A neural network training method and related device

Abstract

A neural network training method and related device, the method comprising: a first device receives first channel sample information from a second device; the first device determines a first neural network, the first neural network is trained based on the first channel sample information, and the first neural network is used to infer second channel sample information based on the first channel sample information. Through this method, the air interface signaling overhead can be effectively reduced, and the channel environment can be adapted to improve communication performance.

Description

Technical Field

[0001] The present application relates to the field of communication technology, and in particular to a neural network training method and related devices. Background Art

[0002] A wireless communication system may include three parts: a transmitter, a channel, and a receiver. The channel is used to transmit signals exchanged between the transmitter and the receiver. For example, the transmitter may be an access network device, such as a base station (BS), and the receiver may be a terminal device. For another example, the transmitter may be a terminal device and the receiver may be an access network device.

[0003] In order to optimize the performance of the communication system, the above-mentioned transmitter and receiver can be optimized. The transmitter and receiver can have independent mathematical models, so they are usually optimized independently based on their respective mathematical models. For example, a mathematical channel model can be used to generate channel samples to optimize the transmitter and receiver.

[0004] Since the mathematical channel model is non-ideal and nonlinear, the channel samples generated by the mathematical channel model defined by the protocol are difficult to reflect the actual channel environment. When a large number of actual channel samples are transmitted between the sender and the receiver, too many air interface resources will be occupied, affecting the data transmission efficiency. Summary of the invention

[0005] A first aspect of an embodiment of the present application provides a neural network training method, comprising:

[0006] The first device receives first channel sample information from the second device; the first device determines a first neural network, the first neural network is trained based on the first channel sample information, and the first neural network is used to infer the second channel sample information.

[0007] Optionally, the first device is an access network device and the second device is a terminal device. It is understandable that the first device may be an access network device, a chip used in an access network device, or a circuit used in an access network device, and the second device may be a terminal device, a chip used in a terminal device, or a circuit used in a terminal device.

[0008] In one possible design, the method includes: a first device trains a first neural network according to first channel sample information. The first neural network is used to generate new channel sample information, such as second channel sample information.

[0009] In one possible design, the method includes: the first device receives information of the first neural network from the third device, and determines the first neural network according to the information of the first neural network. The first neural network is trained by the third device according to the first channel sample information.

[0010] In the method, a first neural network can be trained based on the first channel sample information. The first neural network is used to infer the second channel sample information. By using this method, the air interface signaling overhead when transmitting the channel sample information can be effectively reduced.

[0011] In one possible design, the second channel sample information is used to train a second neural network and / or a third neural network, and the second neural network and / or the third neural network are used for transmitting target information between the first device and the second device.

[0012] In one possible design, the method includes: the first device trains a second neural network and / or a third neural network based on the second channel sample information, or based on the second channel sample information and the first channel sample information, and the second neural network and / or the third neural network is used for the first device to transmit target information with the second device.

[0013] In one possible design, the method includes: a first device receives information of a second neural network and / or information of a third neural network from a third device, and the second neural network and / or the third neural network are used for the first device to transmit target information with the second device. The second neural network and / or the third neural network are trained by the third device according to the second channel sample information, or according to the second channel sample information and the first channel sample information.

[0014] This method can effectively reduce the air interface signaling overhead, and at the same time can adapt to the channel environment. The trained second neural network and the third neural network are closer to the actual channel environment, improving communication performance. The speed of training the second neural network and the third neural network is also greatly improved.

[0015] In one possible design, the method further includes: the first device sends a first reference signal to the second device. Optionally, the first reference signal includes a demodulation reference signal DMRS or a channel state information reference signal CSI-RS. Optionally, the sequence type of the first reference signal includes a ZC sequence or a Glod sequence.

[0016] In one possible design, the first channel sample information includes but is not limited to a second reference signal and / or channel state information (CSI). The second reference signal is a first reference signal propagated through a channel, or is described as a first reference signal received by a second device from a first device.

[0017] In one possible design, the information of the first neural network includes: a model change of the first neural network relative to a reference neural network.

[0018] In one possible design, the information of the first neural network includes one or more of the following: weights of the neural network, activation functions of neurons, the number of neurons in each layer of the neural network, the cascade relationship between layers of the neural network, and the network type of each layer in the neural network.

[0019] In one possible design, the first neural network is a generative neural network. Optionally, the first neural network is a generative adversarial network (GAN) or a variational autoencoder (VAE).

[0020] Through this method, second channel sample information with the same distribution or similar distribution as the first channel sample information can be obtained, so that the second channel sample information is closer to the actual channel environment.

[0021] In one possible design, the method further includes: the first device receiving capability information of the second device from the second device. The capability information is used to indicate one or more of the following information:

[0022] 1) Whether the second device supports using a neural network to replace or implement the function of the communication module;

[0023] The communication module includes but is not limited to: an OFDM modulation module, an OFDM demodulation module, a constellation mapping module, a constellation demapping module, a channel coding module, a channel decoding module, a precoding module, an equalization module, an interleaving module and / or a deinterleaving module.

[0024] 2) whether the second device supports the network type of the third neural network;

[0025] 3) whether the second device supports receiving information of the third neural network through signaling;

[0026] 4) a reference neural network stored in a second device;

[0027] 5) memory space of the second device that can be used to store the third neural network; and,

[0028] 6) Information about the computing power of the second device that can be used to run the neural network.

[0029] Through this method, the first device can receive capability information sent by the second device. The capability information is used to notify the first device of relevant information of the second device, and the first device can perform operations on the third neural network according to the capability information to ensure that the second device can normally use the third neural network.

[0030] In one possible design, the method also includes: the first device sends information of the third neural network to the second device.

[0031] In this embodiment, after the first device completes training of the third neural network, the first device sends information of the third neural network to the second device. The information of the third neural network includes but is not limited to: weights of the neural network, activation functions of neurons, the number of neurons in each layer of the neural network, the cascade relationship between layers of the neural network, and / or the network type of each layer in the neural network. Exemplarily, different activation functions may be indicated for different neurons in the information of the third neural network.

[0032] In one possible design, when the third neural network is preconfigured (or predefined) in the second device, the information of the third neural network may also be the model change of the third neural network. The model change includes, but is not limited to: the weight of the changed neural network, the changed activation function, the number of neurons in one or more layers of the changed neural network, the changed inter-layer cascade relationship of the neural network, and / or the network type of one or more layers in the changed neural network. Exemplarily, the preconfiguration may be preconfigured by the access network device through signaling, and the predefinition may be predefined by a protocol, for example: the protocol predefines the third neural network in the terminal device as neural network A.

[0033] In the embodiments of the present application, there are multiple implementation schemes for the information of the third neural network, which improves the implementation flexibility of the scheme.

[0034] In a second aspect, an embodiment of the present application provides a method for training a neural network, including: a second device performs channel estimation based on a first reference signal received from a first device to determine first channel sample information; the second device sends the first channel sample information to the first device; the second device receives information of a third neural network from the first device, and the third neural network is used to transmit target information between the second device and the first device.

[0035] In one possible design, the method also includes: the second device sending capability information of the second device to the first device.

[0036] For an introduction to the first channel sample information, the third neural network information, and the capability information of the second device, please refer to the first aspect and will not be repeated here.

[0037] In a third aspect, the present application embodiment provides a neural network training method, including:

[0038] The first device sends a first reference signal to the second device; the first device receives information of a first neural network from the second device; the first neural network is used to infer and obtain second channel sample information.

[0039] In the method, the second device (eg, terminal device) trains a first neural network based on the first channel sample information. The first neural network is used to infer the second channel sample information. By using this method, the air interface signaling overhead when transmitting the channel sample information can be effectively reduced.

[0040] In one possible design, the second channel sample information is used to train a second neural network and / or a third neural network, and the second neural network and / or the third neural network are used for transmitting target information between the first device and the second device.

[0041] In one possible design, the method includes: the first device trains a second neural network and / or a third neural network based on the second channel sample information, or based on the second channel sample information and the first channel sample information, and the second neural network and / or the third neural network is used for the first device to transmit target information with the second device.

[0042] In one possible design, the method includes: a first device receives information of a second neural network and / or information of a third neural network from a third device, and the second neural network and / or the third neural network are used for the first device to transmit target information with the second device. The second neural network and / or the third neural network are trained by the third device according to the second channel sample information, or according to the second channel sample information and the first channel sample information.

[0043] This method can effectively reduce the air interface signaling overhead, and at the same time can adapt to the channel environment. The trained second neural network and the third neural network are closer to the actual channel environment, improving communication performance. The speed of training the second neural network and the third neural network is also greatly improved.

[0044] Specifically, the introduction of the first reference signal, the first neural network, the information of the first neural network, the second neural network and / or the third neural network, etc. can be referred to the first aspect and will not be elaborated herein.

[0045] In one possible design, the method also includes: the first device sends information of the third neural network to the second device.

[0046] Specifically, the introduction of the information of the third neural network can be found in the first aspect and will not be repeated here.

[0047] In one possible design, the method further includes:

[0048] The first device receives capability information from the second device, where the capability information is used to indicate one or more of the following information of the second device:

[0049] 1) Whether it supports the use of neural networks to replace or implement the functions of the communication module;

[0050] 2) Whether the network type of the first neural network is supported;

[0051] 3) Whether the network type of the third neural network is supported;

[0052] 4) whether to support receiving information of a reference neural network through signaling, where the reference neural network is used to train the first neural network;

[0053] 5) Whether to support receiving information of the third neural network through signaling;

[0054] 6) stored reference neural network;

[0055] 7) memory space that can be used to store the first neural network and / or the third neural network;

[0056] 8) information about the computing power available to run the neural network; and,

[0057] 9) Location information of the second device.

[0058] In a fourth aspect, an embodiment of the present application provides a neural network training method, comprising:

[0059] The second device performs channel estimation based on the first reference signal received from the first device to determine first channel sample information; the second device determines a first neural network, which is trained based on the first channel sample information; the second device sends information about the first neural network to the first device.

[0060] Specifically, the introduction of the first reference signal, the first neural network, the first channel sample information, and the information of the first neural network can be found in the third aspect and will not be elaborated here.

[0061] In one possible design, the method further includes:

[0062] The second device receives information of the third neural network from the first device. Please refer to the third aspect for information of the third neural network, which will not be described in detail here.

[0063] In a possible design, the method further includes: the second device sends capability information to the first device. Please refer to the third aspect for the capability information, which will not be described in detail here.

[0064] In a fifth aspect, an embodiment of the present application proposes a neural network training method, comprising: a third device receives first channel sample information from a first device; the third device trains a first neural network based on the first channel sample information, and the first neural network is used to infer second channel sample information.

[0065] In one possible design, the method further includes: training a second neural network and / or a third neural network based on the second channel sample information, sending information of the second neural network and / or the third neural network to the first device, and using the second neural network and / or the third neural network to transmit target information between the first device and the second device.

[0066] In a sixth aspect, a device is provided, which may be an access network device, or a device in an access network device, or a device that can be used in combination with an access network device.

[0067] In one possible design, the device may include a module corresponding to the method / operation / step / action described in the first aspect, and the module may be a hardware circuit, or software, or a combination of a hardware circuit and software. In one design, the device may include a processing module and a transceiver module. Exemplarily,

[0068] The transceiver module is used to receive first channel sample information from the second device;

[0069] The processing module is used to determine a first neural network, where the first neural network is trained based on the first channel sample information, and the first neural network is used to infer the second channel sample information.

[0070] For an introduction to the first neural network, first channel sample information, second channel sample information and other operations, please refer to the first aspect and will not be repeated here.

[0071] In one possible design, the device may include a module corresponding to the method / operation / step / action described in the third aspect, and the module may be a hardware circuit, or software, or a combination of a hardware circuit and software. In one design, the device may include a processing module and a transceiver module. Exemplarily,

[0072] The transceiver module is used to send a first reference signal to a second device; and receive information of a first neural network from the second device; the first neural network is used to infer and obtain second channel sample information.

[0073] For an introduction to the first neural network, first channel sample information, second channel sample information and other operations, please refer to the third aspect and will not be elaborated here.

[0074] In a seventh aspect, a device is provided, which may be a terminal device, or a device in a terminal device, or a device that can be used in combination with a terminal device.

[0075] In one possible design, the device may include a module corresponding to the method / operation / step / action described in the second aspect, and the module may be a hardware circuit, or software, or a combination of a hardware circuit and software. In one design, the device may include a processing module and a transceiver module. Exemplarily,

[0076] A processing module, configured to perform channel estimation according to a first reference signal received from a first device, and determine first channel sample information;

[0077] A transceiver module, used for sending first channel sample information to the first device;

[0078] The transceiver module is also used to receive information of the third neural network from the first device, and the third neural network is used to transmit target information between the second device and the first device.

[0079] For an introduction to the first reference signal, the first channel sample information, the third neural network, and other operations, please refer to the second aspect and will not be repeated here.

[0080] In one possible design, the device may include a module corresponding to the method / operation / step / action described in the fourth aspect, and the module may be a hardware circuit, or software, or a combination of a hardware circuit and software. In one design, the device may include a processing module and a transceiver module. Exemplarily,

[0081] A processing module, configured to perform channel estimation according to a first reference signal received from a first device, and determine first channel sample information;

[0082] The processing module is also used to determine a first neural network, where the first neural network is trained based on the first channel sample information;

[0083] The transceiver module is used to send information of the first neural network to the first device.

[0084] For an introduction to the first reference signal, the first channel sample information, the first neural network, and other operations, please refer to the fourth aspect and will not be repeated here.

[0085] In an eighth aspect, a device is provided, which may be an AI node, or a device in an AI node, or a device that can be used in conjunction with an AI node.

[0086] In one possible design, the device may include a module corresponding to the method / operation / step / action described in the fifth aspect, and the module may be a hardware circuit, or software, or a combination of a hardware circuit and software. In one design, the device may include a processing module and a transceiver module. Exemplarily,

[0087] The transceiver module is used to receive first channel sample information from the first device;

[0088] The processing module is used to train a first neural network according to the first channel sample information, and the first neural network is used to infer the second channel sample information.

[0089] In one possible design, the processing module is also used to train the second neural network and / or the third neural network based on the second channel sample information; the transceiver module is also used to send information of the second neural network and / or the third neural network to the first device, and the second neural network and / or the third neural network are used for the first device to transmit target information with the second device.

[0090] In a ninth aspect, an embodiment of the present application provides a device.

[0091] In one possible design, the device includes a processor for implementing the method described in the first aspect above. The device may also include a memory for storing instructions and data. The memory is coupled to the processor, and when the processor executes the instructions stored in the memory, the method described in the first aspect above can be implemented. The device may also include a communication interface, and the communication interface is used for the device to communicate with other devices. Exemplarily, the communication interface may be a transceiver, circuit, bus, module or other type of communication interface. In one possible design, the device includes:

[0092] A memory for storing program instructions;

[0093] The processor is configured to receive first channel sample information from a second device using a communication interface;

[0094] The processor is also used to determine a first neural network, where the first neural network is trained based on the first channel sample information, and the first neural network is used to infer the second channel sample information.

[0095] For an introduction to the first neural network, first channel sample information, second channel sample information and other operations, please refer to the first aspect and will not be repeated here.

[0096] In one possible design, the device includes a processor for implementing the method described in the third aspect above. The device may also include a memory for storing instructions and data. The memory is coupled to the processor, and when the processor executes the instructions stored in the memory, the method described in the third aspect above may be implemented. The device may also include a communication interface, and the communication interface is used for the device to communicate with other devices. Exemplarily, the communication interface may be a transceiver, circuit, bus, module or other type of communication interface. In one possible design, the device includes:

[0097] A memory for storing program instructions;

[0098] The processor is used to send a first reference signal to a second device using a communication interface; and receive information of a first neural network from the second device; the first neural network is used to infer and obtain second channel sample information.

[0099] For an introduction to the first neural network, first channel sample information, second channel sample information and other operations, please refer to the third aspect and will not be elaborated here.

[0100] In a tenth aspect, an embodiment of the present application provides a device.

[0101] In one possible design, the device includes a processor for implementing the method described in the second aspect above. The device may also include a memory for storing instructions and data. The memory is coupled to the processor, and when the processor executes the instructions stored in the memory, the method described in the second aspect above may be implemented. The device may also include a communication interface, and the communication interface is used for the device to communicate with other devices. Exemplarily, the communication interface may be a transceiver, circuit, bus, module or other type of communication interface. In one possible design, the device includes:

[0102] A memory for storing program instructions;

[0103] The processor is used to utilize the communication interface to perform channel estimation based on the first reference signal received from the first device, determine the first channel sample information, send the first channel sample information to the first device, and receive the information of the third neural network from the first device, the third neural network is used for the second device to transmit the target information with the first device.

[0104] For an introduction to the first reference signal, the first channel sample information, the third neural network, and other operations, please refer to the second aspect and will not be repeated here.

[0105] In one possible design, the device includes a processor for implementing the method described in the fourth aspect above. The device may also include a memory for storing instructions and data. The memory is coupled to the processor, and when the processor executes the instructions stored in the memory, the method described in the fourth aspect above may be implemented. The device may also include a communication interface, and the communication interface is used for the device to communicate with other devices. Exemplarily, the communication interface may be a transceiver, circuit, bus, module or other type of communication interface. In one possible design, the device includes:

[0106] A memory for storing program instructions;

[0107] a processor, configured to use a communication interface to perform channel estimation according to a first reference signal received from a first device, determine first channel sample information, determine a first neural network and send the information of the first neural network to the first device, wherein the first neural network is trained according to the first channel sample information;

[0108] For an introduction to the first reference signal, the first channel sample information, the first neural network, and other operations, please refer to the fourth aspect and will not be repeated here.

[0109] In the eleventh aspect, an embodiment of the present application provides a device. The device includes a processor for implementing the method described in the fifth aspect above. The device may also include a memory for storing instructions and data. The memory is coupled to the processor, and when the processor executes the instructions stored in the memory, the method described in the fifth aspect above can be implemented. The device may also include a communication interface, which is used for the device to communicate with other devices. Exemplarily, the communication interface may be a transceiver, circuit, bus, module or other type of communication interface. In one possible design, the device includes:

[0110] A memory for storing program instructions;

[0111] The processor is used to receive first channel sample information from a first device using a communication interface, and obtain a first neural network through training according to the first channel sample information, wherein the first neural network is used to infer second channel sample information.

[0112] In one possible design, the processor is also used to train a second neural network and / or a third neural network based on the second channel sample information; the processor also uses the communication interface to send information of the second neural network and / or the third neural network to the first device, and the second neural network and / or the third neural network are used for the first device to transmit target information with the second device.

[0113] In the twelfth aspect, an embodiment of the present application also provides a computer-readable storage medium, comprising instructions, which, when executed on a computer, enables the computer to execute any one of the methods in the first to fifth aspects.

[0114] In the thirteenth aspect, an embodiment of the present application also provides a computer program product, comprising instructions, which, when executed on a computer, enable the computer to execute any one of the methods in the first to fifth aspects.

[0115] In a fourteenth aspect, an embodiment of the present application provides a chip system, which includes a processor and may also include a memory, for implementing any method in the first to fifth aspects above. The chip system may be composed of a chip, or may include a chip and other discrete devices.

[0116] In a fifteenth aspect, an embodiment of the present application further provides a communication system, wherein the communication system includes:

[0117] The device of the sixth aspect and the device of the seventh aspect;

[0118] The device of the sixth aspect, the device of the seventh aspect, and the device of the eighth aspect;

[0119] The device of the ninth aspect and the device of the tenth aspect;

[0120] The device of the ninth aspect, the device of the tenth aspect, and the device of the eleventh aspect. BRIEF DESCRIPTION OF THE DRAWINGS

[0121] Figure 1 A schematic diagram of a network architecture provided for an embodiment of the present application;

[0122] Figure 2 A schematic diagram of the hardware structure of a communication device provided in an embodiment of the present application;

[0123] Figure 3 A schematic diagram of a neuron structure provided in an embodiment of the present application;

[0124] Figure 4 A schematic diagram of the layer relationship of a neural network provided in an embodiment of the present application;

[0125] Figure 5 A schematic diagram of a convolutional neural network (CNN) provided in an embodiment of the present application;

[0126] Figure 6 A schematic diagram of a feedforward neural network RNN ​​provided in an embodiment of the present application;

[0127] Figure 7 A schematic diagram of a generative adversarial network (GAN) provided in an embodiment of the present application;

[0128] Figure 8A schematic diagram of a variational autoencoder provided in an embodiment of the present application;

[0129] Fig. 9 A schematic diagram of a neural network architecture for jointly optimizing constellation modulation / demodulation for transmission and reception provided in an embodiment of the present application;

[0130] Figure 10-12 A flowchart of a neural network training method provided in an embodiment of the present application;

[0131] Fig.13 A schematic diagram of the structure of a generator network of the first neural network provided in an embodiment of the present application;

[0132] Fig.14 A schematic diagram of the structure of a discriminator network of the first neural network provided in an embodiment of the present application;

[0133] Fig.15a A schematic diagram of a generator network structure of a first neural network provided in an embodiment of the present application;

[0134] Fig.15b A schematic diagram of a discriminator network structure of a first neural network provided in an embodiment of the present application;

[0135] Fig.16a and Fig.16b A schematic diagram of a network structure provided for an embodiment of the present application;

[0136] Fig.17 A schematic diagram of a communication device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0137] In a wireless communication system, including communication equipment, wireless resources can be used for wireless communication between communication equipment. Among them, the communication equipment may include access network equipment and terminal equipment, and the access network equipment may also be referred to as access side equipment. Wireless resources may include link resources and / or air interface resources. Air interface resources may include at least one of time domain resources, frequency domain resources, code resources and space resources. In the embodiment of the present application, at least one (individual) may also be described as one (individual) or multiple (individual)), and multiple (individual) may be two (individual), three (individual), four (individual) or more (individual), and the present application does not limit it.

[0138] In the embodiments of the present application, " / " can indicate that the objects associated before and after are in an "or" relationship, for example, A / B can indicate A or B; "and / or" can be used to describe that there are three relationships between the associated objects, for example, A and / or B can indicate: A exists alone, A and B exist at the same time, and B exists alone, where A and B can be singular or plural. In order to facilitate the description of the technical solutions of the embodiments of the present application, in the embodiments of the present application, the words "first" and "second" can be used to distinguish between technical features with the same or similar functions. The words "first" and "second" do not limit the quantity and execution order, and the words "first" and "second" do not necessarily limit the difference. In the embodiments of the present application, the words "exemplary" or "for example" are used to indicate examples, illustrations or explanations, and any embodiment or design described as "exemplary" or "for example" should not be interpreted as being more preferred or more advantageous than other embodiments or design. The use of words such as "exemplary" or "for example" is intended to present related concepts in a specific way for easy understanding.

[0139] like Figure 1 The communication system in the embodiment of the present application may include an access network device (such as Figure 1 The system may be a base station (as shown) and a terminal device, or a system including two or more terminal devices. In this communication system, the access network device may send configuration information to the terminal device, and the terminal device may configure accordingly according to the configuration information. The access network device may send downlink data to the terminal device, and / or the terminal device may send uplink data to the access network device. In a communication system (such as the Internet of Vehicles) including two or more terminal devices, terminal device 1 may send configuration information to terminal device 2, and terminal device 2 may configure accordingly according to the configuration information. Terminal device 1 may send data to terminal device 2, and terminal device 2 may also send data to terminal device 1. Optionally, Figure 1 In the communication system shown, the access network device can implement one or more of the following artificial intelligence (AI) functions: model training, and reasoning. Figure 1In the communication system shown, the network side may include a node independent of the access network device, which is used to implement one or more of the following AI functions: model training, and reasoning. The node may be called an AI node, a model training node, an reasoning node, a wireless intelligent controller, or other names, without limitation. Exemplarily, the model training function and the reasoning function may be implemented by the access network device; or, the model training function and the reasoning function may be implemented by the AI ​​node; or, the model training function may be implemented by the AI ​​node, and the information of the model may be sent to the access network device, and the reasoning function may be implemented by the access network device. Optionally, if the AI ​​node implements the reasoning function, the AI ​​node may send the reasoning result to the access network device for use by the access network device, and / or the AI ​​node may send the reasoning result to the terminal device through the access network device for use by the terminal device; if the access network device implements the reasoning function, the access network device may use the reasoning result, or send the reasoning result to the terminal device for use by the terminal network device. If the AI ​​node is used to implement the model training function and the reasoning function, the AI ​​node may be separated into two nodes, one of which implements the model training function and the other implements the reasoning function.

[0140] The embodiments of the present application do not limit the specific number of network elements in the communication system involved.

[0141] The terminal device involved in the embodiments of the present application may also be referred to as a terminal or an access terminal, and may be a device with wireless transceiver functions. The terminal device may communicate with one or more core networks (CN) via an access network device. The terminal device may be a user unit, a user station, a mobile station, a mobile station, a remote station, a remote terminal, a mobile device, a user terminal, a user equipment (UE), a user agent or a user device, etc. The terminal device may be deployed on land, including indoors or outdoors, handheld or vehicle-mounted; it may also be deployed on the water (such as a ship, etc.); it may also be deployed in the air (such as an airplane, a balloon, and a satellite, etc.). The terminal device may be a cellular phone, a cordless phone, a session initiation protocol (SIP) phone, a smart phone, a mobile phone, a wireless local loop (WLL) station, or a personal digital assistant (PDA). The terminal device may also be a handheld device, a computing device or other device with wireless communication function, a vehicle-mounted device, a wearable device, a drone device, a terminal in the Internet of Things or the Internet of Vehicles, a terminal in the fifth generation (5G) mobile communication network, a relay user device, or a terminal in a future evolved mobile communication network (etc. Among them, the relay user device may be, for example, a 5G residential gateway (RG). For another example, the terminal device may be a virtual reality (VR) terminal, an augmented reality (AR) terminal, a wireless terminal in industrial control, a wireless terminal in self-driving, a wireless terminal in remote medical, a wireless terminal in smart grid, a wireless terminal in transportation safety, a wireless terminal in a smart city, or a smart home (smart home). The embodiment of the present application does not limit this. In the embodiment of the present application, the device for realizing the function of the terminal can be a terminal; it can also be a device that can support the terminal to realize the function, such as a chip system, which can be installed in the terminal or used in combination with the terminal. In the embodiment of the present application, the chip system can be composed of chips, or it can include chips and other discrete devices.

[0142] The access network device can be regarded as a sub-network of the operator network, and is an implementation system between the service node and the terminal device in the operator network. To access the operator network, the terminal device can first pass through the access network device, and then connect to the service node of the operator network through the access network device. The access network device in the embodiment of the present application is a device located in the (radio) access network ((R)AN) and capable of providing wireless communication functions for the terminal device. The access network equipment includes base stations, for example, including but not limited to: the next generation node B (gNB) in the 5G system, the evolved node B (eNB) in the long term evolution (LTE) system, the radio network controller (RNC), the node B (NB), the base station controller (BSC), the base transceiver station (BTS), the home base station (for example, home evolved nodeB, or home node B, HNB), the base band unit (BBU), the transmission point (TRP), the transmitting point (TP), the small base station equipment (pico), the mobile switching center, or the access network equipment in the future network. In the system using different wireless access technologies, the name of the equipment with the access network equipment function may be different. In the embodiment of the present application, the device for realizing the function of the access network equipment may be the access network equipment; it may also be a device that can support the access network equipment to realize the function, such as a chip system, which can be installed in the access network equipment or used in combination with the access network equipment.

[0143] The technical solution provided in the embodiments of the present application can be applied to various communication systems, such as: LTE system, 5G system, wireless-fidelity (WiFi) system, future sixth-generation mobile communication system, or a system integrating multiple communication systems, etc., which is not limited in the embodiments of the present application. Among them, 5G can also be called new radio (NR).

[0144] The technical solution provided in the embodiment of the present application can be applied to various communication scenarios, for example, it can be applied to one or more of the following communication scenarios: enhanced mobile broadband (eMBB) communication, ultra-reliable low-latency communication (URLLC), machine type communication (MTC), massive machine type communication (mMTC), device-to-device (D2D) communication, vehicle to everything (V2X) communication, vehicle to vehicle (V2V) communication, and Internet of things (IoT), etc. In the embodiment of the present application, the term "communication" can also be described as "transmission", "information transmission", "data transmission", or "signal transmission". Transmission may include sending and / or receiving. In the embodiment of the present application, the technical solution is described by taking the communication between the access network device and the terminal device as an example. Those skilled in the art may also use this technical solution for communication between other scheduling entities and subordinate entities, such as communication between a macro base station and a micro base station, and / or communication between terminal device 1 and terminal device 2.

[0145] In addition, the network architecture and business scenarios described in this application are intended to more clearly illustrate the technical solutions of this application and do not constitute a limitation on the technical solutions provided by this application. Those skilled in the art will appreciate that with the evolution of the network architecture and the emergence of new business scenarios, the technical solutions provided by this application are also applicable to similar technical problems.

[0146] Figure 2Schematic diagram of the hardware structure of the communication device in the embodiment of the present application. The communication device may be a possible implementation of the AI ​​node, access network device or terminal device in the embodiment of the present application. The communication device may be an AI node, may be a device in an AI node, or may be a device that can be used with an AI node. The communication device may be an access network device, may be a device in an access network device, or may be a device that can be used with an access network device. The communication device may be a terminal device, may be a device in a terminal device, or may be a device that can be used with a terminal device. Among them, the device may be a chip system. In the embodiment of the present application, the chip system may be composed of chips, or may include chips and other discrete devices. The connection in the embodiment of the present application is an indirect coupling or communication connection between devices, units or modules, which may be electrical, mechanical or other forms, and is used for information exchange between devices, units or modules.

[0147] like Figure 2 As shown, the communication device includes at least one processor 204 for implementing the technical solution provided in the embodiment of the present application. Optionally, the communication device may also include a memory 203. The memory 203 is used to store instructions 2031 and / or data 2032. The memory 203 is connected to the processor 204. The processor 204 can operate in conjunction with the memory 203. The processor 204 may execute instructions stored in the memory 203 to implement the technical solution provided in the embodiment of the present application. The communication device may also include a transceiver 202 for receiving and / or sending signals. Optionally, the communication device may also include one or more of the following: an antenna 206, an I / O (input / output, Input / Output) interface 210 and a bus 212. The transceiver 202 further includes a transmitter 2021 and a receiver 2022. The processor 204, the transceiver 202, the memory 203 and the I / O interface 210 are connected to each other through the bus 212, and the antenna 206 is connected to the transceiver 202. The bus 212 may include an address bus, a data bus, and / or a control bus, etc. Figure 2 In the figure, only one thick line is used to represent the bus 212, but this does not mean that the bus 212 has only one bus or only one type of bus.

[0148] The processor 204 may be a general-purpose processor, a digital signal processor, an application-specific integrated circuit, a field programmable gate array and / or other programmable logic devices, such as discrete gates or transistor logic devices and / or discrete hardware components. A general-purpose processor may be a microprocessor or any conventional processor, etc. The processor 204 may implement or execute the various methods, steps and logic block diagrams disclosed in the embodiments of the present application. The steps of the method disclosed in the embodiments of the present application may be embodied as being executed by a hardware processor, or may be executed by a combination of hardware and software modules in the processor. Exemplarily, the processor 204 may be a central processing unit (CPU), or may be a dedicated processor, such as, but not limited to, a digital signal processor (DSP), an application-specific integrated circuit (ASIC) and / or a field programmable gate array (FPGA). The processor 204 may also be a neural processing unit (NPU). In addition, the processor 204 may also be a combination of multiple processors. In the technical solution provided in the embodiments of the present application, the processor 204 may be used to execute the relevant steps in the subsequent method embodiments. Processor 204 may be a processor specifically designed to perform the above steps and / or operations, or may be a processor that performs the above steps and / or operations by reading and executing instructions 2031 stored in memory 203. Processor 204 may need to use data 2032 in the process of performing the above steps and / or operations.

[0149] The transceiver 202 includes a transmitter 2021 and a receiver 2022. In an optional implementation, the transmitter 2021 is used to send a signal through at least one antenna among the antennas 206. The receiver 2022 is used to receive a second reference signal through at least one antenna among the antennas 206.

[0150] In the embodiment of the present application, the transceiver 202 is used to support the communication device to perform the receiving function and the sending function. The processor with the processing function is regarded as the processor 204. The receiver 2022 can also be called an input port, a receiving circuit, a receiving bus or other devices that implement the receiving function, etc., and the transmitter 2021 can be called a transmitting port, a transmitting circuit, a transmitting bus or other devices that implement the transmitting function. The transceiver 202 can also be called a communication interface.

[0151] The processor 204 can be used to execute the instructions stored in the memory 203, for example, to control the transceiver 202 to receive messages and / or send messages, and complete the function of the communication device in the method embodiment of the present application. As an implementation method, the function of the transceiver 202 can be considered to be implemented by a transceiver circuit or a dedicated transceiver chip. In the embodiment of the present application, the transceiver 202 receiving a message can be understood as the transceiver 202 inputting a message, and the transceiver 202 sending a message can be understood as the transceiver 202 outputting a message.

[0152] The memory 203 may be a non-volatile memory, such as a hard disk drive (HDD) or a solid-state drive (SSD), or may be a volatile memory (volatile memory), such as a random-access memory (RAM). The memory is any other medium that can be used to carry or store the desired program code in the form of an instruction or data structure and can be accessed by a computer, but is not limited thereto. The memory in the embodiment of the present application may also be a circuit or any other device that can implement a storage function, for storing program instructions and / or data. The memory 203 is specifically used to store instructions 2031 and data 2032. The processor 204 can perform the steps and / or operations in the embodiment of the method of the present application by reading and executing the instructions 2031 stored in the memory 203. The data 2032 may be needed in the process of performing the operations and / or steps in the embodiment of the method of the present application.

[0153] Optionally, the communication device may further include an I / O interface 210, wherein the I / O interface 210 is used to receive instructions and / or data from a peripheral device, and output instructions and / or data to the peripheral device.

[0154] Next, some concepts involved in the embodiments of the present application are introduced:

[0155] Machine learning (ML) has attracted widespread attention from academia and industry in recent years. Due to the great advantages of machine learning in dealing with structured information and massive data, many researchers in the field of communications have also turned their attention to machine learning. Communication technologies based on machine learning have great potential in signal classification, channel estimation, and / or performance optimization. Most communication systems are designed block by block, which means that these communication systems are composed of multiple modules. For this communication architecture based on module design, many technologies can be developed to optimize the performance of each module. However, the optimal performance of each module does not necessarily mean the optimal performance of the entire communication system. End-to-end optimization (i.e., optimizing the entire communication system) may be better than optimizing a single model. Machine learning provides an advanced and powerful tool for maximizing end-to-end performance as much as possible. In wireless communication systems, channel conditions change rapidly in complex and large-scale communication scenarios. Many traditional communication models, such as large-scale multiple input multiple output (MIMO) models, rely heavily on channel state information, and the performance of these models deteriorates under nonlinear time-varying channels. Therefore, accurately obtaining the channel state information (CSI) of time-varying channels is crucial to system performance. By utilizing machine learning techniques, it is possible for communication systems to learn sudden changes in channel models and provide timely feedback on channel status.

[0156] Based on the above considerations, the use of machine learning technology in wireless communications can adapt to new demands in future wireless communication scenarios.

[0157] Machine learning is an important technical approach to achieve artificial intelligence. Machine learning can include supervised learning, unsupervised learning, and reinforcement learning.

[0158] Supervised learning uses machine learning algorithms to learn the mapping relationship from sample values ​​to sample labels based on the collected sample values ​​and sample labels, and uses machine learning models to express the learned mapping relationship. Among them, sample labels can also be referred to as labels. The process of training a machine learning model is the process of learning this mapping relationship. For example, in signal detection, the noisy received signal is the sample, and the real constellation point corresponding to the signal is the label. Machine learning expects to learn the mapping relationship between sample values ​​and labels through training, that is, to make the machine learning model learn a signal detector. During training, the model parameters are optimized by calculating the error between the model's predicted value and the real label. After the mapping relationship is learned, the learned mapping can be used to predict the sample label of a new sample. The mapping relationship learned by supervised learning can include linear mapping or nonlinear mapping. According to the type of label, the learning task can be divided into classification task and regression task.

[0159] Unsupervised learning uses algorithms to discover the inherent patterns of samples based on the collected sample values. There is a type of algorithm in unsupervised learning that uses the samples themselves as supervisory signals, that is, the model learns the mapping relationship from sample to sample, which is also called self-supervised learning. During training, the model parameters are optimized by calculating the error between the model's predicted value and the sample itself. For example, self-supervised learning can be used in applications such as signal compression and decompression recovery. Common algorithms include autoencoders and generative adversarial networks.

[0160] Reinforcement learning is different from supervised learning. It is a type of algorithm that learns problem-solving strategies by interacting with the environment. Unlike supervised and unsupervised learning, reinforcement learning problems do not have clear "correct" action label data. The algorithm needs to interact with the environment to obtain reward signals from the environment, and then adjust the decision-making actions to obtain a larger reward signal value. For example, in downlink power control, the reinforcement learning model adjusts the downlink transmission power of each user according to the total system throughput fed back by the wireless network, and then expects to obtain a higher system throughput. The goal of reinforcement learning is also to learn the mapping relationship between the state of the environment and the optimal decision action. However, because the label of the "correct action" cannot be obtained in advance, the network cannot be optimized by calculating the error between the action and the "correct action". Reinforcement learning training is achieved through iterative interaction with the environment.

[0161] Deep neural network (DNN) is a specific implementation form of machine learning. According to the universal approximation theorem, DNN can theoretically approximate any continuous function, thus enabling DNN to learn any mapping. Traditional communication systems require extensive expert knowledge to design communication modules, while DNN-based deep learning communication systems can automatically discover implicit pattern structures from large data sets, establish mapping relationships between data, and achieve performance that is superior to traditional modeling methods.

[0162] The idea of ​​DNN comes from the neuron structure of the brain. Each neuron performs a weighted sum operation on its input value and generates an output through an activation function. Figure 3 As shown, Figure 3 is a schematic diagram of the neuron structure. Assume that the input of the neuron is x = [x0, x1, ...x n ], and the weight corresponding to the input is w=[w0, w1, ...w n ], the bias of the weighted sum is b, and the activation function can be diversified. As an example, if the activation function of a neuron is: y = f(x) = max{0, x}, then the output of the execution of a neuron is: Among them, w i x i Indicates wi With x i The data type of b is an integer or floating point number, and w i The data type is an integer or a floating point number. DNN generally has a multi-layer structure, and each layer of DNN can contain one or more neurons. The input layer of DNN passes the received values ​​to the middle hidden layer after processing by neurons. Similarly, the hidden layer passes the calculation results to the next hidden layer or the next output layer to generate the final output of DNN. Figure 4 As shown, Figure 4 Schematic diagram of the layer relationship of the neural network.

[0163] DNN generally has one or more hidden layers, which can affect the ability to extract information and fit functions. Increasing the number of hidden layers of DNN or expanding the number of neurons in each layer can improve the function fitting ability of DNN. The parameters of each neuron include weights, biases, and activation functions. The set of parameters of all neurons in DNN is called DNN parameters (or neural network parameters). The weights and biases of neurons can be optimized through the training process, so that DNN has the ability to extract data features and express mapping relationships. DNN generally uses supervised learning or unsupervised learning strategies to optimize neural network parameters.

[0164] Depending on how the network is constructed, DNNs can include feedforward neural networks (FNNs), convolutional neural networks (CNNs), and recurrent neural networks (RNNs).

[0165] Figure 4 The figure shows an FNN, which is characterized by the neurons in adjacent layers being fully connected to each other, which usually requires a large amount of storage space and leads to high computational complexity.

[0166] like Figure 5 As shown, Figure 5is a schematic diagram of CNN. CNN is a neural network used to process data with a grid-like structure. For example, time series data (discrete sampling on the time axis) and image data (two-dimensional discrete sampling) can be considered as data with a similar grid structure. CNN does not use all the input information for calculations at one time, but uses a window (such as a window of fixed size) to intercept part of the information for convolution operations, which greatly reduces the amount of calculation of neural network parameters. In addition, depending on the type of information intercepted by the window (such as people and objects in the same picture are different types of information), each window can use different convolution kernel operations, which enables CNN to better extract the features of the input data. Among them, the convolution layer is used to extract features and obtain feature maps. The pooling layer is used to compress the input feature map to make the feature map smaller and simplify the network calculation complexity. The fully connected layer is used to map the learned "distributed feature representation" to the sample labeling space. Exemplarily, Figure 5 In the image, the probability of judging that the image is the sun is 0.7, the probability of judging that the image is the moon is 0.1, the probability of judging that the image is a car is 0.05, and the probability of judging that the image is a house is 0.02.

[0167] like Figure 6 As shown, Figure 6 The figure is a schematic diagram of RNN. RNN is a type of DNN that uses feedback time series information. Its input includes the new input value at the current moment and its own output value at the previous moment. RNN is suitable for obtaining sequence features that are correlated in time, and is particularly suitable for applications such as speech recognition and channel coding. Figure 6 , a neuron produces multiple outputs at multiple times. For example, at time 0, the input is x0 and s0, the output is y0 and s1, at time 1, the input is x1 and s1, the output is y1 and s2, ..., at time t, the input is x t and t , the output is y t and t+1 .

[0168] A generative neural network is a special type of deep learning neural network. Unlike general neural networks that mainly perform classification and prediction tasks, a generative neural network can learn the probability distribution function obeyed by a set of training samples, so it can be used to model random variables and establish conditional probability distributions between variables. Common generative neural networks include generative adversarial networks (GANs) and variational autoencoders (VAEs).

[0169] See also Figure 7 , Figure 7The schematic diagram of the Generative Adversarial Network (GAN) is shown in Figure 1. The GAN network can include two parts according to its function: a generator network (referred to as the generator) and a discriminator network (referred to as the discriminator). Specifically, the generator network processes the input random noise and outputs a generated sample. The discriminator network compares the generated sample output by the generator network with the training sample in the training set, determines whether the generated sample and the training sample approximately obey similar probability distributions, and outputs a true or false judgment. For example, taking the generated sample and the training sample obeying a normal distribution as an example, when the mean of the generated sample is consistent with the mean of the training sample, and when the difference between the variance of the generated sample and the variance of the training sample is less than a threshold value (such as 0.01), it is considered that the generated sample and the training sample obey similar probability distributions. There is a game relationship between the generator network and the discriminator network: the generator network hopes to generate samples that obey the distribution of the training set as much as possible, and the discriminator network hopes to distinguish the difference between the generated sample and the training set as much as possible. By jointly training the generator network and the discriminator network, the two can reach a balanced state, that is, the probability distribution of the generated samples output by the generator network is similar to the probability distribution of the training samples, and the discriminator network believes that the generated samples and the training set follow a similar distribution. Optionally, this similar distribution can be called the same distribution.

[0170] See also Figure 8 , Figure 8 The diagram is a diagram of a variational autoencoder. VAE can be divided into three parts according to its function: an encoder network (referred to as encoder), a decoder network (referred to as decoder), and a discriminator network (referred to as discriminator). The encoder network compresses the samples in the input training set into intermediate variables, and the decoder network attempts to restore the intermediate variables to samples in the training set. At the same time, certain constraints can be placed on the form of the intermediate variables. Similar to the discriminant network in GAN, a discriminant network can also be used in VAE to determine whether the intermediate variables obey the distribution of random noise. By jointly training the encoder network, the decoder network, and the discriminator network, the decoder network can use the input random noise to generate generated samples that conform to the distribution of the training set.

[0171] The above-mentioned FNN, CNN, RNN, GAN and VAE are neural network structures, and these network structures are all constructed based on neurons.

[0172] Thanks to the advantages of machine learning in modeling and extracting information features, communication solutions based on machine learning can be designed to achieve better performance. This includes but is not limited to CSI compression feedback, adaptive constellation point design, and / or robust precoding. These solutions optimize transmission performance or reduce processing complexity by replacing the sending or receiving modules in the original communication system with neural network models. In order to support different application scenarios, different neural network model information can be pre-defined or configured so that the neural network model can adapt to the requirements of different scenarios.

[0173] In the embodiment of the present application, in the communication system based on the neural network, some communication modules of the access network device and / or the terminal device may adopt the neural network model. Fig. 9 As shown, Fig. 9 Schematic diagram of the neural network architecture for jointly optimizing constellation modulation / demodulation for both transmission and reception. Among them, the constellation mapping neural network of the transmitting end (also called the transmitting end) and the constellation demapping neural network of the receiving end (also called the receiving end) both adopt neural network models. The constellation mapping neural network of the transmitting end maps the bit stream into constellation symbols, and the constellation demapping neural network demaps (or demodulates) the received constellation symbols into the log-likelihood ratio of the bit information. By collecting channel data to train the neural network, better end-to-end communication performance can be achieved. In the communication system, the transmitting end performs a series of processing on the bit stream to be transmitted, which may include one or more of the following: channel coding, constellation symbol mapping modulation, orthogonal frequency division multiplexing (OFDM) modulation, layer mapping, precoding, and up-conversion. For the convenience of expression, Fig. 9 Only constellation mapping symbol modulation and OFDM modulation are listed.

[0174] It should be noted that the neural network involved in the embodiments of the present application is not limited to any specific application scenario, but can be applied to any communication scenario, such as CSI compression feedback, adaptive constellation point design, and / or robust precoding.

[0175] The above neural network needs to be adaptively trained to ensure communication performance. For example, for each different wireless system parameter (including one or more of wireless channel type, bandwidth, number of receiving antennas, number of transmitting antennas, modulation order, number of paired users, channel coding method, and coding code rate), a set of corresponding neural network model information (including neural network structure information and neural network parameters) is defined. Taking the adaptive modulation constellation design based on artificial intelligence (AI) as an example, when the number of transmitting antennas is 1 and the number of receiving antennas is 2, a set of neural network model information is required to generate the corresponding modulation constellation; when the number of transmitting antennas is 1 and the number of receiving antennas is 4, another set of neural network model information is required to generate the corresponding modulation constellation. Similarly, different wireless channel types, bandwidths, modulation orders, number of paired users, channel coding methods, and / or coding code rates may have different corresponding neural network model information.

[0176] In order to make the (transmitter and receiver) neural network achieve better performance under actual channel conditions, it is necessary to use actual channel sample information for joint training when training the neural network. Generally speaking, the larger the number of channel sample information used in training, the better the training effect. The sources of the channel sample information include: 1. The receiving end obtains it based on actual measurements; 2. It is generated using a mathematical (channel) model.

[0177] The details are as follows: 1. The receiving end (such as UE) of the signal (such as a reference signal or a synchronization signal, etc.) measures the actual channel to obtain channel sample information. The channel sample information can more accurately reflect the channel environment. If the network element used to train the neural network using the channel sample information is the transmitting end of the signal, the receiving end measures the channel sample information and feeds back the channel sample information to the transmitting end (such as a base station). The transmitting end trains the neural network based on the channel sample information. In order to improve the training effect and make the trained neural network perform better, the receiving end of the signal needs to feed back a large amount of channel sample information to the transmitting end of the signal. Feedback of a large amount of channel sample information will occupy a large amount of air interface resources and affect the data transmission efficiency between the transmitting and receiving ends.

[0178] 2. In a possible implementation, for different channel types, mathematical expressions can be used to model the channel model. That is, a mathematical (channel) model is used to simulate the real channel. For example, mathematical channel models such as tapped delay line (TDL) and clustered delay line (CDL) can be defined in the protocol. Each channel type can be further divided into multiple subcategories, such as TDL and CDL channels include five subcategories A, B, C, D, and E respectively; each subcategory is further subdivided into multiple typical channel scenarios according to specific parameters, such as 10 nanoseconds (ns), 30ns, 100ns, 300ns or 1000ns according to the multipath delay interval. Therefore, when training the neural network of the transmitter and receiver, a mathematical channel model that is more similar to the actual environment can be selected to generate a large amount of channel sample information that is similar to the actual channel, and the channel sample information can be used for training.

[0179] Although the use of mathematical channel models can greatly reduce the signaling overhead of obtaining channel samples, there is also a disadvantage that the mathematical channel model does not match the actual channel model. For example, the TDL channel model assumes a limited number of reflection paths, and the channel coefficient of each path obeys a simple Rayleigh distribution. However, the number of reflection paths of the actual channel varies in different environments, and the Rayleigh distribution cannot accurately describe the distribution of the channel coefficient of each path. In addition, the multipath delay interval often varies from scene to scene, and the coarse-grained division into several typical values ​​will inevitably cause modeling errors. Therefore, the mathematical channel modeling method is difficult to accurately describe the actual channel model, which in turn affects the training effect of the neural network. That is, there is a problem of data model mismatch.

[0180] In summary, how to generate channel sample information close to the actual channel scenario while saving air interface resources has become a problem that needs to be solved urgently.

[0181] Next, the technical solution proposed in this application is introduced in conjunction with the accompanying drawings. Fig.10 , Fig.10 A flowchart of a neural network training method proposed in an embodiment of the present application. A neural network training method proposed in an embodiment of the present application includes:

[0182] Optionally, 1001. The first device sends a first reference signal to the second device.

[0183] In this embodiment, the first device is an access network device and the second device is a terminal device. It is understandable that the first device can be an access network device or a chip in the access network device, and the second device can be a terminal device or a chip in the terminal device. The first device can also be a terminal device or a chip in the terminal device, and the second device can also be an access network device or a chip in the access network device, which is not limited here.

[0184] The first device sends a first reference signal to the second device, and the first reference signal may be a synchronization signal, a synchronization signal block (synchronization signal and PBCH block, SSB), a demodulation reference signal (demodulation reference signal, DMRS) or a channel state information reference signal (channel state information reference signal, CSI-RS), etc. The first reference signal may also be referred to as the first signal. The first reference signal may also be a newly defined reference signal, which is not limited here. The sequence type of the first reference signal includes but is not limited to: a ZC (zadoff-chu) sequence or a gold (glod) sequence. Optionally, the DMRS may be a DMRS of a physical downlink control channel (physical downlink control channel, PDCCH) or a DMRS of a physical downlink shared channel (physical downlink shared channel, PDSCH).

[0185] Optionally, before step 1001, when the second device needs to use a neural network (the neural network used by the second device is referred to as a third neural network), the second device sends capability information to the first device. The capability information is used to indicate one or more of the following information of the second device:

[0186] 1) Whether it supports the use of neural networks to replace or implement the functions of the communication module;

[0187] The communication module includes but is not limited to: an OFDM modulation module, an OFDM demodulation module, a constellation mapping module, a constellation demapping module, a channel coding module, a channel decoding module, a precoding module, an equalization module, an interleaving module and / or a deinterleaving module.

[0188] 2) whether the network type of the third neural network is supported, or indicating the network type of the supported third neural network;

[0189] The network type of the third neural network includes one or more of the following: fully connected neural network, radial basis neural network, convolutional neural network, recurrent neural network, Hopfield neural network, restricted Boltzmann machine or deep belief network, etc. The third neural network can be any of the above neural networks, and the third neural network can also be a combination of the above multiple neural networks, which is not limited here.

[0190] 3) Whether to support receiving information of the third neural network through signaling;

[0191] 4) a stored preconfigured third neural network;

[0192] For example, if a predefined or preconfigured third neural network already exists in the second device, the identification or index of the third neural network may be carried in the capability information.

[0193] Optionally, the capability information may also carry an identifier or index of another predefined or preconfigured neural network already existing in the second device.

[0194] 5) memory space available for storing a third neural network;

[0195] 6) Information about the computing power available to run the neural network;

[0196] The computing power information refers to the computing power information for running a neural network, such as the computing speed of the processor, and / or the amount of data that the processor can process.

[0197] Optionally, 1002: The second device performs channel estimation according to a first reference signal received from the first device to determine first channel sample information.

[0198] In this embodiment, after receiving the first reference signal, the second device performs channel estimation on the first reference signal to determine first channel sample information.

[0199] Specifically, for ease of description, after the first reference signal is transmitted through the channel, the signal received at the second device is called the second reference signal. Assuming the first reference signal is x and the second reference signal is y, the channel for propagating the first reference signal can be understood as a function with a transition probability P(y|x).

[0200] The second device may pre-configure the information of the first reference signal for the first device through signaling, or the information of the first reference signal may be agreed upon by a protocol, without limitation. In the embodiment of the present application, the type of signaling is not limited, for example, it may be a broadcast message, system information, radio resource control (RRC) signaling, media access control (MAC) control element (CE), or downlink control information (DCI).

[0201] Since the second device already knows the information of the first reference signal x sent by the first device in advance, when the second device receives the second reference signal y, the second device performs channel estimation based on the second reference signal y and the first reference signal x sent, and determines the first channel sample information. At this time, the channel experienced by the first reference signal from being sent to being received can be estimated, such as the amplitude change and phase rotation experienced. The first channel sample information includes but is not limited to the second reference signal and / or channel state information (CSI).

[0202] The channel estimation algorithm used by the second device when performing channel estimation includes but is not limited to: least square estimation method (LS) or linear minimum mean square error estimation method (LMMSE).

[0203] 1003. The second device sends first channel sample information to the first device.

[0204] In this embodiment, the second device sends the first channel sample information to the first device.

[0205] 1004. The first device obtains a first neural network by training according to the first channel sample information.

[0206] In this embodiment, the first device obtains a first neural network by training according to the first channel sample information. The first neural network is used to generate new channel sample information.

[0207] Optionally, the first neural network is a generative neural network. In the embodiment of the present application, the first neural network is described as GAN as an example. It can be understood that the first neural network can be other types of neural networks, such as VAE, etc., which are not limited here. According to the functional division, the first neural network includes a generator network and a discriminator network, wherein the generator network is used to generate second channel sample information, and the discriminator network is used to determine whether the newly generated second channel sample information and the first channel sample information from the second device obey similar probabilistic statistical characteristics. Using machine learning methods, the generator network and the discriminator network of the first neural network are jointly trained, so that the second channel sample information output by the generator network can converge to the probability distribution of the first channel sample information from the second device. The following are examples for explanation.

[0208] Exemplarily, when the first channel sample information is CSI (CSI is denoted as h), the structural diagram of the first neural network can be found in Figure 13-14 . Fig.13 Schematic diagram of the structure of a generator network of the first neural network in an embodiment of the present application. Fig.14Schematic diagram of the structure of a discriminator network of the first neural network in an embodiment of the present application. Figure 13-14 In the first neural network shown, the generator network includes a 5-layer convolutional layer network. The input of the generator network is random noise z, which includes but is not limited to Gaussian white noise. The output of the generator network is the second channel sample information The discriminator network consists of 3 convolutional layers and 3 fully connected layers. Part of the input to the discriminator network is the output information of the generator network. Another part of the input signal of the discriminator network includes the first channel sample information h from the second device. The output c of the discriminator network is a binary variable, which indicates whether the second channel sample information output by the generator network obeys the probability distribution of the first channel sample, that is, Whether it obeys the probability distribution of h.

[0209] Exemplarily, when the first channel sample information is the second reference signal (y) described above, the structural diagram of the first neural network can be found in Figure 15a-15b . Fig.15a This is a schematic diagram of a generator network structure of the first neural network in an embodiment of the present application. Fig.15b Schematic diagram of a discriminator network structure of the first neural network in an embodiment of the present application.

[0210] Figure 15a-15b The generator network in the first neural network shown is composed of a 7-layer convolutional layer network. The input of the generator network includes random noise z and training sequence x, and the output is a sample The discriminator network consists of 4 convolutional layers and 4 fully connected layers, and the input includes samples generated by the generator network. The sample second reference signal y from the second device and the corresponding first reference signal x. The output c of the discriminator network is a binary variable representing Whether it obeys the probability distribution of P(y|x).

[0211] Assuming that the first reference signal sent by the first device is x, and after propagation through the channel, the second reference signal received at the second device is y, then the channel can be understood as a function with a transition probability P(y|x). The first neural network hopes to learn this probability transition feature of the channel, and for each input x, it can generate an output y with a probability of P(y|x), that is, the first neural network simulates the process of signal propagation through the channel.

[0212] Specifically, the first device uses the first channel sample information to train the reference neural network. The neural network obtained after the training is called the first neural network. In the embodiment of the present application, for the convenience of description, the neural network before training with the first channel sample information can be called the reference neural network. Exemplarily, the first device uses the first channel sample information including 5000 samples from the second device to train the reference neural network, thereby training the first neural network.

[0213] In an optional implementation, the reference neural network may preconfigure parameters of some neurons or the structure of the neural network. The reference neural network may also preconfigure information of other neural networks, which is not limited here. For example, the reference neural network may be a neural network trained under a predefined channel environment for generating channel sample information (such as CSI or a second reference signal, etc.). The first device further trains the reference neural network using the first channel sample information to obtain a first neural network.

[0214] In another optional implementation, the reference neural network may be an initialized neural network. For example, the parameters of each neuron in the reference neural network are zero, random values, or other agreed values, and / or the structure of the neural network is a random or agreed initial structure, etc., which are not limited here. The first device starts training the reference neural network using the first channel sample information, thereby training the first neural network.

[0215] 1005. The first device uses the first neural network to infer and obtain second channel sample information.

[0216] In this embodiment, after the first device obtains the trained first neural network, the first neural network is used to infer the second channel sample information.

[0217] Exemplarily, the first device inputs random noise into the first neural network, and the second channel sample information can be inferred by the first neural network. Exemplarily, when the first neural network is a GAN, the random noise is input into the generator network of the first neural network to obtain the second channel sample information. Exemplarily, when the first neural network is a VAE, the random noise is input into the decoder network of the first neural network to obtain the second channel sample information.

[0218] The format of the second channel sample information is the same as that of the first channel sample information. For example, when the first channel sample information is CSI(h), the second channel sample information is When the first channel sample information is the second reference signal (y), the second channel sample information is the second reference signal

[0219] Optionally, 1006, the first device trains the second neural network and / or the third neural network according to the second channel sample information.

[0220] Exemplarily, when the first channel sample information is CSI (denoted as CSI h), a structural diagram of the first device training the second neural network and / or the third neural network according to the second channel sample information is shown in Fig.16a The input of the first neural network (GAN generator network) is Gaussian white noise, and the output is the second channel sample information The second sample information and Gaussian white noise are input to the signal channel module. Fig.16a The signal passes through the channel module, and the transmitting signal and the second sample information The multiplication result is then added to Gaussian white noise, and the result is used as the input of the receiving end. Through machine learning methods, the receiving end neural network (for example: constellation demapping neural network) and the sending end neural network (constellation mapping neural network) can be trained end-to-end.

[0221] Exemplarily, when the first channel sample information is the second reference signal (y) described above, a structural diagram of the first device training the second neural network and / or the third neural network according to the second channel sample information is shown in FIG. Fig.16b The first neural network (GAN generator network) inputs Gaussian white noise and the transmitted signal x, and the output of the first neural network is the second reference signal The first neural network is connected between the receiving end and the transmitting end, and the receiving end neural network (for example: constellation demapping neural network) and the transmitting end neural network (constellation mapping neural network) can be trained end-to-end through machine learning methods.

[0222] In this embodiment, after the first device generates the second channel sample information according to the first neural network, the first device trains the second neural network and / or the third neural network according to the second channel sample information. The second neural network and / or the third neural network are used for the first device to transmit target information with the second device.

[0223] Optionally, the first device trains the second neural network and / or the third neural network based on the first channel sample information and the second channel sample information.

[0224] The second neural network and the third neural network are described below:

[0225] In the embodiment of the present application, the second neural network is applied to the sending end, which can be understood as the second neural network performing the sending processing of the data, for example: Fig. 9The constellation mapping neural network shown. Optionally, the second neural network can perform operations such as rate matching and / or OFDM modulation. Rate matching means that the bits to be sent are repeated and / or punctured to match the carrying capacity of the physical channel so that the bit rate required by the transmission format is achieved during channel mapping. OFDM modulation mainly realizes the movement of the baseband spectrum to the radio frequency band, thereby realizing the wireless transmission function. Exemplarily, the second neural network can perform one or more functions of source coding (such as scrambling, and / or signal compression, etc.), channel coding, constellation mapping, OFDM modulation, precoding and / or filtering.

[0226] The network type of the second neural network includes, but is not limited to: a fully connected neural network, a radial basis function (RBF) neural network, a CNN, a recurrent neural network, a Hopfield neural network, a restricted Boltzmann machine or a deep belief network, etc. The second neural network can be any of the above neural networks, or a combination of the above multiple neural networks, which is not limited here.

[0227] In the embodiment of the present application, for the convenience of description, the first device is used as the transmitting end for illustration, that is, the second neural network is applied in the first device. It is understandable that the second device can also be used as the transmitting end, that is, the second neural network is applied in the second device.

[0228] In the embodiment of the present application, the third neural network is applied to the receiving end, which can be understood as the third neural network receiving and processing the data from the sending end. Fig. 9 Specifically, the third neural network can perform source decoding (such as descrambling and / or signal decompression), channel decoding, constellation demapping, OFDM demodulation, signal detection and / or equalization.

[0229] The network type of the third neural network includes, but is not limited to: a fully connected neural network, a radial basis neural network, a convolutional neural network, a recurrent neural network, a Hopfield neural network, a restricted Boltzmann machine or a deep belief network, etc. The third neural network can be any of the above neural networks, and the third neural network can also be a combination of the above multiple neural networks, which is not limited here.

[0230] It should be noted that the second neural network can use the same neural network as the third neural network, or different neural networks, or partially the same neural network, and there is no limitation here.

[0231] In the embodiment of the present application, for the convenience of description, the second device is used as the receiving end for illustration, that is, the third neural network is applied in the second device. It is understandable that the first device can also be used as the receiving end, that is, the third neural network is applied in the first device.

[0232] The second neural network and / or the third neural network are used to transmit target information between the first device and the second device. Based on the above description, it can be understood that when the second neural network and the third neural network perform different processing, the target information is different.

[0233] Exemplarily, when the second neural network is a constellation mapping neural network and the third neural network is a constellation demapping neural network, the target information is a modulation symbol sequence.

[0234] Exemplarily, when the second neural network is a CSI compression neural network and the third neural network is a CSI decompression neural network, the target information is compressed CSI or CSI.

[0235] Exemplarily, when the second neural network is a channel coding neural network and the third neural network is a channel decoding neural network, the target information is the bit information to be sent before encoding or the bit information to be sent after encoding.

[0236] Exemplarily, when the second neural network is an OFDM modulation neural network and the third neural network is an OFDM demodulation neural network, the target information is an OFDM symbol.

[0237] Exemplarily, when the second neural network is a precoding neural network and the third neural network is a balanced neural network, the target information is a transmission signal, and the transmission signal reaches the receiving end after being transmitted through a channel.

[0238] The first device can train the second neural network alone, the first device can also train the third neural network alone, and the first device can also jointly train the second neural network and the third neural network.

[0239] Exemplarily, when the second neural network is a precoding neural network of the transmitting end (e.g., the first device), the first device can train the second neural network alone. The function of the precoding neural network is to generate the transmission weighting coefficients of each transmitting antenna according to the spatial characteristics of the channel so that multiple transmission data streams can be spatially isolated to improve the received signal-to-interference-noise ratio of the signal. The input of the precoding neural network is the CSI of the channel, and the output is the precoding weights on each transmitting antenna. That is, the target information of the second neural network is the precoding weights on each transmitting antenna.

[0240] Exemplarily, when the third neural network is an equalizer neural network of the receiving end (e.g., the second device), the first device can train the third neural network alone. The function of the equalizer neural network is to offset the distortion effect of the channel on the propagation signal and restore the signal emitted by the transmitting end. The input of the equalizer neural network is the CSI of the channel and the second reference signal of the receiving end, and the output is the signal restored after equalization. That is, the target information of the third neural network is the signal restored after equalization.

[0241] The first device can use the second channel sample information generated by the first neural network to train the second neural network alone, or use the second channel sample information and the locally generated transmission data to train the third neural network alone, and send the trained third neural network information to the second device for use. The first device can also jointly train the second and third neural networks to achieve optimal end-to-end communication performance.

[0242] When the first device completes training of the third neural network, step 1007 may be performed. The completion of the training of the third neural network includes but is not limited to: the number of training samples used in the training reaches a preset threshold, for example, the number of second channel sample information for training the third neural network reaches 45,000, and the third neural network is considered to have been trained. Alternatively, the number of training times reaches a preset threshold, for example, the number of training times of the third neural network reaches 30,000 times, and the third neural network is considered to have been trained.

[0243] Optionally, 1007, the first device sends information of the third neural network to the second device.

[0244] In this embodiment, after the first device completes training of the third neural network, the first device sends information about the third neural network to the second device. The information about the third neural network includes, but is not limited to: weights of the neural network, activation functions of neurons, the number of neurons in each layer of the neural network, the cascade relationship between layers of the neural network, and / or the network type of each layer in the neural network. Exemplarily, the information about the third neural network may indicate the same or different activation functions for different neurons, without limitation.

[0245] Optionally, when the third neural network is preconfigured (or predefined) in the second device, the information of the third neural network may also be the model change of the third neural network. The model change includes, but is not limited to: the weight of the changed neural network, the changed activation function, the number of neurons in each layer of the changed neural network, the changed inter-layer cascade relationship of the neural network, and / or the network type of each layer in the changed neural network. Exemplarily, the preconfiguration may be pre-configuration of the terminal device by the access network device through signaling, and the predefinition may be pre-defined by the protocol, for example: the protocol pre-defines the third neural network in the terminal device as neural network A.

[0246] The first device sends information about the third neural network to the second device through signaling.

[0247] In the embodiment of the present application, signaling includes but is not limited to: broadcast messages (such as master information block (MIB)), system messages (such as system information block (SIB)), radio resource control (RRC) signaling, medium access control control element (MAC CE) and / or downlink control information (DCI). Among them, MAC CE and / or DCI can be a common message for multiple devices, or a specific message specific to a certain device (such as a second device).

[0248] The second device applies the third neural network in the second device according to the information of the third neural network.

[0249] After the first device completes training the second neural network, the first device applies the second neural network.

[0250] When the second neural network is preconfigured (or predefined) in the first device, the first device updates the local second neural network according to the training result of step 1006. When the third neural network is preconfigured in the second device, the second device updates the local third neural network according to the information of the third neural network in step 1007. When there is no third neural network in the second device, the information of the third neural network in step 1007 may include a complete third neural network. The second device locally configures the trained third neural network according to the information of the third neural network.

[0251] Afterwards, if the target is downlink information, the first device may use the second neural network to process the target information (i.e., the input of the second neural network) or obtain the target information (i.e., the output of the second neural network). The second device may use the third neural network to process the target information (i.e., the input of the third neural network) or recover the target information (i.e., the output of the third neural network). Exemplarily, the first device uses the second neural network to obtain the modulation symbol, and the second device uses the third neural network to recover the bit from the received modulation symbol.

[0252] If the target is uplink information, the second device can use the third neural network to process the target information (i.e., the input of the third neural network) or obtain the target information (i.e., the output of the third neural network). The first device can use the second neural network to process the target information (i.e., the input of the second neural network) or restore the target information (i.e., the output of the second neural network). Exemplarily, the second device compresses CSI using the third neural network to obtain compressed CSI; the first device uses the second neural network to restore CSI from the received compressed CSI.

[0253] In an embodiment of the present application, a second device (e.g., a terminal device) may send a relatively small amount of first channel sample information to a first device (e.g., an access network device). The first device trains a first neural network based on the first channel sample information. The first neural network is used to infer the second channel sample information. The first device trains a second neural network and / or a third neural network based on the second channel sample information, or based on the second channel sample information and the first channel sample information, and the second neural network and / or the third neural network are used to transmit target information between the first device and the second device. Through this method, the air interface signaling overhead can be effectively reduced, and at the same time, it can adapt to the channel environment. The trained second neural network and the third neural network are closer to the actual channel environment, thereby improving communication performance. The speed of training the second neural network and the third neural network is also greatly improved.

[0254] Similar to the above Fig.10, the relevant steps of training to obtain the first neural network, the relevant training steps of the second neural network and / or the relevant training steps of the third neural network can be alternatively implemented by other devices independent of the first device. In the embodiment of the present application, for the convenience of description, the other device is referred to as the third device. The third device can be the AI ​​node, mobile edge computing device, or cloud server mentioned above, etc., without limitation. The reference signal pool learned by the third device can be agreed in the protocol after offline learning, or sent to the first device through the interface between the third device and the first device, or forwarded to the first device through other network elements, without limitation. The samples needed by the third device to perform model training, such as the first channel sample information, can be sent directly or indirectly by the first device to the third device, or directly or indirectly by the second device to the third device, without limitation.

[0255] See also Fig.11 , Fig.11 This is another flow chart of a neural network training method proposed in an embodiment of the present application. A neural network training method proposed in an embodiment of the present application includes:

[0256] Optionally, 1101. The first device sends a first reference signal to the second device.

[0257] 1102. The second device performs channel estimation according to the first reference signal received from the first device, and determines first channel sample information.

[0258] Optionally, 1103, the second device sends first channel sample information to the first device.

[0259] Steps 1101-1103 are the same as the aforementioned steps 1001-1003 and are not repeated here.

[0260] 1104. The first device sends first channel sample information to the third device.

[0261] In this embodiment, after the first device receives the first channel sample information from the second device, the first device sends the first channel sample information to the third device.

[0262] 1105. The third device obtains a first neural network by training according to the first channel sample information.

[0263] 1106. The third device uses the first neural network to infer and obtain second channel sample information.

[0264] Optionally, 1106 may be replaced by: the third device sends information of the first neural network to the first device. The first device uses the first neural network to infer the second channel sample information. Optionally, the first device sends the second channel sample information to the third device.

[0265] 1107. The third device trains the second neural network and / or the third neural network according to the second channel sample information.

[0266] Steps 1105-1107 are similar to the aforementioned steps 1004-1006 and are not described in detail here.

[0267] 1108. The third device sends the information of the second neural network and / or the information of the third neural network to the first device.

[0268] In this embodiment, after the third device trains the third neural network, the third device can send information about the third neural network to the second device through the first device. Optionally, the third device can also send information about the third neural network to the first device through other devices (such as other access network devices). Optionally, if there is a direct communication link between the third device and the first device, the third device can also directly send the information about the third neural network to the first device.

[0269] After the third device obtains the second neural network through training, the third device may send information of the second neural network to the second device.

[0270] 1109. The first device sends information of the third neural network to the second device.

[0271] Step 1109 is the same as the aforementioned step 1007 and will not be repeated here.

[0272] Optional, Fig.11 The embodiment shown can be replaced by: the subject used to train the first neural network is different from the subject used to train the second neural network and / or the subject used to train the second neural network. For example, the former is the first device and the latter is the third device; or the former is the third device and the latter is the first device.

[0273] In the embodiment of the present application, the neural network is trained in the above manner, which effectively reduces the air interface signaling overhead, and can adapt to the channel environment. The trained second neural network and the third neural network are closer to the actual channel environment, and the communication performance is improved. The speed of training the second neural network and the third neural network is also greatly improved. The steps involving neural network training can be performed by other devices (third devices), effectively reducing the computing load and power consumption of the first device (such as access network equipment).

[0274] Similar to the above Fig.10 or Fig.11, the relevant steps of training the first neural network can alternatively be implemented by a second device (such as a terminal device).

[0275] See also Fig.12 , Fig.12 A flowchart of another method for training a neural network proposed in an embodiment of the present application. In this method, a first neural network is trained by a second device. A method for training a neural network proposed in an embodiment of the present application includes:

[0276] Optionally, 1201, the second device sends capability information to the first device.

[0277] In this embodiment, Figure 10-11 The embodiments shown are similar, and the first device is an access network device, and the second device is a terminal device. The first device can be an access network device, a chip in an access network device, a module or circuit in an access network device, etc., and the second device can be a terminal device, a chip in a terminal device, a module or circuit in a terminal, etc. The first device can also be a terminal device, a chip in a terminal device, a module or circuit in a terminal, etc., and the second device can also be an access network device or a chip in an access network device, a module or circuit in an access network device, etc., and there is no limitation here.

[0278] In the embodiment of the present application, for ease of description, the first neural network that is not trained using the first channel sample information or the first neural network before updating may be referred to as a reference neural network. After the reference neural network is trained using the first channel sample information, the first neural network is obtained.

[0279] The reference neural network is similar to the reference neural network in the aforementioned step 1004. The second device starts training the reference neural network using the first channel sample information, thereby training to obtain the first neural network.

[0280] In step 1201, the second device sends capability information to the first device, where the capability information is used to indicate one or more of the following information:

[0281] 1) Whether it supports the use of neural networks to replace or implement the functions of the communication module;

[0282] The communication module includes but is not limited to: an OFDM modulation module, an OFDM demodulation module, a constellation mapping module, a constellation demapping module, a channel coding module, a channel decoding module, a precoding module, an equalization module, an interleaving module and / or a deinterleaving module.

[0283] 2) whether the network type of the first neural network is supported, or the network type of the first neural network that is supported;

[0284] For example, the capability information indicates whether the second device supports VAE or GAN. For another example, the capability information indicates whether the second device supports GAN, supports VAE, supports GAN and VAE, or supports neither GAN nor VAE.

[0285] 3) whether the network type of the third neural network is supported, or indicating the network type of the supported third neural network;

[0286] The network type of the third neural network includes one or more of the following: fully connected neural network, radial basis neural network, convolutional neural network, recurrent neural network, Hopfield neural network, restricted Boltzmann machine or deep belief network, etc. The third neural network can be any of the above neural networks, and the third neural network can also be a combination of the above multiple neural networks, which is not limited here.

[0287] 4) whether it supports receiving information of a reference neural network through signaling, where the reference neural network is a first neural network that has not been trained with the first channel sample information or is used to train the first neural network;

[0288] 5) Whether to support receiving information of the third neural network through signaling;

[0289] 6) stored pre-configured reference neural networks;

[0290] For example, if a predefined or preconfigured reference neural network already exists in the second device, the identification or index of the reference neural network may be carried in the capability information.

[0291] Optionally, the capability information may also carry an identifier or index of another predefined or preconfigured neural network already existing in the second device.

[0292] 7) memory space that can be used to store the first neural network and / or the third neural network;

[0293] 8) Information about the computing power available to run the neural network;

[0294] The computing power information refers to the computing power information for running a neural network, such as the computing speed of the processor, and / or the amount of data that the processor can process.

[0295] 9) Location information of the second device.

[0296] For example, the longitude, latitude, and / or altitude of the second device is reported. For another example, the channel environment of the second device is reported, such as an office, a subway station, a park, or a street.

[0297] The capability information includes but is not limited to the above information. For example, the capability information may also include: the type of channel sample information supported by the second device for training the first neural network. Exemplary: the channel sample information supported by the second device (terminal device) for training the first neural network is CSI, or the channel sample information supported by the second device for training the first neural network is the second reference signal (y).

[0298] Step 1201 is an optional step. When step 1201 is executed, the second device may send the capability information when it first accesses the first device, or the second device may periodically send the capability information to the first device, which is not limited here.

[0299] Optionally, after the first device receives the capability information from the second device, when the capability information indicates that the reference neural network is not configured (not stored) in the second device. The first device may send the information of the reference neural network to the second device. The information of the reference neural network includes: the weights of the neural network, the activation function of the neurons, the number of neurons in each layer of the neural network, the cascade relationship between the layers of the neural network, and / or the network type of each layer in the neural network. The information of the reference neural network may be referred to as the specific information of the reference neural network. Among them, the activation functions of different neurons may be the same or different, without limitation.

[0300] Optionally, 1202, the first device sends a first reference signal to the second device.

[0301] Same as the above step 1001, which will not be described again here.

[0302] Optionally, 1203: The second device performs channel estimation according to the first reference signal received from the first device to determine first channel sample information.

[0303] Same as step 1002, which will not be described again here.

[0304] 1204. The second device trains a first neural network according to the first channel sample information.

[0305] The method for the second device to obtain the first neural network by training according to the first channel sample information is similar to the method for the first device to obtain the first neural network by training according to the first channel sample information in 1004. It will not be described in detail here.

[0306] Exemplarily, the second device trains the first neural network based on the first channel sample information and the reference neural network.

[0307] Optionally, the second device may determine the reference neural network according to any one of the following methods 1 to 3:

[0308] Method 1: The information of the reference neural network is agreed upon by the protocol, or the second device determines the information of the reference neural network.

[0309] In a possible implementation, the protocol agrees on (predefines) information of a reference neural network, that is, agrees on specific information of the reference neural network.

[0310] In the embodiment of the present application, the information of the reference neural network includes: the weight of the neural network, the neuron activation function, the number of neurons in each layer of the neural network, the inter-layer cascade relationship of the neural network, and / or the network type of each layer in the neural network. The information of the reference neural network can also be referred to as the specific information of the reference neural network. Among them, the activation functions of different neurons can be the same or different, without limitation.

[0311] In one possible implementation, the protocol stipulates information of multiple reference neural networks. Each reference neural network corresponds to a channel environment. The channel environment includes one or more of the following: offices, shopping malls, subway stations, city streets, squares, and suburbs. As mentioned above, the second device reports the location information of the second device to the first device, and the first device can determine the channel environment where the second device is located, thereby obtaining the reference neural network corresponding to the channel environment. For example, the access network device can maintain a map information, and the channel environment corresponding to different location information marks in the map information. For example: The map information is shown in Table 1:

[0312] Table 1

[0313] Location information / {longitude, latitude, altitude} Channel environment {x1, y1, z1} office {x2, y2, z2} metro station {x3, y3, z3} garden {x4, y4, z4} street

[0314] Optionally, if the second device reports the location information of the second device, that is, the channel environment of the second device, to the first device through capability information, the first device can determine the information of the reference neural network corresponding to the channel environment.

[0315] In a possible implementation, the second device determines the information of the reference neural network, that is, determines the specific information of the reference neural network. As described above, the second device can report the information of the reference neural network to the first device through capability information.

[0316] Method 2: The reference neural network is indicated by the first device to the second device.

[0317] In one possible implementation, the first device sends information of the reference neural network to the second device, that is, the first device sends specific information of the reference neural network to the second device.

[0318] In one possible implementation, the protocol stipulates information of multiple candidate reference neural networks, each candidate reference neural network corresponds to a unique index. The first device sends an identifier to the second device through signaling, and the identifier is used to indicate the index of the reference neural network configured for the second device among the multiple candidate reference neural networks.

[0319] In a possible implementation, the first device configures information of multiple candidate reference neural networks for the second device through a first signaling, that is, configures specific information of the multiple candidate reference neural networks. Each candidate reference neural network corresponds to a unique index. The first device sends an identifier to the second device through a second signaling, and the identifier is used to indicate the index of the reference neural network configured for the second device among the multiple candidate reference neural networks.

[0320] In a possible implementation, as described above, the second device may send information of multiple candidate reference neural networks stored or possessed by the second device to the first device through capability information. Each candidate reference neural network corresponds to a unique index. The first device sends an identifier to the second device through signaling, and the identifier is used to indicate the index of the reference neural network configured for the second device among the multiple candidate reference neural networks.

[0321] In a possible implementation, the first device sends the type of the reference neural network to the second device through signaling. The protocol stipulates the information of various types of reference neural networks, or the second device autonomously determines the information of various types of reference neural networks.

[0322] 1205. The second device sends information of the first neural network to the first device.

[0323] After the second device obtains the first neural network through training, the second device sends information of the first neural network to the first device.

[0324] Optionally, the information of the first neural network may be specific information of the first neural network. The specific information may include: weights of the neural network, activation functions of neurons, the number of neurons in each layer of the neural network, the cascade relationship between layers of the neural network, and / or the network type of each layer in the neural network. The activation functions of different neurons may be the same or different, without limitation.

[0325] Optionally, the information of the first neural network may also be a model change of the first neural network relative to a reference neural network. The model change includes, but is not limited to: a changed weight of the neural network, a changed activation function, a changed number of neurons in one or more layers of the neural network, a changed inter-layer cascade relationship of the neural network, and / or a changed network type of one or more layers of the neural network. When the information of the first neural network is a model change, the overhead of air interface signaling can be effectively reduced.

[0326] 1206. The first device determines the first neural network according to the information of the first neural network.

[0327] The first device determines the first neural network based on the information of the first neural network from the second device.

[0328] Optionally, when the information of the first neural network is specific information of the first neural network, the first device determines the first neural network according to the specific information.

[0329] Optionally, when the information of the first neural network is a model change, the first device determines the first neural network based on the model change and a reference neural network.

[0330] 1207. The first device obtains second channel sample information according to the first neural network inference.

[0331] 1208. The first device trains the second neural network and / or the third neural network according to the second channel sample information.

[0332] 1209. The first device sends information of the third neural network to the second device.

[0333] Steps 1207-1209 are the same as steps 1005-1007, and will not be repeated here.

[0334] In this embodiment, the second device trains the first neural network according to the first channel sample information and the reference neural network.

[0335] In this embodiment, the second device (e.g., a terminal device) trains a first neural network based on the first channel sample information. The first neural network is used to infer the second channel sample information. The first device trains a second neural network and / or a third neural network based on the second channel sample information, and the second neural network and / or the third neural network are used to transmit target information between the first device and the second device. It effectively reduces the air interface signaling overhead and can adapt to the channel environment. The trained second neural network and the third neural network are closer to the actual channel environment, thereby improving communication performance. The speed of training the second neural network and the third neural network is also greatly improved. The computing task of training the first neural network is assigned to the second device, which effectively reduces the computing burden of the first device (e.g., an access network device) and improves communication performance.

[0336] In the embodiments provided by the present application, the method provided by the embodiments of the present application is introduced from the perspective of the first device, the second device, the third device, and the interaction therebetween. In order to realize the functions in the method provided by the embodiments of the present application, the first device, the second device, and the third device may include a hardware structure and / or a software module, and the functions are realized in the form of a hardware structure, a software module, or a hardware structure plus a software module. Whether a function in the above 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.

[0337] See also Fig.17 , Fig.17 A schematic diagram of a communication device provided in an embodiment of the present application is provided. The communication device 1700 includes a transceiver module 1710 and a processing module 1720, wherein:

[0338] Optionally, the communication device 1700 is used to implement the function of the first device in the above method embodiment.

[0339] In a possible implementation, the transceiver module 1710 is configured to send a first reference signal to a second device;

[0340] The transceiver module 1710 is further configured to receive first channel sample information from a second device;

[0341] Processing module 1720 is used to determine a first neural network, where the first neural network is trained based on first channel sample information, the first neural network is used to infer second channel sample information, the second channel sample information is used to train a second neural network and / or a third neural network, and the second neural network and / or the third neural network is used for the first device to transmit target information with the second device.

[0342] In a possible implementation, the transceiver module 1710 is configured to send a first reference signal to a second device;

[0343] The transceiver module 1710 is further configured to receive information of the first neural network from the second device;

[0344] Processing module 1720 is used to determine a second neural network and / or a third neural network, wherein the second neural network and / or the third neural network is used for transmitting target information between the first device and the second device, and the second neural network and / or the third neural network is trained based on the second channel sample information, and the second channel sample information is inferred by the first neural network.

[0345] Optionally, the communication device 1700 is used to implement the function of the second device in the above method embodiment.

[0346] In a possible implementation, the processing module 1720 is configured to perform channel estimation according to a first reference signal received from the first device to determine first channel sample information;

[0347] The transceiver module 1710 is configured to send first channel sample information to the first device;

[0348] The transceiver module 1710 is also used to receive information of the third neural network from the first device, and the third neural network is used to transmit target information between the second device and the first device.

[0349] In a possible implementation, the processing module 1720 is configured to perform channel estimation according to a first reference signal received from the first device to determine first channel sample information;

[0350] The processing module 1720 is further configured to determine a first neural network, where the first neural network is trained according to the first channel sample information;

[0351] The transceiver module 1710 is used to send the information of the first neural network to the first device.

[0352] Optionally, the communication device 1700 is used to implement the function of the third device in the above method embodiment.

[0353] In a possible implementation, the transceiver module 1710 is configured to receive first channel sample information from a first device;

[0354] Processing module 1720 is used to determine a first neural network, where the first neural network is trained based on first channel sample information, the first neural network is used to infer second channel sample information, the second channel sample information is used to train a second neural network and / or a third neural network, and the second neural network and / or the third neural network is used for the first device to transmit target information with the second device.

[0355] Optionally, the above-mentioned communication device may also include a storage unit for storing data and / or instructions (also referred to as codes or programs). The above-mentioned various units may interact or couple with the storage unit to implement the corresponding method or function. For example, the processing unit 1720 may read the data or instructions in the storage unit so that the communication device implements the method in the above-mentioned embodiment. The coupling in the embodiment of the present application is an indirect coupling or communication connection between devices, units or modules, which may be electrical, mechanical or other forms, and is used for information exchange between devices, units or modules.

[0356] The division of modules in the embodiments of the present application is schematic and is only a logical function division. There may be other division methods in actual implementation. In addition, each functional module in each embodiment of the present application may be integrated into a processor, or may exist physically separately, or two or more modules may be integrated into one module. The above-mentioned integrated modules may be implemented in the form of hardware or in the form of software functional modules.

[0357] In one example, the unit in any of the above communication devices may be one or more integrated circuits configured to implement the above method, such as: one or more application specific integrated circuits (ASIC), or one or more microprocessors (digital singnal processors, DSP), or one or more field programmable gate arrays (FPGA), or a combination of at least two of these integrated circuit forms. For another example, when the unit in the communication device can be implemented in the form of a processing element scheduler, the processing element can be a general-purpose processor, such as a central processing unit (CPU) or other processor that can call a program. For another example, these units can be integrated together and implemented in the form of a system-on-a-chip (SOC).

[0358] The technical solution provided in the embodiment of the present application can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present invention is generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, an AI node, an access network device, a terminal device or other programmable device. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions may be transmitted from one website site, computer, server or data center to another website site, computer, server or data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium may be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more available media integrated. The available medium may be a magnetic medium (eg, a floppy disk, a hard disk, a magnetic tape), an optical medium (eg, a digital video disc (DVD)), or a semiconductor medium, etc.

[0359] In the embodiments of the present application, under the premise of no logical contradiction, the embodiments may reference each other, for example, the methods and / or terms between method embodiments may reference each other, for example, the functions and / or terms between device embodiments may reference each other, for example, the functions and / or terms between device embodiments and method embodiments may reference each other.

[0360] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.

Claims

1. A neural network training method, characterized in that: include: The first device sends a first reference signal to the second device; The first device receives first channel sample information from the second device; The first device determines a first neural network, the first neural network is trained based on the first channel sample information, the first neural network is used to infer second channel sample information, the second channel sample information is used to train a second neural network and / or a third neural network, and the second neural network and / or the third neural network are used for the first device to transmit target information with the second device.

2. The method according to claim 1, characterized in that The first channel sample information is also used to train the second neural network and / or the third neural network.

3. The method according to claim 1, characterized in that The method further comprises: The first device sends information of the third neural network to the second device.

4. The method according to claim 1, characterized in that: The first channel sample information includes channel state information CSI and / or a second reference signal, and the second reference signal is the first reference signal propagated through a channel.

5. The method according to claim 1, characterized in that The first neural network is a generative adversarial network or a variational autoencoder.

6. The method according to claim 1, characterized in that The first reference signal includes a demodulation reference signal DMRS or a channel state information reference signal CSI-RS.

7. The method according to any one of claims 1 to 6, characterized in that The sequence type of the first reference signal includes a ZC sequence or a golden sequence.

8. A method for training a neural network, characterized in that: include: The second device performs channel estimation according to the first reference signal received from the first device to determine first channel sample information; The second device sends the first channel sample information to the first device, where the first channel sample information is used to train a third neural network; The second device receives information of the third neural network from the first device, and the third neural network is used for the second device to transmit target information with the first device.

9. The method according to claim 8, characterized in that The first channel sample information includes channel state information CSI and / or a second reference signal, and the second reference signal is a first reference signal received by the second device.

10. The method according to claim 8, characterized in that The third neural network includes one or more of the following: a fully connected neural network, a radial basis neural network, a convolutional neural network, a recurrent neural network, a Hopfield neural network, a restricted Boltzmann machine or a deep belief network.

11. The method according to claim 8, characterized in that The first reference signal includes a demodulation reference signal DMRS or a channel state information reference signal CSI-RS.

12. The method according to any one of claims 8 to 11, characterized in that The sequence type of the first reference signal includes a ZC sequence or a golden sequence.

13. A method for training a neural network, characterized in that: include: The first device sends a first reference signal to the second device; The first device receives information of the first neural network from the second device; The first device determines a second neural network and / or a third neural network, the second neural network and / or the third neural network are used for transmitting target information between the first device and the second device, the second neural network and / or the third neural network are trained based on second channel sample information, and the second channel sample information is inferred by the first neural network.

14. The method according to claim 13, characterized in that The information of the first neural network includes: a model change amount of the first neural network relative to a reference neural network, and the reference neural network is used to train the first neural network.

15. The method according to claim 13, characterized in that The information of the first neural network includes one or more of the following: weights of the neural network, activation functions of neurons, the number of neurons in each layer of the neural network, the cascade relationship between layers of the neural network, and the network type of each layer in the neural network.

16. The method according to claim 13, characterized in that The method further comprises: The first device sends information of the third neural network to the second device.

17. The method according to claim 13, characterized in that The method further comprises: The first device receives capability information from the second device, where the capability information is used to indicate one or more of the following information of the second device: Whether it supports the use of neural networks to replace or implement the functions of the communication module; Whether the network type of the first neural network is supported; whether the network type of the third neural network is supported; whether to support receiving information of a reference neural network through signaling, where the reference neural network is used to train the first neural network; whether to support receiving information of the third neural network through signaling; the stored reference neural network; A memory space that can be used to store the first neural network and / or the third neural network; Information about the amount of computing power available to run the neural network; and, The location information of the second device.

18. The method according to claim 13, characterized in that The first neural network is a generative adversarial network or a variational autoencoder.

19. The method according to claim 13, characterized in that The first reference signal includes a demodulation reference signal DMRS or a channel state information reference signal CSI-RS.

20. The method according to any one of claims 13 to 19, characterized in that The sequence type of the first reference signal includes a ZC sequence or a golden sequence.

21. A method for training a neural network, characterized in that: The method is applied to a second device, where the second device is a terminal device, and the method includes: The second device performs channel estimation according to the first reference signal received from the first device to determine first channel sample information; The second device determines a first neural network, where the first neural network is trained according to the first channel sample information; The second device sends information of the first neural network to the first device.

22. The method according to claim 21, characterized in that The information of the first neural network includes: a model change amount of the first neural network relative to a reference neural network, and the reference neural network is used to train the first neural network.

23. The method according to claim 21, characterized in that The information of the first neural network includes one or more of the following: weights of the neural network, activation functions of neurons, the number of neurons in each layer of the neural network, the cascade relationship between layers of the neural network, and / or the network type of each layer in the neural network.

24. The method according to claim 21, characterized in that The method further comprises: The second device receives information of a third neural network from the first device.

25. The method according to claim 21, characterized in that The method further comprises: The second device sends capability information to the first device, where the capability information is used to indicate one or more of the following information of the second device: Whether it supports the use of neural networks to replace or implement the functions of the communication module; Whether the network type of the first neural network is supported; Whether the network type of the third neural network is supported; whether to support receiving information of a reference neural network through signaling, where the reference neural network is used to train the first neural network; whether to support receiving information of the third neural network through signaling; the stored reference neural network; A memory space that can be used to store the first neural network and / or the third neural network; Information about the amount of computing power available to run the neural network; and, The location information of the second device.

26. The method according to claim 21, characterized in that The first neural network is a generative adversarial network or a variational autoencoder.

27. The method according to claim 21, characterized in that The first reference signal includes a demodulation reference signal DMRS or a channel state information reference signal CSI-RS.

28. The method according to any one of claims 21 to 27, characterized in that The sequence type of the first reference signal includes a ZC sequence or a golden sequence.

29. A neural network training method, characterized in that: include: The third device receives the first channel sample information from the first device; The third device obtains a first neural network trained according to the first channel sample information, the first neural network is used to infer second channel sample information, the second channel sample information is used to train a second neural network and / or a third neural network, and the second neural network and / or the third neural network are used for the first device to transmit target information with the second device.

30. A communication device, characterized in that: It comprises a processing module and a transceiver module for implementing the method described in any one of claims 1 to 7.

31. A communication device, characterized in that: The method comprises a processor and a memory, wherein the memory is coupled to the processor, and the processor is used to execute the method according to any one of claims 1 to 7.

32. A communication device, characterized in that: It comprises a processing module and a transceiver module for implementing the method described in any one of claims 8 to 12.

33. A communication device, characterized in that: The method comprises a processor and a memory, wherein the memory is coupled to the processor, and the processor is used to execute the method according to any one of claims 8 to 12.

34. A communication device, characterized in that: It comprises a processing module and a transceiver module for implementing the method described in claim 29.

35. A communication device, characterized in that: The method comprises a processor and a memory, wherein the memory is coupled to the processor, and the processor is configured to execute the method of claim 29.

36. A communication device, characterized in that: It comprises a processing module and a transceiver module for implementing the method described in any one of claims 13-20.

37. A communication device, characterized in that: The method comprises a processor and a memory, wherein the memory is coupled to the processor, and the processor is used to execute the method according to any one of claims 13 to 20.

38. A communication device, characterized in that: It comprises a processing module and a transceiver module for implementing the method described in any one of claims 21-28.

39. A communication device, characterized in that: The method comprises a processor and a memory, wherein the memory is coupled to the processor, and the processor is used to execute the method according to any one of claims 21 to 28.

40. A communication system, characterized in that: include: The communication device according to claim 30 or 31, and the communication device according to claim 32 or 33; The communication device according to claim 30 or 31, the communication device according to claim 32 or 33, and the communication device according to claim 34 or 35; or, The communication device according to claim 36 or 37, and the communication device according to claim 38 or 39.

41. A computer-readable storage medium, characterized in that The computer-readable storage medium stores instructions, and when the instructions are executed on a computer, the computer executes the method according to any one of claims 1 to 29.

42. A computer program product, characterized in that The method comprises instructions which, when executed on a computer, cause the computer to execute the method according to any one of claims 1 to 29.

Images