设备接收训练结束指示。
对于第二通信装置进行性能评估的描述，参见图5所示实施例的步骤508。
第二通信装置确定了可以结束对机器学习模型系统的训练，就可以向第一通信装置和中心控制设备传输训练结束指示，训练结束指示用于通知对机器学习模型系统的训练。
基于统一技术方案，本申请还提供了相应的装置实施例。为了实现上述本申请实施例提供的方法中的各功能，第一通信装置、第二通信装置、第三通信装置均可以包括硬件结构和/或软件模块，以硬件结构、软件模块、或硬件结构加软件模块的形式来实现上述各功能。上述各功能中的某个功能以硬件结构、软件模块、还是硬件结构加软件模块的方式来执行，取决于技术方案的特定应用和设计约束条件。
下面描述本申请实施例提供的第一通信装置的结构。
1、第一通信装置的结构。
在本申请实施例中，第一通信装置(发送端通信装置)包含于通信系统，该通信系统还包括第二通信装置(接收端通信装置)，第一通信装置部署有第一机器学习模型，第二通信装置部署有第二机器学习模型，第一机器学习模型和第二机器学习模型用于实现第一通信装置与第二通信装置之间的通信。
如图9所示，第一通信装置900包括：处理单元901、收发单元902；
处理单元901用于：获取第一信息，第一信息同时承载有第一传输资源和第二传输资源的指示信息；其中，第一传输资源用于向第二通信装置传输第一数据，第一数据为第一机器学习模型的第一输出；第二传输资源用于接收来自第二通信装置的第一反馈数据，第一反馈数据用于指示第一梯度，第一梯度用于更新第一机器学习模型；
收发单元902用于：在第一传输资源上，向第二通信装置传输第一数据。
在一种可选的实施方式中，第一梯度为第二通信装置根据第二输出计算所得，第二输出为第二数据输入第二机器学习模型所得的输出，第二数据为第一数据经过信道传输后所得的数据。
在一种可选的实施方式中，第一信息包括传输资源单元的指示信息，第一传输资源和第二传输资源包含于传输资源单元。
在一种可选的实施方式中，第一信息同时承载有第三传输资源和第四传输资源的指示信息；其中，第三传输资源用于向第二通信装置传输第三数据，第三数据为第一机器学习模型的第三输出；第四传输资源用于接收来自第二通信装置的第二反馈数据，第二反馈数据用于指示第二梯度，第二梯度用于更新第一机器学习模型；
收发单元902用于：在第三传输资源上，向第二通信装置传输第三数据。
在一种可选的实施方式中，第一传输资源和第二传输资源位于：时域、频域、空域、码域、功率域中至少一项的固定传输资源位置上；或者，用于传输业务数据的时域、频域、空域、码域、功率域中至少一项的资源块中。
在一种可选的实施方式中，收发单元902用于：接收来自第二通信装置的第一信息；或者，接收来自第三通信装置的第一信息，第三通信装置用于控制第一通信装置与第二通信装置之间的通信；
处理单元901具体用于：从收发单元902获取第一信息。
在一种可选的实施方式中，收发单元902还用于：从处理单元901获取第一信息；向第二通信装置传输第一信息。
在一种可选的实施方式中，收发单元902还用于：接收来自第二通信装置的训练指示，训练指示用于指示对第一机器学习模型的训练。
上面介绍了本申请实施例提供的第一通信装置，接下来结合图10说明本申请实施例提供的第二通信装置的结构。
2、第二通信装置的结构。
在本申请实施例中，第二通信装置(接收端通信装置)包含于通信系统，该通信系统还包括第一通信装置(发送端通信装置)，第一通信装置部署有第一机器学习模型，第二通信装置部署有第二机器学习模型，第一机器学习模型和第二机器学习模型用于实现第一通信装置与第二通信装置之间的通信。
如图10所示，第二通信装置1000包括：处理单元1000、收发单元1002；
处理单元1001用于：获取第一信息，第一信息同时承载有第一传输资源和第二传输资源的指示信息；其中，第一传输资源用于接收来自第一通信装置的第二数据，第二数据为第一数据经过信道传输后所得的数据，第一数据为第一机器学习模型的第一输出；第二传输资源用于向第一通信装置传输第一反馈数据，第一反馈数据用于指示第一梯度，第一梯度用于更新第一机器学习模型；
收发单元1002用于：在第二传输资源上，向第一通信装置传输第一反馈数据。
在一种可选的实施方式中，第一梯度为第二通信装置根据第二输出计算所得，第二输 出为第二数据输入第二机器学习模型所得的输出。
在一种可选的实施方式中，第一信息包括传输资源单元的指示信息，第一传输资源和第二传输资源包含于传输资源单元。
在一种可选的实施方式中，第一信息同时承载有第三传输资源和第四传输资源的指示信息；其中，第三传输资源用于接收来自第一通信装置的第四数据，第四数据为第三数据经过信道传输后所得的数据，第三数据为第一机器学习模型的第三输出；第四传输资源用于向第一通信装置传输第二反馈数据，第二反馈数据用于指示第二梯度，第二梯度用于更新第一机器学习模型；收发单元1002用于：在第四传输资源上，向第一通信装置传输第二反馈数据。
在一种可选的实施方式中，第一传输资源和第二传输资源位于：时域、频域、空域、码域、功率域中至少一项的固定传输资源位置上；或者，用于传输业务数据的时域、频域、空域、码域、功率域中至少一项的资源块中。
在一种可选的实施方式中，收发单元1002用于：接收来自第一通信装置的第一信息；或者，接收来自第三通信装置的第一信息，第三通信装置用于控制第一通信装置与第二通信装置之间的通信；
处理单元1001具体用于：从收发单元1002获取第一信息。
在一种可选的实施方式中，收发单元1002还用于：从处理单元1001获取第一信息；向第一通信装置传输第一信息。
在一种可选的实施方式中，处理单元1001还用于：生成训练指示，训练指示用于指示对第一机器学习模型的训练；
收发单元1002还用于：向第一通信装置发送该训练指示。
在一种可选的实施方式中，处理单元1001还用于：生成训练指示，训练指示用于指示对机器学习模型网络的训练；
收发单元1002还用于：向第三通信装置发送该训练指示。
上面介绍了本申请实施例提供的第一通信装置和第二通信装置，接下来结合图11说明本申请实施例提供的中心控制设备的结构。
3、中心控制设备的结构。
在本申请实施例中，中心控制设备也称为第三通信装置，第三通信装置包含于通信系统，该通信系统还包括第一通信装置(发送端通信装置)和第二通信装置(接收端通信装置)，第三通信装置用于控制第一通信装置和第二通信装置的之间的通信，第一通信装置部署有第一机器学习模型，第二通信装置部署有第二机器学习模型，第一机器学习模型和第二机器学习模型用于实现第一通信装置与第二通信装置之间的通信。
如图11所示，第三通信装置1100包括：处理器1101、收发单元1102；
处理单元1101用于：获取第一信息，第一信息同时承载有第一传输资源和第二传输资源的指示信息；其中，第一传输资源用于第一通信装置向第二通信装置传输第一数据，第一数据为第一机器学习模型的第一输出；第二传输资源用于第二通信装置向第一通信装置传输第一反馈数据，第一反馈数据用于指示第一梯度，第一梯度用于更新第一机器学习模 型；
收发单元1102用于：向第一通信装置和/或第二通信装置传输第一信息。
在一种可选的实施方式中，第一信息包括传输资源单元的指示信息，第一传输资源和第二传输资源包含于传输资源单元。
在一种可选的实施方式中，第一信息同时承载有第三传输资源和第四传输资源的指示信息；其中，第三传输资源用于第一通信装置向第二通信装置传输第三数据，第三数据为第一机器学习模型的第三输出；第四传输资源用于第二通信装置向第一通信装置传输第二反馈数据，第二反馈数据用于指示第二梯度，第二梯度用于更新第一机器学习模型。
在一种可选的实施方式中，第一传输资源和第二传输资源位于：时域、频域、空域、码域、功率域中至少一项的固定传输资源位置上；或者，用于传输业务数据的时域、频域、空域、码域、功率域中至少一项的资源块中。
在一种可选的实施方式中，收发单元1102用于，接收来自第二通信装置的训练指示，训练指示用于指示对机器学习模型系统的训练。
在一种可选的实施方式中，收发单元1102还用于：接收来自第二通信装置的训练指示，训练指示用于指示对机器学习模型网络的训练。
所属领域的技术人员可以清楚地了解到，为描述的方便和简洁，上述描述的系统，装置和单元的具体工作过程，可以参考前述方法实施例中的对应过程，在此不再赘述。
可选的，除了上述结构，本申请实施例的通信装置还可以是如下所示的结构：
参见图12，本申请实施例还提供了一种通信装置1200，用于实现上述方法中终端、网络设备的功能，即第一通信装置、第二通信装置或第三通信装置的功能。该通信装置可以是第一通信装置、第二通信装置或第三通信装置，也可以是第一通信装置、第二通信装置或第三通信装置中的装置，或者是能够和第一通信装置、第二通信装置或第三通信装置匹配使用的装置。其中，该通信装置1200可以为芯片系统。本申请实施例中，芯片系统可以由芯片构成，也可以包含芯片和其他分立器件。通信装置1200包括至少一个处理器1210，用于实现本申请实施例提供的方法中第一通信装置、第二通信装置或第三通信装置的功能。通信装置1200还可以包括通信接口1220。在本申请实施例中，通信接口1220可以是收发器、电路、总线、模块或其它类型的通信接口，用于通过传输介质和其它设备进行通信。例如，通信接口1220用于通信装置1200中的装置可以和其它设备进行通信。
处理器1210可以执行第一通信装置900中处理单元910所执行的功能；通信接口1220可以用于执行通信装置900中收发单元920所执行的功能。
当通信装置1200用于执行第一通信装置所执行的操作时，处理器1210用于获取第一信息，第一信息同时承载有第一传输资源和第二传输资源的指示信息；其中，第一传输资源用于向第二通信装置传输第一数据，第一数据为第一机器学习模型的第一输出；第二传输资源用于接收来自第二通信装置的第一反馈数据，第一反馈数据用于指示第一梯度，第一梯度用于更新第一机器学习模型；通信接口1220用于在第一传输资源上，向第二通信装置传输第一数据。
当通信装置1200用于执行第二通信装置所执行的操作时，处理器1210用于：获取第 一信息，第一信息同时承载有第一传输资源和第二传输资源的指示信息；其中，第一传输资源用于接收来自第一通信装置的第二数据，第二数据为第一数据经过信道传输后所得的数据，第一数据为第一机器学习模型的第一输出；第二传输资源用于向第一通信装置传输第一反馈数据，第一反馈数据用于指示第一梯度，第一梯度用于更新第一机器学习模型；通信接口1220用于在第二传输资源上，向第一通信装置传输第一反馈数据。
当通信装置1200用于执行第三通信装置所执行的操作时，处理器1210用于：获取第一信息，第一信息同时承载有第一传输资源和第二传输资源的指示信息；其中，第一传输资源用于第一通信装置向第二通信装置传输第一数据，第一数据为第一机器学习模型的第一输出；第二传输资源用于第二通信装置向第一通信装置传输第一反馈数据，第一反馈数据用于指示第一梯度，第一梯度用于更新第一机器学习模型；通信接口1220用于向第一通信装置和/或第二通信装置传输第一信息。
通信接口1220还用于执行上述方法实施例中第一通信装置、第二通信装置或第三通信装置执行的其它接收或发送的步骤或操作。处理器1210还可以用于执行上述方法实施例第一通信装置、第二通信装置或第三通信装置执行的除收发之外的其它对应的步骤或操作，在此不再一一赘述。
通信装置1200还可以包括至少一个存储器1230，用于存储程序指令和/或数据。存储器1230和处理器1210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1210可能执行存储器1230中存储的程序指令。在一种可能的实现中，至少一个存储器中的至少一个可以与处理器集成在一起。在另一种可能的实现中，存储器1230位于该通信装置1200之外。
本申请实施例中不限定上述通信接口1220、处理器1210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1210以及通信接口1220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。该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。
本申请实施例中，处理器1210可以是一个或多个中央处理器(Central Processing Unit，CPU)，在处理器1210是一个CPU的情况下，该CPU可以是单核CPU，也可以是多核CPU。处理器1210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，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。
本申请实施例中，存储器1230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期望的程序代码并能够由计算机存取的任何其他介质，但不限于此。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。
可选的，除了上述结构，本申请实施例的通信装置还可以是如下所示的结构：
参加图13，本申请实施例还提供了一种装置1300，可用于实现上述方法中第一通信装置、第二通信装置或第三通信装置的功能，该装置1300可以是通信装置或者通信装置中的芯片。该通信装置包括：
至少一个输入输出接口1310和逻辑电路1320。输入输出接口1310可以是输入输出电路，也可以称为通信接口。逻辑电路1320可以是信号处理器、芯片，或其他可以实现本申请方法的集成电路。
其中，至少一个输入输出接口1310用于信号或数据的输入或输出。举例来说，当该装置为第一通信装置或者用于第一通信装置时，输入输出接口1310用于与第二通信装置通信。举例来说，当该装置为第二通信装置或者用于第二通信装置时，输入输出接口1310用于输出第一反馈数据。
其中，逻辑电路1320用于执行本申请实施例提供的任意一种方法的部分或全部步骤。逻辑电路可以实现上述第一通信装置900中的处理单元901、第二通信装置1000中的处理单元1001或第三通信装置1100中的处理单元1101所实现的功能。举例来说，当该装置为第一通信装置或者用于第一通信装置时，用于执行上述方法实施例中各种可能的实现方式中第一通信装置(发送端通信装置)执行的步骤，例如逻辑电路1320用于获取第一输出。当该装置为第二通信装置或者用于第二通信装置时，用于执行上述方法实施例中各种可能的实现方法中第二通信装置(接收端通信装置)执行的步骤，例如逻辑电路1320用于确定第一梯度。
当上述通信装置为应用于第一通信装置的芯片时，该终端芯片实现上述方法实施例中第一通信装置的功能。该终端芯片从终端中的其它模块(如射频模块或天线)接收信息，该信息是第二通信装置或第三通信装置发送给第一通信装置的；或者，该终端芯片向第二通信装置中的其它模块(如射频模块或天线)发送信息，该信息是第一通信装置发送给第二通信装置或第三通信装置的。
当上述通信装置为应用于第二通信装置的芯片时，该第二通信装置芯片实现上述方法实施例中第二通信装置的功能。该第二通信装置芯片从第二通信装置中的其它模块(如射频模块或天线)接收信息，该信息是第一通信装置或第三通信装置发送给第二通信装置的；或者，该第二通信装置芯片向第二通信装置中的其它模块(如射频模块或天线)发送信息，该信息是第二通信装置发送给第一通信装置或第三通信装置的。
当上述通信装置为应用于第三通信装置的芯片时，该第三通信装置芯片实现上述方法实施例中第三通信装置的功能。该第三通信装置芯片从第三通信装置中的其它模块(如射频模块或天线)接收信息，该信息是第一通信装置或第二通信装置发送给第三通信装置的；或者，该第三通信装置芯片向第三通信装置中的其它模块(如射频模块或天线)发送信息，该信息是第三通信装置发送给第一通信装置或第二通信装置的。
基于与上述方法实施例相同构思，本申请实施例还提供一种计算机可读存储介质，所述计算机可读存储介质存储有计算机程序，所述计算机程序被硬件(例如处理器等)执行，以实现本申请实施例中由任意装置执行的任意一种方法的部分或全部步骤。
基于与上述方法实施例相同构思，本申请实施例还提供了一种包括程序指令的计算机程序产品，当所述计算机程序产品在计算机上运行时，使得所述这个计算机执行以上各方面的任意一种方法的部分或者全部步骤。
基于与上述方法实施例相同构思，本申请还提供一种芯片或芯片系统，该芯片可包括处理器。该芯片还可包括存储器(或存储模块)和/或收发器(或通信模块)，或者，该芯片与存储器(或存储模块)和/或收发器(或通信模块)耦合，其中，收发器(或通信模块)可用于支持该芯片进行有线和/或无线通信，存储器(或存储模块)可用于存储程序，该处理器调用该程序可用于实现上述方法实施例、方法实施例的任意一种可能的实现方式中由终端或者网络设备执行的操作。该芯片系统可包括以上芯片，也可以包含上述芯片和其他分立器件，如存储器(或存储模块)和/或收发器(或通信模块)。
基于与上述方法实施例相同构思，本申请还提供一种通信系统，该通信系统可包括以上第一通信装置、第二通信装置和第三通信装置中的至少一项。该通信系统可用于实现上述方法实施例、方法实施例的任意一种可能的实现方式中由终端或者网络设备执行的操作。示例性的，该通信系统可具有如图1a或图1b所示结构。
在本申请所提供的几个实施例中，应该理解到，所揭露的系统，装置和方法，可以通过其它的方式实现。例如，以上所描述的装置实施例仅仅是示意性的，例如，所述单元的划分，仅仅为一种逻辑功能划分，实际实现时可以有另外的划分方式，例如多个单元或组件可以结合或者可以集成到另一个系统，或一些特征可以忽略，或不执行。另一点，所显示或讨论的相互之间的耦合或直接耦合或通信连接可以是通过一些接口，装置或单元的间接耦合或通信连接，可以是电性，机械或其它的形式。
所述作为分离部件说明的单元可以是或者也可以不是物理上分开的，作为单元显示的部件可以是或者也可以不是物理单元，即可以位于一个地方，或者也可以分布到多个网络单元上。可以根据实际的需要选择其中的部分或者全部单元来实现本实施例方案的目的。
另外，在本申请各个实施例中的各功能单元可以集成在一个处理单元中，也可以是各个单元单独物理存在，也可以两个或两个以上单元集成在一个单元中。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，本申请的技术方案本质上或者说对现有技术做出贡献的部分或者该技术方案的全部或部分可以以软件产品的形式体现出来，该计算机软件产品存储在一个存储介质中，包括若干指令用以使得一台计算机设备(可以是个人计算机，服务器，或者网络设备等)执行本申请各个实施例所述方法的全部或部分步骤。而前述的存储介质包括：U盘、移动硬盘、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。

Claims (68)

  1. 一种数据传输方法，其特征在于，所述方法应用于通信系统中的第一通信装置，所述通信系统还包括第二通信装置，所述第一通信装置部署有第一机器学习模型，所述第二通信装置部署有第二机器学习模型，所述第一机器学习模型和所述第二机器学习模型用于实现所述第一通信装置与所述第二通信装置之间的通信，所述方法包括：
    所述第一通信装置获取第一信息，所述第一信息包括第一传输资源和第二传输资源的指示信息；其中，所述第一传输资源用于所述第一通信装置向所述第二通信装置发送第一数据，所述第一数据为所述第一机器学习模型的第一输出；所述第二传输资源用于所述第一通信装置接收来自所述第二通信装置的第一反馈数据，所述第一反馈数据用于指示第一梯度，所述第一梯度用于更新所述第一机器学习模型；
    所述第一通信装置在所述第一传输资源上，向所述第二通信装置发送所述第一数据。
  2. 根据权利要求1所述的方法，其特征在于，所述第一梯度为所述第二通信装置根据第二输出计算所得，所述第二输出为第二数据输入所述第二机器学习模型所得的输出，所述第二数据为所述第一数据经过信道传输后所得的数据。
  3. 根据权利要求1或2所述的方法，其特征在于，所述第一信息包括传输资源单元的指示信息，所述第一传输资源和所述第二传输资源包含于所述传输资源单元。
  4. 根据权利要求3所述的方法，其特征在于，所述传输资源单元的指示信息包括训练控制信息TCI。
  5. 根据权利要求1至4中任一项所述的方法，其特征在于，所述第一信息包括第三传输资源和第四传输资源的指示信息；其中，所述第三传输资源用于所述第一通信装置向所述第二通信装置发送第三数据，所述第三数据为所述第一机器学习模型的第三输出；所述第四传输资源用于所述第一通信装置接收来自所述第二通信装置的第二反馈数据，所述第二反馈数据用于指示第二梯度，所述第二梯度用于更新所述第一机器学习模型；
    所述第一通信装置在所述第三传输资源上，向所述第二通信装置发送所述第三数据。
  6. 根据权利要求5所述的方法，其特征在于，所述第二梯度为所述第二通信装置根据第四输出计算所得，所述第四输出为第四数据输入所述第二机器学习模型所得的输出，所述第四数据为所述第三数据经过信道传输后所得的数据。
  7. 根据权利要求5或6所述的方法，其特征在于，所述第一传输资源为第一回合的前向传输资源，所述第二传输资源为所述第一回合的反向传输资源；所述第三传输资源为第二回合的前向传输资源，所述第四传输资源为所述第二回合的反向传输资源。
  8. 根据权利要求1至7中任一项所述的方法，其特征在于，所述第一传输资源和所述第二传输资源位于：
    时域、频域、空域、码域、功率域中至少一项的固定传输资源位置上；或者，
    用于传输业务数据的时域、频域、空域、码域、功率域中至少一项的资源位置上。
  9. 根据权利要求1至8中任一项所述的方法，其特征在于，所述第一通信装置获取第一信息，包括：
    所述第一通信装置接收来自所述第二通信装置的所述第一信息；或者，
    所述第一通信装置接收来自第三通信装置的所述第一信息，所述第三通信装置用于控制所述第一通信装置与所述第二通信装置之间的通信。
  10. 根据权利要求1至8中任一项所述的方法，其特征在于，在所述第一通信装置获取第一信息之后，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所述第一信息。
  11. 根据权利要求1至10中任一项所述的方法，其特征在于，所述方法还包括：
    所述第一通信装置将训练数据输入所述第一机器学习模型，得到所述第一输出；
    在所述第一通信装置在所述第一传输资源上，向所述第二通信装置发送所述第一数据之后，所述方法还包括；
    所述第一通信装置在所述第二传输资源上，接收来自所述第二通信装置的所述第一反馈数据，所述第一反馈数据用于指示所述第一梯度；
    所述第一通信装置根据所述第一梯度更新所述第一机器学习模型。
  12. 根据权利要求1至11中任一项所述的方法，其特征在于，在所述第一通信装置获取第一信息之前，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训练指示，所述训练指示用于指示对所述第一机器学习模型的训练。
  13. 一种数据传输方法，其特征在于，所述方法应用于通信系统中的第二通信装置，所述通信系统还包括第一通信装置，所述第一通信装置部署有第一机器学习模型，所述第二通信装置部署有第二机器学习模型，所述第一机器学习模型和所述第二机器学习模型用于实现所述第一通信装置与所述第二通信装置之间的通信，所述方法包括：
    所述第二通信装置获取第一信息，所述第一信息包括第一传输资源和第二传输资源的指示信息；其中，所述第一传输资源用于所述第二通信装置接收来自所述第一通信装置的第二数据，所述第二数据为第一数据经过信道传输后所得的数据，所述第一数据为所述第一机器学习模型的第一输出；所述第二传输资源用于所述第二通信装置向所述第一通信装置发送第一反馈数据，所述第一反馈数据用于指示第一梯度，所述第一梯度用于更新所述第一机器学习模型；
    所述第二通信装置在所述第二传输资源上，向所述第一通信装置发送所述第一反馈数据。
  14. 根据权利要求13所述的方法，其特征在于，所述方法还包括：
    所述第二通信装置将所述第二数据输入所述第二机器学习模型，并根据输出计算得到第三梯度，所述第三梯度用于更新所述第二机器学习模型。
  15. 根据权利要求13或14所述的方法，其特征在于，所述第一梯度为所述第二通信装置根据第二输出计算所得，所述第二输出为所述第二数据输入所述第二机器学习模型所得的输出。
  16. 根据权利要求13至15中任一项所述的方法，其特征在于，所述第一信息包括传输资源单元的指示信息，所述第一传输资源和所述第二传输资源包含于所述传输资源单元。
  17. 根据权利要求16所述的方法，其特征在于，所述传输资源单元的指示信息包括传 输资源单元TCI。
  18. 根据权利要求13至17中任一项所述的方法，其特征在于，所述第一信息包括第三传输资源和第四传输资源的指示信息；其中，所述第三传输资源用于所述第二通信装置接收来自所述第一通信装置的第四数据，所述第四数据为第三数据经过信道传输后所得的数据，所述第三数据为所述第一机器学习模型的第三输出；所述第四传输资源用于所述第二通信装置向所述第一通信装置发送第二反馈数据，所述第二反馈数据用于指示第二梯度，所述第二梯度用于更新所述第一机器学习模型；
    所述第二通信装置在所述第四传输资源上，向所述第一通信装置发送所述第二反馈数据。
  19. 根据权利要求18所述的方法，其特征在于，所述第二梯度为所述第二通信装置根据第四输出计算所得，所述第四输出为所述第四数据输入所述第二机器学习模型所得的输出。
  20. 根据权利要求18或19所述的方法，其特征在于，所述方法还包括：
    所述第二通信设备将所述第四数据输入所述第二机器学习模型，并根据输出计算得到第四梯度，所述第四梯度用于更新所述第二机器学习模型。
  21. 根据权利要求18至20中任一项所述的方法，其特征在于，所述第一传输资源为第一回合的前向传输资源，所述第二传输资源为所述第一回合的反向传输资源；所述第三传输资源为第二回合的前向传输资源，所述第四传输资源为所述第二回合的反向传输资源。
  22. 根据权利要求13至21中任一项所述的方法，其特征在于，所述第一传输资源和所述第二传输资源位于：
    时域、频域、空域、码域、功率域中至少一项的固定传输资源位置上；或者，
    用于传输业务数据的时域、频域、空域、码域、功率域中至少一项的资源位置上。
  23. 根据权利要求13至22中任一项所述的方法，其特征在于，所述第二通信装置获取第一信息，包括：
    所述第二通信装置接收来自所述第一通信装置的所述第一信息；或者，
    所述第二通信装置接收来自第三通信装置的所述第一信息，所述第三通信装置用于控制所述第一通信装置与所述第二通信装置之间的通信。
  24. 根据权利要求13至22中任一项所述的方法，其特征在于，在所述第二通信装置获取第一信息之后，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所述第一信息。
  25. 根据权利要求13至24中任一项所述的方法，其特征在于，所述方法还包括：
    所述第二通信装置将所述第二数据输入所述第二机器学习模型，得到所述第二输出；
    所述第二通信装置根据所述第二输出计算得到所述第一梯度。
  26. 根据权利要求13至25中任一项所述的方法，其特征在于，在所述第二通信装置获取第一信息之前，所述方法还包括：
    所述第二通信装置向所述第一通信装置传输训练指示，所述训练指示用于指示对所述第一机器学习模型的训练。
  27. 一种数据传输方法，其特征在于，所述方法应用于通信系统中的第三通信装置，所述通信系统还包括第一通信装置和第二通信装置，所述第三通信装置用于控制所述第一通信装置和所述第二通信装置的之间的通信，所述第一通信装置部署有第一机器学习模型，所述第二通信装置部署有第二机器学习模型，所述第一机器学习模型和所述第二机器学习模型用于实现所述第一通信装置与所述第二通信装置之间的通信，所述方法包括：
    所述第三通信装置获取第一信息，所述第一信息包括第一传输资源和第二传输资源的指示信息；其中，所述第一传输资源用于所述第一通信装置向所述第二通信装置发送第一数据，所述第一数据为所述第一机器学习模型的第一输出；所述第二传输资源用于所述第二通信装置向所述第一通信装置发送第一反馈数据，所述第一反馈数据用于指示第一梯度，所述第一梯度用于更新所述第一机器学习模型；
    所述第三通信装置向所述第一通信装置和/或所述第二通信装置发送所述第一信息。
  28. 根据权利要求27所述的方法，其特征在于，所述第一信息包括传输资源单元的指示信息，所述第一传输资源和所述第二传输资源包含于所述传输资源单元。
  29. 根据权利要求27或28所述的方法，其特征在于，所述第一信息包括第三传输资源和第四传输资源的指示信息；其中，所述第三传输资源用于所述第一通信装置向所述第二通信装置发送第三数据，所述第三数据为所述第一机器学习模型的第三输出；所述第四传输资源用于所述第二通信装置向所述第一通信装置发送第二反馈数据，所述第二反馈数据用于指示第二梯度，所述第二梯度用于更新所述第一机器学习模型。
  30. 根据权利要求27至29中任一项所述的方法，其特征在于，所述第一传输资源和所述第二传输资源位于：
    时域、频域、空域、码域、功率域中至少一项的固定传输资源位置上；或者，
    用于传输业务数据的时域、频域、空域、码域、功率域中至少一项的资源位置上。
  31. 一种通信装置，其特征在于，所述通信装置为通信系统中的第一通信装置，所述通信系统还包括第二通信装置，所述第一通信装置部署有第一机器学习模型，所述第二通信装置部署有第二机器学习模型，所述第一机器学习模型和所述第二机器学习模型用于实现所述第一通信装置与所述第二通信装置之间的通信，所述第一通信装置包括：
    处理单元、收发单元；
    所述处理单元用于：获取第一信息，所述第一信息包括第一传输资源和第二传输资源的指示信息；其中，所述第一传输资源用于向所述第二通信装置发送第一数据，所述第一数据为所述第一机器学习模型的第一输出；所述第二传输资源用于接收来自所述第二通信装置的第一反馈数据，所述第一反馈数据用于指示第一梯度，所述第一梯度用于更新所述第一机器学习模型；
    所述收发单元用于：在所述第一传输资源上，向所述第二通信装置发送所述第一数据。
  32. 根据权利要求31所述的装置，其特征在于，所述第一梯度为所述第二通信装置根据第二输出计算所得，所述第二输出为第二数据输入所述第二机器学习模型所得的输出，所述第二数据为所述第一数据经过信道传输后所得的数据。
  33. 根据权利要求31或32所述的装置，其特征在于，所述第一信息包括传输资源单元 的指示信息，所述第一传输资源和所述第二传输资源包含于所述传输资源单元。
  34. 根据权利要求33所述的装置，其特征在于，所述传输资源单元的指示信息包括训练控制信息TCI。
  35. 根据权利要求31至34中任一项所述的装置，其特征在于，所述第一信息包括第三传输资源和第四传输资源的指示信息；其中，所述第三传输资源用于所述第一通信装置向所述第二通信装置发送第三数据，所述第三数据为所述第一机器学习模型的第三输出；所述第四传输资源用于所述第一通信装置接收来自所述第二通信装置的第二反馈数据，所述第二反馈数据用于指示第二梯度，所述第二梯度用于更新所述第一机器学习模型；
    所述收发单元用于：在所述第三传输资源上，向所述第二通信装置发送所述第三数据。
  36. 根据权利要求35所述的装置，其特征在于，所述第二梯度为所述第二通信装置根据第四输出计算所得，所述第四输出为第四数据输入所述第二机器学习模型所得的输出，所述第四数据为所述第三数据经过信道传输后所得的数据。
  37. 根据权利要求35或36所述的装置，其特征在于，所述第一传输资源为第一回合的前向传输资源，所述第二传输资源为所述第一回合的反向传输资源；所述第三传输资源为第二回合的前向传输资源，所述第四传输资源为所述第二回合的反向传输资源。
  38. 根据权利要求31至37中任一项所述的装置，其特征在于，所述第一传输资源和所述第二传输资源位于：
    时域、频域、空域、码域、功率域中至少一项的固定传输资源位置上；或者，
    用于传输业务数据的时域、频域、空域、码域、功率域中至少一项的资源位置上。
  39. 根据权利要求31至38中任一项所述的装置，其特征在于，所述收发单元具体用于：
    接收来自所述第二通信装置的所述第一信息；或者，
    接收来自第三通信装置的所述第一信息，所述第三通信装置用于控制所述第一通信装置与所述第二通信装置之间的通信。
  40. 根据权利要求31至38中任一项所述的装置，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所述第一信息。
  41. 根据权利要求31至40中任一项所述的装置，其特征在于，
    所述处理单元还用于：将训练数据输入所述第一机器学习模型，得到所述第一输出；
    所述收发单元还用于：在所述第二传输资源上，接收来自所述第二通信装置的所述第一反馈数据，所述第一反馈数据用于指示所述第一梯度；
    所述处理单元还用于：根据所述第一梯度更新所述第一机器学习模型。
  42. 根据权利要求31至41中任一项所述的装置，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训练指示，所述训练指示用于指示对所述第一机器学习模型的训练。
  43. 一种通信装置，其特征在于，所述通信装置为通信系统中的第二通信装置，所述通信系统还包括第一通信装置，所述第一通信装置部署有第一机器学习模型，所述第二通信装置部署有第二机器学习模型，所述第一机器学习模型和所述第二机器学习模型用于实现所述第一通信装置与所述第二通信装置之间的通信，所述第二通信装置包括：
    处理单元、收发单元；
    所述处理单元用于：获取第一信息，所述第一信息包括第一传输资源和第二传输资源的指示信息；其中，所述第一传输资源用于接收来自所述第一通信装置的第二数据，所述第二数据为第一数据经过信道传输后所得的数据，所述第一数据为所述第一机器学习模型的第一输出；所述第二传输资源用于向所述第一通信装置发送第一反馈数据，所述第一反馈数据用于指示第一梯度，所述第一梯度用于更新所述第一机器学习模型；
    所述收发单元用于：在所述第二传输资源上，向所述第一通信装置发送所述第一反馈数据。
  44. 根据权利要求43所述的装置，其特征在于，所述处理单元还用于：
    将所述第二数据输入所述第二机器学习模型，并根据输出计算得到第三梯度，所述第三梯度用于更新所述第二机器学习模型。
  45. 根据权利要求43或44所述的装置，其特征在于，所述第一梯度为所述第二通信装置根据第二输出计算所得，所述第二输出为所述第二数据输入所述第二机器学习模型所得的输出。
  46. 根据权利要求43至45中任一项所述的装置，其特征在于，所述第一信息包括传输资源单元的指示信息，所述第一传输资源和所述第二传输资源包含于所述传输资源单元。
  47. 根据权利要求46所述的装置，其特征在于，所述传输资源单元的指示信息包括传输资源单元TCI。
  48. 根据权利要求43至47中任一项所述的装置，其特征在于，所述第一信息包括第三传输资源和第四传输资源的指示信息；其中，所述第三传输资源用于所述第二通信装置接收来自所述第一通信装置的第四数据，所述第四数据为第三数据经过信道传输后所得的数据，所述第三数据为所述第一机器学习模型的第三输出；所述第四传输资源用于所述第二通信装置向所述第一通信装置发送第二反馈数据，所述第二反馈数据用于指示第二梯度，所述第二梯度用于更新所述第一机器学习模型；
    所述收发单元还用于：在所述第四传输资源上，向所述第一通信装置发送所述第二反馈数据。
  49. 根据权利要求48所述的装置，其特征在于，所述第二梯度为所述第二通信装置根据第四输出计算所得，所述第四输出为所述第四数据输入所述第二机器学习模型所得的输出。
  50. 根据权利要求48或49所述的装置，其特征在于，所述处理单元还用于：
    将所述第四数据输入所述第二机器学习模型，并根据输出计算得到第四梯度，所述第四梯度用于更新所述第二机器学习模型。
  51. 根据权利要求48至50中任一项所述的装置，其特征在于，所述第一传输资源为第一回合的前向传输资源，所述第二传输资源为所述第一回合的反向传输资源；所述第三传输资源为第二回合的前向传输资源，所述第四传输资源为所述第二回合的反向传输资源。
  52. 根据权利要求43至51中任一项所述的装置，其特征在于，所述第一传输资源和所述第二传输资源位于：
    时域、频域、空域、码域、功率域中至少一项的固定传输资源位置上；或者，
    用于传输业务数据的时域、频域、空域、码域、功率域中至少一项的资源位置上。
  53. 根据权利要求43至52中任一项所述的装置，其特征在于，所述收发单元具体用于：
    接收来自所述第一通信装置的所述第一信息；或者，
    接收来自第三通信装置的所述第一信息，所述第三通信装置用于控制所述第一通信装置与所述第二通信装置之间的通信。
  54. 根据权利要求43至52中任一项所述的装置，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所述第一信息。
  55. 根据权利要求43至54中任一项所述的装置，其特征在于，所述处理单元还用于：
    将所述第二数据输入所述第二机器学习模型，得到所述第二输出；
    根据所述第二输出计算得到所述第一梯度。
  56. 根据权利要求43至55中任一项所述的装置，其特征在于，所述收发单元还用于：
    向所述第一通信装置传输训练指示，所述训练指示用于指示对所述第一机器学习模型的训练。
  57. 一种通信装置，其特征在于，所述通信装置为通信系统中的第三通信装置，所述通信系统还包括第一通信装置和第二通信装置，所述第三通信装置用于控制所述第一通信装置和所述第二通信装置的之间的通信，所述第一通信装置部署有第一机器学习模型，所述第二通信装置部署有第二机器学习模型，所述第一机器学习模型和所述第二机器学习模型用于实现所述第一通信装置与所述第二通信装置之间的通信，所述第三通信装置包括：处理单元、收发单元；
    所述处理单元用于：获取第一信息，所述第一信息包括第一传输资源和第二传输资源的指示信息；其中，所述第一传输资源用于所述第一通信装置向所述第二通信装置发送第一数据，所述第一数据为所述第一机器学习模型的第一输出；所述第二传输资源用于所述第二通信装置向所述第一通信装置发送第一反馈数据，所述第一反馈数据用于指示第一梯度，所述第一梯度用于更新所述第一机器学习模型；
    所述收发单元用于：向所述第一通信装置和/或所述第二通信装置发送所述第一信息。
  58. 根据权利要求57所述的装置，其特征在于，所述第一信息包括传输资源单元的指示信息，所述第一传输资源和所述第二传输资源包含于所述传输资源单元。
  59. 根据权利要求57或58所述的装置，其特征在于，所述第一信息包括第三传输资源和第四传输资源的指示信息；其中，所述第三传输资源用于所述第一通信装置向所述第二通信装置发送第三数据，所述第三数据为所述第一机器学习模型的第三输出；所述第四传输资源用于所述第二通信装置向所述第一通信装置发送第二反馈数据，所述第二反馈数据用于指示第二梯度，所述第二梯度用于更新所述第一机器学习模型。
  60. 根据权利要求57至59中任一项所述的装置，其特征在于，所述第一传输资源和所述第二传输资源位于：
    时域、频域、空域、码域、功率域中至少一项的固定传输资源位置上；或者，
    用于传输业务数据的时域、频域、空域、码域、功率域中至少一项的资源位置上。
  61. 一种通信装置，其特征在于，所述通信装置为通信系统中的第一通信装置，所述通信系统还包括第二通信装置，所述第一通信装置部署有第一机器学习模型，所述第二通信装置部署有第二机器学习模型，所述第一机器学习模型和所述第二机器学习模型用于实现所述第一通信装置与所述第二通信装置之间的通信，所述第一通信装置包括：
    处理器、收发器；
    所述收发器用于收发数据或信息；
    所述处理器用于执行权利要求1至12中任一项所述的数据传输方法。
  62. 一种通信装置，其特征在于，所述通信装置为通信系统中的第二通信装置，所述通信系统还包括第一通信装置，所述第一通信装置部署有第一机器学习模型，所述第二通信装置部署有第二机器学习模型，所述第一机器学习模型和所述第二机器学习模型用于实现所述第一通信装置与所述第二通信装置之间的通信，所述第二通信装置包括：
    处理器、收发器；
    所述收发器用于收发数据或信息；
    所述处理器用于执行权利要求13至26中任一项所述的数据传输方法。
  63. 一种通信装置，其特征在于，所述通信装置为通信系统中的第三通信装置，所述通信系统还包括第一通信装置和第二通信装置，所述第三通信装置用于控制所述第一通信装置和所述第二通信装置的之间的通信，所述第一通信装置部署有第一机器学习模型，所述第二通信装置部署有第二机器学习模型，所述第一机器学习模型和所述第二机器学习模型用于实现所述第一通信装置与所述第二通信装置之间的通信，所述第三通信装置包括：
    处理器、收发器；
    所述收发器用于收发数据或信息；
    所述处理器用于执行权利要求27至30中任一项所述的数据传输方法。
  64. 一种包含指令的计算机程序产品，其特征在于，当其在计算机上运行时，使得如权利要求1至12或13至26或27至30中任一项所述的方法被执行。
  65. 一种计算机可读存储介质，其特征在于，包括程序指令，当其在计算机上运行时，使得如权利要求1至12或13至26或27至30中任一项所述的方法被执行。
  66. 一种通信设备，其特征在于，包括处理器和存储器，所述处理器与所述存储器耦合；
    所述存储器，用于存储程序；
    所述处理器，用于执行所述存储器中的程序，使得所述处理器执行如权利要求1至12或13至26或27至30中任一项所述的方法。
  67. 一种通信系统，其特征在于，包括权利要求31至42中任一项所述的第一通信装置和权利要求43至56中任一项所述的第二通信装置。
  68. 根据权利要求67所述的系统，其特征在于，还包括权利要求57至60中任一项所述的第三通信装置。
PCT/CN2022/094051 2021-05-27 2022-05-20 一种数据传输方法和相关装置 Ceased WO2022247739A1 (zh)

Priority Applications (2)

Application Number Priority Date Filing Date Title
EP22810466.7A EP4340283A4 (en) 2021-05-27 2022-05-20 DATA TRANSMISSION METHOD AND ASSOCIATED DEVICE
US18/514,066 US20240089742A1 (en) 2021-05-27 2023-11-20 Data transmission method and related apparatus

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202110587562.0A CN115412212A (zh) 2021-05-27 2021-05-27 一种数据传输方法和相关装置
CN202110587562.0 2021-05-27

Related Child Applications (1)

Application Number Title Priority Date Filing Date
US18/514,066 Continuation US20240089742A1 (en) 2021-05-27 2023-11-20 Data transmission method and related apparatus

Publications (1)

Publication Number Publication Date
WO2022247739A1 true WO2022247739A1 (zh) 2022-12-01

Family

ID=84156135

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/094051 Ceased WO2022247739A1 (zh) 2021-05-27 2022-05-20 一种数据传输方法和相关装置

Country Status (4)

Country Link
US (1) US20240089742A1 (zh)
EP (1) EP4340283A4 (zh)
CN (1) CN115412212A (zh)
WO (1) WO2022247739A1 (zh)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN121693997A (zh) * 2023-07-31 2026-03-17 华为技术有限公司 资源配置方法以及相关装置
CN121001200A (zh) * 2024-05-20 2025-11-21 维沃移动通信有限公司 通信处理的方法、装置及终端

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111629380A (zh) * 2020-05-09 2020-09-04 中国科学院沈阳自动化研究所 面向高并发多业务工业5g网络的动态资源分配方法
US20200343985A1 (en) * 2019-04-23 2020-10-29 DeepSig Inc. Processing communications signals using a machine-learning network
WO2021086308A1 (en) * 2019-10-28 2021-05-06 Google Llc End-to-end machine learning for wireless networks
WO2021089568A1 (en) * 2019-11-04 2021-05-14 Telefonaktiebolaget Lm Ericsson (Publ) Machine learning non-standalone air-interface

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20200343985A1 (en) * 2019-04-23 2020-10-29 DeepSig Inc. Processing communications signals using a machine-learning network
WO2021086308A1 (en) * 2019-10-28 2021-05-06 Google Llc End-to-end machine learning for wireless networks
WO2021089568A1 (en) * 2019-11-04 2021-05-14 Telefonaktiebolaget Lm Ericsson (Publ) Machine learning non-standalone air-interface
CN111629380A (zh) * 2020-05-09 2020-09-04 中国科学院沈阳自动化研究所 面向高并发多业务工业5g网络的动态资源分配方法

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
"3rd Generation Partnership Project; Technical Specification Group Services and System Aspects; Study on traffic characteristics and performance requirements for AI/ML model transfer in 5GS (Release 18)", 3GPP STANDARD; TECHNICAL REPORT; 3GPP TR 22.874, 3RD GENERATION PARTNERSHIP PROJECT (3GPP), MOBILE COMPETENCE CENTRE ; 650, ROUTE DES LUCIOLES ; F-06921 SOPHIA-ANTIPOLIS CEDEX ; FRANCE, no. V0.2.0, 7 January 2021 (2021-01-07), Mobile Competence Centre ; 650, route des Lucioles ; F-06921 Sophia-Antipolis Cedex ; France , pages 1 - 65, XP051975280 *
See also references of EP4340283A4

Also Published As

Publication number Publication date
EP4340283A1 (en) 2024-03-20
US20240089742A1 (en) 2024-03-14
CN115412212A (zh) 2022-11-29
EP4340283A4 (en) 2024-11-13

Similar Documents

Publication Publication Date Title
JP7704348B2 (ja) 無線ネットワークにおいて人工知能ａｉモデルを訓練する方法及び装置
CN111586827B (zh) 功率控制方法及功率控制装置
US20240089742A1 (en) Data transmission method and related apparatus
WO2018202168A1 (zh) 信息传输方法及装置
WO2022188692A1 (zh) 一种旁链路drx配置的确定方法及通信装置
CN116347625A (zh) 一种数据传输方法及装置
CN106982465B (zh) 一种无线帧的传输方法以及无线网络设备
WO2024131889A1 (zh) 一种通信方法、装置、芯片及模组设备
CN118283692A (zh) 一种信道状态信息csi报告的配置方法及相关装置
WO2024059984A1 (zh) 传输块大小的确定方法、装置、设备及存储介质
CN114980325B (zh) 传输块大小确定方法及装置
JP7687583B1 (ja) ネットワークスライス管理装置及び管理方法
JP7687582B1 (ja) ネットワークスライス管理装置及び管理方法
CN121126435B (zh) 一种通信方法、通信装置、可读存储介质以及芯片系统
CN118509137A (zh) 上行链路控制信息的发送方法及装置
WO2026045395A1 (zh) 一种通信方法及通信装置
CN120050206A (zh) 一种语义评估方法及装置
CN121510141A (zh) 一种通信方法、通信装置及存储介质
WO2025148695A1 (zh) 一种通信方法及装置
WO2025092680A1 (zh) 一种通信方法及装置
CN121604019A (zh) 通信方法和通信装置
WO2025067233A1 (zh) 一种通信方法及装置
WO2025054924A1 (zh) 数据传输方法和通信装置
WO2023226992A1 (zh) 通信的方法和装置
CN120415661A (zh) 一种通信方法及装置

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: 22810466

Country of ref document: EP

Kind code of ref document: A1

WWE Wipo information: entry into national phase

Ref document number: 2022810466

Country of ref document: EP

ENP Entry into the national phase

Ref document number: 2022810466

Country of ref document: EP

Effective date: 20231211

NENP Non-entry into the national phase

Ref country code: DE