Neural network training method and communication device

Abstract

The invention provides a neural network training method and a communication device. The method may comprise: a first device sending a first signal; the second device determines a second signal and an intermediate gradient and sends the second signal and the intermediate gradient to the first device, and the second signal is related to the first signal; and the first device receives the intermediate gradient and the second signal, corrects the intermediate gradient according to the first signal and the second signal, and updates neural network parameters according to the corrected intermediate gradient. Therefore, the training of the neural network is more efficient, and the training of the neural network can be realized by using the method provided by the invention even in a communication scene with poor channel reciprocity.

Description

Neural network training method and communication device Technical Field

[0001] The present application relates to the field of communication technology, and more particularly, to a method and a communication device for training a neural network. Background Art

[0002] The growing maturity of artificial intelligence (AI) technology will significantly drive the evolution of future mobile communication network technologies. Currently, extensive research has been conducted on applying AI technology to both the network layer (e.g., network optimization, mobility management, resource allocation, etc.) and the physical layer (e.g., channel coding, channel prediction, receivers, etc.).

[0003] With the advent of the era of large models, some AI models are able to complete increasingly complex tasks and achieve high performance. However, large models with massive parameters require significant computing resources for both training and inference. Conventional communication equipment (such as terminals) struggles to handle the training or inference of such models. Therefore, wireless communications require significant attention to providing diverse support capabilities for AI models, enabling more efficient training and inference.

[0004] Summary of the Invention

[0005] The present application provides a method and communication device for training a neural network, which not only makes the training of the neural network more efficient, but also is applicable to more communication scenarios.

[0006] In a first aspect, a method for training a neural network is provided, which can be performed by an apparatus. The apparatus can be a device (such as a terminal device, a network device, or an AI node), or a component of a device (such as a chip or circuit), which is not limited in this application.

[0007] The method may include: sending a first signal; receiving a first intermediate gradient and a second signal, wherein the second signal is related to the first signal; updating the first intermediate gradient according to the first signal and the second signal to obtain a second intermediate gradient, wherein the second intermediate gradient is used to update the neural network parameters.

[0008] Based on the above technical solution, a device can determine the channel reciprocity error based on the signal it sends (i.e., the first signal) and the signal it receives (i.e., the second signal), as well as the relationship between the two signals, and then can update the received intermediate gradient (i.e., the first intermediate gradient) based on the channel reciprocity error, and then update the neural network parameters based on the updated intermediate gradient (i.e., the second intermediate gradient). In this way, the gradient error caused by channel reciprocity can be overcome, making the training of the neural network more efficient. In addition, for scenarios where channel reciprocity is poor, this technical solution can also be used to implement neural network training, so the above technical solution is applicable to more scenarios.

[0009] In combination with the first aspect, in some implementations of the first aspect, the first signal is any one of the following: a custom sequence, a reference signal, or a data signal.

[0010] The custom sequence may be, for example, a pseudo-random sequence, and may also be referred to as a preset sequence or sequence.

[0011] In combination with the first aspect, in certain implementations of the first aspect, the first signal is a reference signal, and the first intermediate gradient is determined based on the first signal.

[0012] Based on the above technical solution, if the first signal is a reference signal, the opposite device, upon receiving the reference signal, can determine the intermediate gradient based on the reference signal. For example, the opposite device can perform channel estimation based on the reference signal and, based on the channel estimation result, compensate or correct the intermediate gradient before sending it.

[0013] In combination with the first aspect, in some implementations of the first aspect, the first intermediate gradient is determined according to the first signal and / or the data signal.

[0014] In one example, the first intermediate gradient is determined according to the first signal. For example, when the first signal is a data signal, the first intermediate gradient is determined according to the first signal.

[0015] In another example, the first intermediate gradient is determined according to the data signal. For example, when the first signal is a data signal or a custom sequence, the first intermediate gradient is determined according to the data signal.

[0016] In another example, the first intermediate gradient is determined based on the first signal and the data signal. For example, when the first signal is a reference signal, the first intermediate gradient is determined based on the first signal and the data signal.

[0017] In combination with the first aspect, in certain implementations of the first aspect, updating the first intermediate gradient according to the first signal and the second signal includes: determining a channel reciprocity error value according to the first signal and the second signal; and updating the first intermediate gradient according to the channel reciprocity error value.

[0018] Based on the above technical solution, by determining the channel reciprocity error value and correcting the intermediate gradient according to the error value, the gradient error caused by the channel reciprocity can be overcome.

[0019] In combination with the first aspect, in some implementations of the first aspect, the method further includes: sending first indication information, where the first indication information indicates a type of the first signal and / or the second signal.

[0020] Based on the above technical solution, considering that there may be multiple types of the first signal and the second signal, such as different types for different scenarios, the type of the first signal and / or the second signal can be indicated to the opposite device, so that the opposite device can know the type of the first signal and the second signal, and then accurately receive the first signal and send the second signal.

[0021] In combination with the first aspect, in certain implementations of the first aspect, before receiving the first intermediate gradient and the second signal, the method further includes: receiving second indication information, where the second indication information indicates a pattern of the second signal.

[0022] Based on the above technical solution, the opposite device can provide a pattern of the second signal, which facilitates the device to accurately receive the second signal based on the pattern of the second signal.

[0023] In combination with the first aspect, in some implementations of the first aspect, the pattern of the second signal includes at least one of the following: a mask of the second signal, resources occupied by the second signal, and a sequence number of the second signal.

[0024] Based on the above technical solution, the opposite device can indicate the pattern of the second signal by sending at least one of the above items.

[0025] In a second aspect, a method for training a neural network is provided, which can be performed by an apparatus. The apparatus can be a device (such as a terminal device, a network device, or an AI node), or a component of a device (such as a chip or circuit), which is not limited in this application.

[0026] The method may include: receiving a first signal; sending a first intermediate gradient and a second signal, the second signal being related to the first signal, the second signal being used to update the first intermediate gradient, and the first intermediate gradient being used to update neural network parameters.

[0027] In combination with the second aspect, in some implementations of the second aspect, the first signal is any one of the following: a custom sequence, a reference signal, or a data signal.

[0028] In combination with the second aspect, in some implementations of the second aspect, the first intermediate gradient is determined based on the first signal and / or the data signal.

[0029] In one example, the first intermediate gradient is determined according to the first signal. For example, when the first signal is a data signal, the first intermediate gradient is determined according to the first signal.

[0030] In another example, the first intermediate gradient is determined according to the data signal. For example, when the first signal is a data signal or a custom sequence, the first intermediate gradient is determined according to the data signal.

[0031] In another example, the first intermediate gradient is determined based on the first signal and the data signal. For example, when the first signal is a reference signal, the first intermediate gradient is determined based on the first signal and the data signal.

[0032] For example, determining a first intermediate gradient according to a data signal; performing channel estimation according to the first signal and updating the first intermediate gradient according to a result of the channel estimation; and sending the first intermediate gradient includes: sending the updated first intermediate gradient.

[0033] In combination with the second aspect, in some implementations of the second aspect, the method further includes: receiving first indication information, where the first indication information indicates a type of the first signal and / or the second signal.

[0034] In combination with the second aspect, in some implementations of the second aspect, the method further includes: sending second indication information, where the second indication information indicates a pattern of the second signal.

[0035] In combination with the second aspect, in some implementations of the second aspect, the pattern of the second signal includes at least one of the following: a mask of the second signal, resources occupied by the second signal, and a sequence number of the second signal.

[0036] The beneficial effects of the second aspect and each possible design can be referred to the relevant description of the first aspect and will not be repeated here.

[0037] In combination with the first aspect or the second aspect, in some implementations, the second signal and the first signal satisfy:

[0038] A2=f(A1), or A2=f(A1 * )

[0039] Among them, A1 represents the first signal, A2 represents the second signal, A1 * Indicates the conjugate operation on A1.

[0040] For example, the above relationship may be predefined, or may be preconfigured, or indicated.

[0041] Based on the above technical solution, the first signal and the second signal can satisfy the above relationship, so that the opposite device can determine the second signal based on the relationship.

[0042] In combination with the first aspect or the second aspect, in some implementations, the second signal and the first signal satisfy any of the following:

[0043] A2=αA1,A2=αA1 * , or,

[0044] Among them, A1 represents the first signal, A2 represents the second signal, A1 * represents the conjugate operation on A1, |A1| represents the independent normalization of the power of each symbol in A1, ||A1||2 represents the normalization of the power of some or all symbols in A1, and α is a constant.

[0045] For example, the above relationship may be predefined, or may be preconfigured, or indicated.

[0046] In a third aspect, a method for training a neural network is provided, which can be performed by an apparatus. The apparatus can be a device (such as a terminal device, a network device, or an AI node), or a component of a device (such as a chip or circuit), which is not limited in this application.

[0047] The method may include: sending a first reference signal; receiving a first intermediate gradient and a second reference signal, wherein the first intermediate gradient is determined based on the first reference signal; performing channel estimation based on the second reference signal, and updating the first intermediate gradient based on the channel estimation result to obtain a second intermediate gradient, wherein the second intermediate gradient is used to update the neural network parameters.

[0048] Based on the above technical solution, the opposite device can perform channel estimation based on the reference signal sent by the device, and then determine the intermediate gradient based on the result of the channel estimation, and send the intermediate gradient to the device; the device can perform channel estimation based on the reference signal sent by the opposite device, and then update the received intermediate gradient based on the result of the channel estimation, so that the intermediate gradient takes into account the channel reciprocity error, thereby overcoming the gradient error caused by channel reciprocity.

[0049] In combination with the third aspect, in some implementations of the third aspect, the method further includes: sending first indication information, where the first indication information indicates a type of the first reference signal and / or the second reference signal.

[0050] In combination with the third aspect, in certain implementations of the third aspect, before receiving the first intermediate gradient and the second reference signal, the method further includes: receiving second indication information, where the second indication information indicates a pattern of the second reference signal.

[0051] In combination with the third aspect, in certain implementations of the third aspect, the pattern of the second reference signal includes at least one of the following: a mask of the second reference signal, resources occupied by the second reference signal, and a sequence number of the second reference signal.

[0052] In a fourth aspect, a method for training a neural network is provided, which can be performed by an apparatus. The apparatus can be a device (such as a terminal device, a network device, or an AI node), or a component of a device (such as a chip or circuit), which is not limited in this application.

[0053] The method may include: receiving a first reference signal; performing channel estimation based on the first reference signal, and determining a first intermediate gradient based on the result of the channel estimation; sending the first intermediate gradient and a second reference signal, the second reference signal being used to update the first intermediate gradient, and the first intermediate gradient being used to update neural network parameters.

[0054] In combination with the fourth aspect, in some implementations of the fourth aspect, the method further includes: receiving first indication information, where the first indication information indicates a type of the first reference signal and / or the second reference signal.

[0055] In combination with the fourth aspect, in certain implementations of the fourth aspect, the method further includes: sending second indication information, where the second indication information indicates a pattern of the second reference signal.

[0056] In combination with the fourth aspect, in certain implementations of the fourth aspect, the pattern of the second reference signal includes at least one of the following: a mask of the second reference signal, resources occupied by the second reference signal, and a sequence number of the second reference signal.

[0057] The beneficial effects of the fourth aspect and each possible design can be referred to the relevant description of the third aspect and will not be repeated here.

[0058] In a fifth aspect, a method for training a neural network is provided, which can be performed by an apparatus. The apparatus can be a device (such as a terminal device, a network device, or an AI node), or a component of a device (such as a chip or circuit), which is not limited in this application.

[0059] The method may include: obtaining a channel reciprocity error; receiving a first intermediate gradient; updating the first intermediate gradient based on the channel reciprocity error to obtain a second intermediate gradient, and the second intermediate gradient is used to update a neural network parameter.

[0060] Based on this technical solution, channel reciprocity error is taken into account during neural network training, thereby overcoming the gradient error caused by channel reciprocity and making neural network training more efficient. Furthermore, this technical solution can also be used to train neural networks in scenarios where channel reciprocity is poor, making it applicable to a wider range of scenarios.

[0061] In combination with the fifth aspect, in certain implementations of the fifth aspect, the method also includes: sending a first signal; receiving a second signal, where the second signal is related to the first signal; and obtaining a channel reciprocity error, including: determining the channel reciprocity error based on the first signal and the second signal.

[0062] In a sixth aspect, a communication device is provided, the device being configured to execute the method provided in any one of the first to fifth aspects. Specifically, the device may include units and / or modules, such as a processing unit and / or a communication unit, configured to execute the method provided in any one of the above implementations of any one of the first to fifth aspects.

[0063] In one implementation, the apparatus is a communication device. When the apparatus is a communication device, the communication unit may be a transceiver or an input / output interface; the processing unit may be at least one processor. Alternatively, the transceiver may be a transceiver circuit. Alternatively, the input / output interface may be an input / output circuit.

[0064] In another implementation, the apparatus is a chip, chip system, or circuit used in a communication device. When the apparatus is a chip, chip system, or circuit used in a device, the communication unit may be an input / output interface, interface circuit, output circuit, input circuit, pin, or related circuit on the chip, chip system, or circuit; and the processing unit may be at least one processor, processing circuit, or logic circuit.

[0065] In the seventh aspect, a communication device is provided, which includes: a memory for storing programs; and at least one processor for executing computer programs or instructions stored in the memory to execute the method provided by any of the above-mentioned implementation methods of any of the above-mentioned first to fifth aspects.

[0066] In one implementation, the apparatus is a communication device.

[0067] In another implementation, the apparatus is a chip, a chip system, or a circuit used in a communication device.

[0068] In an eighth aspect, the present application provides a processor for executing the methods provided in the above aspects.

[0069] For the operations such as sending and acquiring / receiving involved in the processor, unless otherwise specified, or if they do not conflict with their actual functions or internal logic in the relevant descriptions, they can be understood as operations such as processor output and input, or as sending and receiving operations performed by the radio frequency circuit and antenna. This application does not limit this.

[0070] In a ninth aspect, a computer-readable storage medium is provided, which stores a program code for execution by a device, wherein the program code includes a method provided by any one of the above-mentioned implementation methods for executing any one of the above-mentioned first to fifth aspects.

[0071] In a tenth aspect, a computer program product comprising instructions is provided, which, when run on a computer, enables the computer to execute the method provided by any one of the above-mentioned implementations of any one of the above-mentioned first to fifth aspects.

[0072] In the eleventh aspect, a chip is provided, which includes a processor and a communication interface. The processor reads instructions stored in the memory through the communication interface and executes the method provided by any of the above-mentioned implementation methods of any of the above-mentioned first to fifth aspects.

[0073] Optionally, as an implementation method, the chip also includes a memory, in which a computer program or instruction is stored, and the processor is used to execute the computer program or instruction stored on the memory. When the computer program or instruction is executed, the processor is used to execute the method provided in any one of the above-mentioned implementation methods of any one of the first to fifth aspects.

[0074] In a twelfth aspect, a communication system is provided, comprising a first communication device and a second communication device. The first communication device is configured to execute the method provided in any one of the implementations of the first aspect, and the second communication device is configured to execute the method provided in any one of the implementations of the second aspect; or the first communication device is configured to execute the method provided in any one of the implementations of the third aspect, and the second communication device is configured to execute the method provided in any one of the implementations of the fourth aspect. BRIEF DESCRIPTION OF THE DRAWINGS

[0075] FIG1 is a schematic diagram of a wireless communication system applicable to an embodiment of the present application.

[0076] FIG2 is another schematic diagram of a wireless communication system applicable to an embodiment of the present application.

[0077] Figure 3 is a schematic diagram of the neuron structure.

[0078] FIG4 is a schematic diagram of a neural network training method 400 provided in an embodiment of the present application.

[0079] FIG5 is a schematic flowchart of a method 500 provided according to an embodiment of the present application.

[0080] FIG6 is a schematic diagram of signal transmission and feedback applicable to the method 500 according to an embodiment of the present application.

[0081] FIG7 is a schematic flowchart of a method 700 provided according to another embodiment of the present application.

[0082] FIG8 is a schematic diagram of signal transmission and feedback applicable to the method 700 according to an embodiment of the present application.

[0083] FIG9 is a schematic flowchart of a method 900 provided according to another embodiment of the present application.

[0084] FIG10 is a schematic diagram of signal transmission and feedback applicable to the method 900 according to an embodiment of the present application.

[0085] FIG11 is a schematic diagram of a neural network training method 1100 provided in an embodiment of the present application.

[0086] FIG12 is a schematic flowchart of a method 1200 provided according to another embodiment of the present application.

[0087] FIG13 is a schematic diagram of signal transmission and feedback applicable to the method 1200 according to an embodiment of the present application.

[0088] FIG14 is a schematic diagram showing the decrease of the loss function when the solution of an embodiment of the present application is adopted.

[0089] FIG15 is a schematic diagram of signal processing applicable to an embodiment of the present application.

[0090] FIG16 is a schematic block diagram of a communication device 1600 provided in an embodiment of the present application.

[0091] FIG17 is a schematic diagram of another communication device 1700 provided in an embodiment of the present application.

[0092] FIG18 is a schematic diagram of a chip system 1800 provided in accordance with an embodiment of the present application. DETAILED DESCRIPTION

[0093] The technical solution in this application will be described below with reference to the accompanying drawings.

[0094] First, a brief introduction is given to the communication system to which this application is applicable.

[0095] The technical solutions provided in this application can be applied to various communication systems, such as: fifth generation (5G) or new radio (NR) systems, long term evolution (LTE) systems, LTE frequency division duplex (FDD) systems, LTE time division duplex (TDD) systems, wireless local area networks (WLAN) systems, satellite communication systems, future communication systems, such as sixth generation (6G) mobile communication systems, or a fusion system of multiple systems. The technical solutions provided in this application can also be applied to device to device (D2D) communication, vehicle-to-everything (V2X) communication, machine to machine (M2M) communication, machine type communication (MTC), and Internet of Things (IoT) communication systems or other communication systems.

[0096] A device in a communication system can send signals to or receive signals from another device. These signals may include information, signaling, or data. The term "device" can also be replaced by an entity, network entity, network element, communication device, communication module, node, communication node, and the like. This disclosure uses devices as examples for description. For example, a communication system may include at least one terminal device and at least one network device. A network device can send downlink signals to a terminal device, and / or a terminal device can send uplink signals to a network device.

[0097] The terminal devices in the embodiments of the present application include various devices with wireless communication functions, which can be used to connect people, objects, machines, etc. The terminal devices can be widely used in various scenarios, such as: cellular communication, D2D, V2X, peer to peer (P2P), M2M, MTC, IoT, virtual reality (VR), augmented reality (AR), industrial control, autonomous driving, telemedicine, smart grid, smart furniture, smart office, smart wearable, smart transportation, smart city drones, robots, remote sensing, passive sensing, positioning, navigation and tracking, autonomous delivery, etc. The terminal device can be a terminal in any of the above scenarios, such as an MTC terminal, an IoT terminal, etc. The terminal device may be a user equipment (UE) of the third generation partnership project (3GPP) standard, a terminal, a fixed device, a mobile station device or a mobile device, a subscriber unit, a handheld device, a vehicle-mounted device, a wearable device, a cellular phone, a smart phone, a SIP phone, a wireless data card, a personal digital assistant (PDA), a computer, a tablet computer, a notebook computer, a wireless modem, a handheld device, a laptop computer, a computer with wireless transceiver function, a smart book, a vehicle, a satellite, a global positioning system (GPS) device, a target tracking device, an aircraft (such as a drone, a helicopter, a multi-copter, a quadcopter, or an airplane), a ship, a remote control device, a smart home device, an industrial device, or a device built into the above-mentioned device (such as a communication module, a modem or a chip in the above-mentioned device), or other processing devices connected to a wireless modem. For ease of description, the terminal device will be described below by taking the terminal or UE as an example.

[0098] It should be understood that in some scenarios, a UE can also be used to act as a base station. For example, a UE can act as a scheduling entity that provides sidelink signals between UEs in scenarios such as V2X, D2D, or P2P.

[0099] In the embodiments of the present application, the device for implementing the function of the terminal device can be the terminal device, or it can be a device that can support the terminal device to implement the function, such as a chip system or chip, which can be installed in the terminal device. In the embodiments of the present application, the chip system can be composed of chips, or it can include chips and other discrete devices. In the embodiments of the present application, only the terminal device is used as an example for description, and the embodiments of the present application are not limited to the solutions of the embodiments of the present application.

[0100] The network device in the embodiments of the present application may be a device for communicating with a terminal device, and may also be referred to as an access network device or a radio access network device. For example, the network device may be a base station. The network device in the embodiments of the present application may refer to a radio access network (RAN) node (or device) that connects a terminal device to a wireless network. Base station can broadly cover various names as follows, or replace the following names, such as: NodeB, evolved NodeB (eNB), next generation NodeB (gNB), relay station, access point, transmission point (TRP), transmitting point (TP), master station, auxiliary station, multi-standard wireless (motor slide retainer, MSR) node, home base station, network controller, access node, wireless node, access point (AP), transmission node, transceiver node, baseband unit (BBU), remote radio unit (RRU), active antenna unit (AAU), remote radio head (RRH), central unit (CU), distributed unit (DU), positioning node, etc. The base station can be a macro base station, a micro base station, a relay node, a donor node or the like, or a combination thereof. The base station can also refer to a communication module, modem or chip used to be set in the aforementioned device or apparatus. The base station can also be a mobile switching center and a device that performs base station functions in D2D, V2X, and M2M communications, a network-side device in a 6G network, or a device that performs base station functions in future communication systems. The base station can support networks with the same or different access technologies. The embodiments of this application do not limit the specific technology and specific device form used by the network equipment.

[0101] Base stations can be fixed or mobile. For example, a helicopter or drone can be configured to act as a mobile base station, and at least one cell can move based on the location of the mobile base station. In other examples, a helicopter or drone can be configured to act as a device that communicates with another base station.

[0102] In some deployments, the network devices mentioned in the embodiments of the present application may include a CU, a DU, or both a CU and a DU, or a control plane CU node (central unit-control plane (CU-CP)), a user plane CU node (central unit-user plane (CU-UP)), and a DU node. For example, the network devices may include a gNB-CU-CP, a gNB-CU-UP, and a gNB-DU.

[0103] In some deployments, multiple RAN nodes collaborate to assist terminals in achieving wireless access, with different RAN nodes implementing portions of the base station's functionality. For example, a RAN node can be a CU, DU, CU-CP, CU-UP, or RU. The CU and DU can be separate or included in the same network element, such as the BBU. The RU can be included in a radio frequency device or radio unit, such as an RRU, AAU, or RRH.

[0104] The RAN node can support one or more types of fronthaul interfaces. Different fronthaul interfaces correspond to DUs and RUs with different functions. If the fronthaul interface between the DU and the RU is the common public radio interface (CPRI), the DU is configured to implement one or more baseband functions, and the RU is configured to implement one or more radio frequency functions. If the fronthaul interface between the DU and the RU is the enhanced common public radio interface (eCPRI), relative to CPRI, part of the downlink and / or uplink baseband functions are moved from the DU to the RU for implementation. The division between the DU and the RU is different, corresponding to different types (category, Cat) of eCPRI, such as eCPRI Cat A, B, C, D, E, and F.

[0105] Taking eCPRI Cat A as an example, for downlink transmission, based on layer mapping, the DU is configured to implement layer mapping and one or more functions preceding it (i.e., one or more of coding, rate matching, scrambling, modulation, and layer mapping). Other functions after layer mapping (e.g., RE mapping, digital beamforming (BF), or one or more of inverse fast Fourier transform (IFFT) / cyclic prefix (CP) addition) are moved to the RU for implementation. For uplink transmission, based on RE demapping, the DU is configured to implement demapping and one or more functions preceding it (i.e., one or more of decoding, rate matching, descrambling, demodulation, inverse discrete Fourier transform (IDFT), channel equalization, and RE demapping). Other functions after demapping (e.g., one or more of digital BF or fast Fourier transform (FFT) / CP removal) are moved to the RU for implementation. It is understandable that for the functional description of DU and RU corresponding to various types of eCPRI, reference can be made to the eCPRI protocol, which will not be described in detail here.

[0106] In one possible design, the processing unit for implementing baseband functions in the BBU is called a baseband high layer (BBH) unit, and the processing unit for implementing baseband functions in the RRU / AAU / RRH is called a baseband low layer (BBL) unit.

[0107] In different systems, CU (or CU-CP and CU-UP), DU or RU may also have different names, but those skilled in the art can understand their meanings. For example, in the ORAN system, CU may also be called O-CU (Open CU), DU may also be called O-DU, CU-CP may also be called O-CU-CP, CU-UP may also be called O-CU-UP, and RU may also be called O-RU. Any unit of CU (or CU-CP, CU-UP), DU and RU in this application can be implemented by a software module, a hardware module, or a combination of a software module and a hardware module.

[0108] In the embodiments of the present application, the device for implementing the function of the network device can be a network device, or a device that can support the network device to implement the function, such as a chip system or chip, which can be installed in the network device. In the embodiments of the present application, the chip system can be composed of chips, or it can include chips and other discrete devices. In the embodiments of the present application, only the device for implementing the function of the network device is a network device as an example for description, and does not constitute a limitation on the solutions of the embodiments of the present application.

[0109] Network devices and terminal devices can be deployed on land, including indoors or outdoors, handheld or vehicle-mounted; they can also be deployed on the water surface; they can also be deployed on aircraft, balloons and satellites in the air. The embodiments of this application do not limit the scenarios in which network devices and terminal devices are located. In addition, terminal devices and network devices can be hardware devices, or they can be software functions running on dedicated hardware, software functions running on general-purpose hardware, such as virtualization functions instantiated on a platform (e.g., a cloud platform), or entities including dedicated or general-purpose hardware devices and software functions. This application does not limit the specific forms of terminal devices and network devices.

[0110] In addition, in order to support AI technology in wireless networks, AI nodes may also be introduced into the network.

[0111] Optionally, the AI ​​node can be deployed in one or more of the following locations in the communication system: access network equipment, terminal equipment, or core network equipment. Alternatively, the AI ​​node can be deployed separately, for example, in a location other than any of the above devices, such as a host or cloud server in an over-the-top (OTT) system. The AI ​​node can communicate with other devices in the communication system, such as one or more of the following: network equipment, terminal equipment, or core network elements.

[0112] It is understood that this application does not limit the number of AI nodes. For example, when there are multiple AI nodes, the multiple AI nodes can be divided based on function, such as different AI nodes are responsible for different functions.

[0113] It can also be understood that AI nodes can be independent devices, or they can be integrated into the same device to implement different functions, or they can be network elements in hardware devices, or they can be software functions running on dedicated hardware, or they can be virtualized functions instantiated on a platform (for example, a cloud platform). This application does not limit the specific form of the above-mentioned AI nodes.

[0114] Refer to FIG1 , which is a schematic diagram of a wireless communication system applicable to an embodiment of the present application.

[0115] As shown in Figure 1, the wireless communication system includes a wireless access network 100. The wireless access network 100 can be a next-generation (e.g., 6G or higher) wireless access network, or a traditional (e.g., 5G, 4G, 3G, or 2G) wireless access network. One or more terminal devices (120a-120j, collectively referred to as 120) can be connected to each other or to one or more network devices (110a, 110b, collectively referred to as 110) in the wireless access network 100. Network elements in the wireless communication system are connected through interfaces (e.g., NG, Xn) or air interfaces. In addition, one or more AI modules may be provided in each network element in the wireless communication system. The AI ​​modules deployed in different network elements may be the same or different.

[0116] FIG1 is only a schematic diagram. The wireless communication system may further include other devices, such as core network devices, wireless relay devices and / or wireless backhaul devices, which are not shown in FIG1 .

[0117] Refer to FIG. 2 , which is another schematic diagram of a wireless communication system applicable to an embodiment of the present application.

[0118] As shown in Figure 2, the wireless communication system includes a RAN intelligent controller (RIC). As an example, RIC can be used to implement AI-related functions. As an example, the RIC includes a near-real time RIC (near-real time RIC, near-RT RIC) and a non-real time RIC (non-real time RIC, Non-RT RIC). Among them, the non-real-time RIC mainly processes non-real-time information, such as data that is not sensitive to delay, and the delay of the data can be in the order of seconds. Real-time RIC mainly processes near-real-time information, such as data that is relatively sensitive to delay, and the delay of the data is in the order of tens of milliseconds.

[0119] The near real-time RIC is used for model training and reasoning. For example, it is used to train an AI model and use the AI ​​model for reasoning. The near real-time RIC can obtain network-side and / or terminal-side information from a RAN node (e.g., CU, CU-CP, CU-UP, DU, and / or RU) and / or a terminal. This information can be used as training data or reasoning data. Optionally, the near real-time RIC can deliver the reasoning result to the RAN node and / or the terminal. Optionally, the reasoning result can be exchanged between the CU and the DU, and / or between the DU and the RU. For example, the near real-time RIC delivers the reasoning result to the DU, and the DU sends it to the RU.

[0120] The non-real-time RIC is also used for model training and reasoning. For example, it is used to train an AI model and use the model for reasoning. The non-real-time RIC can obtain network-side and / or terminal-side information from RAN nodes (such as CU, CU-CP, CU-UP, DU and / or RU) and / or terminals. This information can be used as training data or reasoning data, and the reasoning results can be submitted to the RAN node and / or terminal. Optionally, the reasoning results can be exchanged between the CU and the DU, and / or between the DU and the RU. For example, the non-real-time RIC submits the reasoning results to the DU, and the DU sends it to the RU.

[0121] The near real-time RIC and non-real-time RIC may also be separately configured as a network element. Optionally, the near real-time RIC and non-real-time RIC may also be part of other devices. For example, the near real-time RIC is configured in a RAN node (e.g., a CU or DU), while the non-real-time RIC is configured in an OAM, a cloud server, a core network device, or other network device.

[0122] In practical applications, the wireless communication system may include multiple network devices (also called access network devices) and multiple terminal devices at the same time, without limitation. A network device may serve one or more terminal devices at the same time. A terminal device may also access one or more network devices at the same time. The embodiments of the present application do not limit the number of terminal devices and network devices included in the wireless communication system.

[0123] To facilitate understanding of the embodiments of the present application, the following briefly describes the relevant concepts and technologies involved in the present application.

[0124] 1. Artificial Intelligence: This refers to the ability of machines to learn, accumulate experience, and solve problems that humans can solve through experience, such as natural language understanding, image recognition, and chess. Artificial Intelligence can be understood as the intelligence exhibited by machines created by humans. Generally, AI refers to the technology that represents human intelligence through computer programs. The goals of AI include understanding intelligence by constructing computer programs that can perform symbolic reasoning or deduction.

[0125] 2. Machine learning: This is an implementation of artificial intelligence. Machine learning is a method that empowers machines to learn, enabling them to perform functions that cannot be accomplished through direct programming. In practical terms, machine learning utilizes data to train models and then uses these models to make predictions. There are many machine learning methods, such as neural networks (NNs), decision trees, and support vector machines. Machine learning theory primarily involves the design and analysis of algorithms that enable computers to learn automatically. Machine learning algorithms automatically analyze data to identify patterns and use these patterns to make predictions about unknown data.

[0126] 3. Neural Network: A specific embodiment of machine learning. A neural network is a mathematical model that processes information by mimicking the behavioral characteristics of animal neural networks. The concept of a neural network is derived from the neuronal structure of the brain. Each neuron performs a weighted sum operation on its input values, and the result of this weighted summation is passed through an activation function to generate an output.

[0127] See Figure 3, which is a schematic diagram of the neuron structure. As shown in Figure 3, assume that the input of the neuron is x = [x0, x1, ..., x n ], and the weights corresponding to each input are w=[w,w1,…,w n ], the weighted summation bias is b. Where b can be an integer, a decimal, or a complex number, etc. The activation function can be diversified. As an example, assuming that the activation function of a neuron is: y = f(z) = max(0,z), then the output of the neuron is: As another example, suppose the activation function of a neuron is: y = f(z) = z, then the output of the neuron is: As shown in Figure 3. The activation functions of different neurons in a neural network can be the same or different.

[0128] Neural networks generally comprise a multi-layer structure, with each layer comprising one or more logical decision units, referred to as neurons. Increasing the depth and / or width of a neural network can enhance its expressive power, providing more powerful information extraction and abstract modeling capabilities for complex systems. The depth of a neural network can be understood as the number of layers it comprises, and the number of neurons in each layer can be referred to as the width of that layer. In one possible implementation, a neural network comprises an input layer and an output layer. The input layer of the neural network processes the input through neurons and then passes the result to the output layer, which then obtains the output of the neural network. In another possible implementation, a neural network comprises an input layer, a hidden layer, and an output layer. The input layer of the neural network processes the input through neurons and then passes the result to an intermediate hidden layer. The hidden layer then passes the calculation result to the output layer or an adjacent hidden layer, which then obtains the output of the neural network. A neural network can comprise one or more sequentially connected hidden layers, without limitation.

[0129] 4. Loss function: This function measures the difference between the model's predicted value and the true value. During neural network training, the loss function describes the gap or discrepancy between the neural network's output and the ideal target value. Neural network training involves adjusting neural network parameters to reduce the loss function value to a threshold or meet the target requirement. Neural network parameters can include at least one of the following: the number of neural network layers, their width, neuron weights, and parameters in the neuron activation function.

[0130] 5. Gradient: The training method of a neural network can be to use a loss function to evaluate the output of the neural network, and transmit the error in reverse, and iterate the parameters to be optimized (i.e., neural network parameters) through the gradient descent method until the loss function reaches the minimum value.

[0131] As an example, the process of gradient descent can be expressed as: Where θ represents the parameters to be optimized (such as w and b in Figure 3). L represents the loss function. η represents the learning efficiency, which can be used to control the step size of gradient descent. Represents the symbol of partial derivative.

[0132] As mentioned above, neural networks generally include a multi-layer structure. One possible implementation method is that the gradient of the parameters of the previous layer can be recursively calculated from the gradient of the parameters of the next layer. Taking neurons i and j as an example, the weight w between neurons i and j is ij The gradient of can be expressed as: Among them, s i represents the weighted sum of inputs to neuron i.

[0133] 6. Intermediate gradient: It is one or more items in the gradient expression of the neural network parameters, or the product of multiple items.

[0134] For example, suppose the communication system includes neural networks #1, #2, and #3, with the corresponding parameters θ1, θ2, and θ3. The inputs to neural networks #1, #2, and #3 are Q1, Q2, and Q3, respectively. Zi = θi·Q(i-1), Q(i-1) = σ·Z(i-1). Here, Zi represents the output of neural network #i, with i taking values ​​of 1, 2, or 3. σ represents one or more functions that process the data.

[0135] The gradient of the parameters of neural network #3 satisfies Equation 1.

[0136] The gradient of the parameters of neural network #2 satisfies Equation 2.

[0137] The gradient of the parameters of neural network #1 satisfies Equation 3.

[0138] Where L represents the loss function. The intermediate gradient of the parameters of neural network #3 is one or more items in Equation 1, or the product of multiple items. The intermediate gradient of the parameters of neural network #2 is one or more items in Equation 2, or the product of multiple items. The intermediate gradient of the parameters of neural network #1 is one or more items in Equation 3, or the product of multiple items. For example, the intermediate gradient of the parameters of neural network #3 is: For another example, the intermediate gradient of the parameters of neural network #3 is and

[0139] It can be understood that the intermediate gradient is only named for distinction, and its naming does not limit the scope of protection of the embodiments of the present application.

[0140] 7. AI model: It is an algorithm or computer program that can realize AI functions. The AI ​​model represents the mapping relationship between the input and output of the model, or the AI ​​model is a function model that maps input of a certain dimension to output of a certain dimension. The parameters of the function model can be obtained through machine learning training. For example, f(x) = ax 2+b is a quadratic function model, which can be regarded as an AI model. a and b are the parameters of the AI ​​model, and a and b can be obtained through machine learning training. For example, the AI ​​models mentioned in the embodiments below are not limited to neural networks, linear regression models, decision tree models, support vector machines (SVMs), Bayesian networks, Q learning models, or other machine learning (ML) models.

[0141] AI model design primarily includes a data collection phase (e.g., collecting training data and / or inference data), a model training phase, and a model inference phase. It can also include an inference result application phase. In the aforementioned data collection phase, a data source is used to provide training data sets and inference data. In the model training phase, an AI model is obtained by analyzing or training the training data provided by the data source. Learning the AI ​​model through the model training node is equivalent to using the training data to learn the mapping relationship between the AI ​​model's input and output. In the model inference phase, the AI ​​model, trained in the model training phase, performs inference based on the inference data provided by the data source to obtain an inference result. This phase can also be understood as inputting the inference data into the AI ​​model and obtaining an output from the AI ​​model, which is the inference result. The inference result can indicate configuration parameters used (executed) by the execution object and / or the operations performed by the execution object. In the inference result application phase, the inference result is published. For example, the inference result can be centrally planned by an actor entity, such as an actor entity that sends the inference result to one or more actors (e.g., core network devices, access network devices, or terminal devices) for execution. For example, the execution entity can also provide feedback on the performance of the AI ​​model to the data source, facilitating subsequent update and training of the AI ​​model.

[0142] It is understood that the AI ​​model can be implemented as a hardware circuit, software, or a combination of software and hardware, without limitation. Non-limiting examples of software include: program code, program, subroutine, instruction, instruction set, code, code segment, software module, application, or software application.

[0143] 8. Model training: The process of selecting a suitable loss function and using an optimization algorithm to train the model parameters so that the value of the loss function is less than the threshold, or the value of the loss function meets the target requirements.

[0144] 9. Model application: Use the trained model to solve practical problems.

[0145] The following will describe in detail the method provided by the embodiment of the present application in conjunction with the accompanying drawings. The embodiment of the present application proposes two solutions in order to improve the efficiency of model training and be applicable to more scenarios. One solution is: a device calculates the reciprocity error based on the signal it sends and the signal received from the other device side, and then corrects the intermediate gradient, and updates the neural network parameters according to the corrected intermediate gradient. In this way, the gradient error caused by channel reciprocity can be overcome. Another solution is: two devices perform channel estimation based on the reference signal respectively, and correct the intermediate gradient. In this way, the gradient error caused by channel reciprocity can also be overcome.

[0146] The embodiments provided in this application can be applied to the communication system shown in Figure 1 or Figure 2 above, without limitation.

[0147] It should be noted that in this application, "indication" can include direct indication, indirect indication, explicit indication, and implicit indication. When describing that a certain indication information is used to indicate A, it can be understood that the indication information carries A, directly indicates A, or indirectly indicates A.

[0148] In this application, the information indicated by the indication information is referred to as the information to be indicated. In the specific implementation process, there are many ways to indicate the information to be indicated, such as but not limited to, the information to be indicated can be directly indicated, such as the information to be indicated itself or the index of the information to be indicated. The information to be indicated can also be indirectly indicated by indicating other information, wherein there is an association between the other information and the information to be indicated. It is also possible to indicate only a part of the information to be indicated, while the other parts of the information to be indicated are known or agreed in advance. For example, the indication of specific information can also be achieved with the help of the arrangement order of each information agreed in advance (for example, stipulated by the protocol), thereby reducing the indication overhead to a certain extent. In addition, the information to be indicated can be sent together as a whole, or it can be divided into multiple sub-information and sent separately, and the sending period and / or sending time of these sub-information can be the same or different.

[0149] In the following embodiments, the first device and the second device are taken as examples for illustrative description.

[0150] Among them, the first device can be a terminal device or a component of a terminal device (such as a chip or circuit), or the first device can be a network device or a component of a network device (such as a chip or circuit), or the first device can be an AI node or a component of an AI node (such as a chip or circuit).

[0151] The second device may be a terminal device or a component of a terminal device (such as a chip or circuit), or the first device may be a network device or a component of a network device (such as a chip or circuit), or the first device may be an AI node or a component of an AI node (such as a chip or circuit).

[0152] Referring to Figure 4, Figure 4 is a schematic diagram of a neural network training method 400 provided in an embodiment of the present application. The method 400 shown in Figure 4 may include the following steps.

[0153] 401. A first device sends a first signal.

[0154] Correspondingly, the second device receives the first signal. It is understood that in actual transmission, the signal may change after passing through the channel. For the purpose of distinction, the first signal received by the second device is called the third signal, that is, the third signal is the signal after the first signal has passed through the channel.

[0155] The first signal may be a complex symbol, or a real symbol (eg, an imaginary part is 0). For example, the first signal may be a complex value mapped onto an air interface resource (eg, a frequency domain resource).

[0156] Optionally, the first signal is any one of the following: a custom sequence, a reference signal, or a data signal.

[0157] In one example, the first signal is a reference signal, such as a demodulation reference signal (DMRS), a channel state information-reference signal (CSI-RS), or a sounding reference signal (SRS).

[0158] In another example, the first signal is a custom sequence, such as a custom pseudo-noise (PN) sequence.

[0159] In another example, the first signal is a data signal. For example, the first signal is service data. In another example, the first signal is specific data, such as data that does not require coding, modulation, or other processing.

[0160] 402. A first device receives a first intermediate gradient and a second signal, where the second signal is related to the first signal.

[0161] Accordingly, the second device sends the first intermediate gradient and the second signal. It is understood that in actual transmission, the signal may change after passing through the channel. To distinguish, the second signal sent by the second device is called the fourth signal, that is, the second signal is the signal after the fourth signal has passed through the channel.

[0162] For the intermediate gradient, please refer to the previous terminology explanation and will not be elaborated here.

[0163] The first intermediate gradient is used to update the parameters of the neural network. Specifically, the second device provides the first intermediate gradient to the first device, thereby facilitating the first device to update the parameters of the neural network located (or deployed) on the first device based on the first intermediate gradient.

[0164] As an example, the neural network parameters may include at least one of the following: the number of layers and width of the neural network, the weights of neurons, and parameters in the activation function of neurons.

[0165] Optionally, the first intermediate gradient is determined according to the first signal and / or the data signal.

[0166] In one example, the first intermediate gradient is determined based on the first signal.

[0167] For example, when the first signal is a data signal, the first intermediate gradient is determined based on the first signal. In this case, since the first signal is a data signal, it can also be considered that the first intermediate gradient is determined based on the data signal.

[0168] In the embodiments of the present application, the method for determining the first intermediate gradient based on the data signal can refer to existing methods and is not limited thereto. For example, the second device can determine a loss function based on the data signal, and further determine the first intermediate gradient based on the loss function. Alternatively, the first intermediate gradient can be determined based on the data signal, or alternatively, based on the loss function.

[0169] In another example, the first intermediate gradient is determined according to the data signal.

[0170] For example, when the first signal is a data signal or a custom sequence, the first intermediate gradient is determined according to the data signal.

[0171] In another example, the first intermediate gradient is determined according to the first signal and the data signal.

[0172] For example, when the first signal is a reference signal, the first intermediate gradient is determined according to the first signal and the data signal.

[0173] A specific example is given below.

[0174] For example, if the first signal is a reference signal, the determination of the first intermediate gradient may also take the first signal into account.

[0175] Specifically, because the first signal is a reference signal, after receiving the first signal, the second device can perform channel estimation based on the first signal and determine the first intermediate gradient based on the channel estimation result. For example, the second device can obtain the channel condition based on the channel estimation result and then compensate the first intermediate gradient determined based on the loss function based on the channel condition to obtain the compensated first intermediate gradient, and send the compensated first intermediate gradient to the first device.

[0176] Considering that signals may change after passing through a channel, the first intermediate gradient is determined based on the first signal. Alternatively, the first intermediate gradient can be determined based on a third signal (i.e., the first signal after passing through the channel). Specifically, a first device sends a first signal to a second device. After passing through the channel, the second device receives the third signal. The second device performs channel estimation based on the received third signal and determines the first intermediate gradient based on the channel estimation results.

[0177] The second signal is correlated with the first signal. Based on this, the second signal can be determined based on the first signal. Specifically, after receiving the first signal sent by the first device, the second device can determine the second signal based on the correlation between the second signal and the first signal, and then send the second signal to the first device.

[0178] Considering that signals may change after passing through the channel, the second signal can be correlated with the first signal, and the fourth signal can also be correlated with the third signal. Specifically, the first device sends the first signal to the second device. After passing through the channel, the second device receives the third signal. The third signal is correlated with the fourth signal, so the second device determines the fourth signal based on this correlation. The second device sends the fourth signal to the first device. After passing through the channel, the first device receives the second signal. For ease of description, the following mainly uses the first and second signals as examples.

[0179] Optionally, the second signal is related to the first signal, including: the second signal and the first signal satisfy Formula 4 or Formula 5. A2=f(A1) Formula 4 A2=f(A1 * ) Formula 5

[0180] Among them, A1 represents the first signal, A2 represents the second signal, A1 * Indicates the conjugate operation on A1.

[0181] Considering that the signal may change after passing through the channel, Formula 4 can also be replaced by: A4 = f(A3), and Formula 5 can also be replaced by: A4 = f(A3 * ). Wherein, A3 represents the third signal, that is, the first signal after passing through the channel. A4 represents the fourth signal, and the second signal is the signal after the fourth signal passes through the channel. A3 * Indicates the conjugate operation on A3.

[0182] The specific form of the function f is not limited. Several possible forms are listed below. The meanings of the parameters mentioned below can be referred to the previous description.

[0183] The first possible form is A2 = αA1. α represents a constant. Considering that the signal may change after passing through the channel, A2 = αA1 can also be replaced by A4 = αA3.

[0184] The second possible form, A2 = αA1 * Considering that the signal may change after passing through the channel, A2=αA1 * Can also be replaced by: A4 = αA3 * .

[0185] The third possible form, Considering that the signal may change after passing through the channel, Can also be replaced by:

[0186] The fourth possible form, Considering that the signal may change after passing through the channel, Can also be replaced by: |A| represents the amplitude of the element A. Taking the first signal A1 as an example, assuming that the first signal includes multiple symbols (such as complex symbols or real symbols), |A1| represents the independent normalization processing of the power of each symbol in the multiple symbols.

[0187] The fifth possible form, Considering that the signal may change after passing through the channel, Can also be replaced by: ||x|| p represents the p-norm of the x-vector. For example, When p=2, it can be understood that the power is normalized. Taking the first signal A1 as an example, assuming that the first signal includes multiple symbols (such as complex symbols or real symbols), ||A1||2 represents the power normalization of some or all of the multiple symbols.

[0188] The sixth possible form, Considering that the signal may change after passing through the channel, Can also be replaced by:

[0189] The forms listed above are only examples, and any variations of the above forms are applicable to the embodiments of the present application.

[0190] Optionally, the correlation form of the second signal and the first signal, such as Formula 4 or Formula 5 or any possible form above, can be predefined, or can be sent from the first device to the second device, or can be sent from the second device to the first device, without limitation.

[0191] 403. The first device updates the first intermediate gradient according to the first signal and the second signal to obtain a second intermediate gradient, and the second intermediate gradient is used to update the neural network parameters.

[0192] Optionally, the first device updates the first intermediate gradient based on the first signal and the second signal, including: the first device determines a channel reciprocity error value based on the first signal and the second signal; and updates (or corrects, or adjusts, or compensates) the first intermediate gradient based on the channel reciprocity error value.

[0193] Illustratively, the channel reciprocity error E determined by the first device according to the first signal and the second signal satisfies: E=A2A1.

[0194] For example, the second intermediate gradient obtained by the first device satisfies: gE * Where g represents the first intermediate gradient and the superscript * indicates the conjugate operation.

[0195] Based on the embodiments of the present application, the first device can determine the channel reciprocity error based on the signal it sends (i.e., the first signal) and the received signal (i.e., the second signal), as well as the relationship between the two signals. Based on the channel reciprocity error, the first device can update the received intermediate gradient (i.e., the first intermediate gradient), and then update the neural network parameters based on the updated intermediate gradient (i.e., the second intermediate gradient). In this way, the gradient error caused by channel reciprocity can be overcome and is applicable to more scenarios.

[0196] Optionally, the method 400 further includes: the first device sending first indication information, where the first indication information indicates a type of the first signal and / or the second signal. Correspondingly, the second device receives the first indication information.

[0197] The first indication information may be implemented by one or more bits or a bitmap, which will be described below with reference to several examples.

[0198] Example 1: The first indication information indicates the types of the first signal and the second signal.

[0199] As an example, the types of the first signal and the second signal may exist in the form of a table, function, text, or string, such as for storage or transmission. Table 1 below is an example of presenting the types of the first signal and the second signal in table form.

[0200] Table 1

[0201] Type A indicates that the first signal is a custom sequence and the second signal is related to the first signal. Type B indicates that the first signal is a data signal and the second signal is related to the first signal, or that the second signal is a subset of the first signal. Type C indicates that the first signal is a reference signal and the second signal is related to the first signal. Type D indicates that the first signal is a reference signal and the second signal is a reference signal, and the first and second signals can be independent. Type D will be described in detail later in conjunction with method 1100.

[0202] The specific form of the correlation between the second signal and the first signal can be as shown in Formula 4, Formula 5, or any one of the first to sixth possible forms. Furthermore, Table 1 also lists the specific forms of the correlation between the first signal and the second signal. In different types, the specific forms of the correlation between the first signal and the second signal can be the same or different.

[0203] Taking Table 1 as an example, for example, the first indication information can be implemented by 2 bits. For example, assuming that the 2 bits are set to "00", it means that the types of the first signal and the second signal belong to type A, that is, the first signal is a custom sequence, and the second signal is related to the first signal. For another example, assuming that the 2 bits are set to "01", it means that the types of the first signal and the second signal belong to type B, that is, the first signal is a data signal, and the second signal is related to the first signal. For another example, assuming that the 2 bits are set to "10", it means that the types of the first signal and the second signal belong to type C, that is, the first signal is a reference signal, and the second signal is related to the first signal. For another example, assuming that the 2 bits are set to "11", it means that the types of the first signal and the second signal belong to type D, that is, the first signal is a reference signal, and the second signal is a reference signal.

[0204] The above is an example and is not limiting. For example, the first indication information can also be implemented through a bitmap. For example, a 4-bit bitmap is used to indicate the type of the first signal and the second signal, and the 4 bits correspond to type A, type B, type C, and type D respectively. If the bit value is "1", it means that the first signal and the second signal of this type are enabled. For example, if the value of the bitmap is "1000", it means that the type of the first signal and the second signal is type A. For another example, if the value of the bitmap is "0100", it means that the type of the first signal and the second signal is type B. For another example, if the value of the bitmap is "0010", it means that the type of the first signal and the second signal is type C. For another example, if the value of the bitmap is "0001", it means that the type of the first signal and the second signal is type D.

[0205] Example 2: The first indication information indicates the type of the first signal.

[0206] As an example, the type of the first signal may exist in the form of a table, function, text, or character string, such as for storage or transmission. Table 2 below is an example of presenting the type of the first signal in table form.

[0207] Table 2

[0208] Taking Table 2 as an example, the first indication information can be implemented using 2 bits. For example, assuming that the 2 bits are set to "00", it indicates that the first signal is a custom sequence. For another example, assuming that the 2 bits are set to "01", it indicates that the first signal is a data signal. For another example, assuming that the 2 bits are set to "10", it indicates that the first signal is a reference signal.

[0209] The above is an example and is not limiting. For example, the first indication information can also be implemented through a bitmap. For example, a 3-bit bitmap is used to indicate the type of the first signal, and the 3 bits correspond to a custom sequence, a data signal, and a reference signal, respectively. If the bit value is "1", it indicates that the first signal of this type is enabled. For example, if the value of the bitmap is "100", it indicates that the first signal is a custom sequence. For another example, if the value of the bitmap is "010", it indicates that the first signal is a data signal. For another example, if the value of the bitmap is "001", it indicates that the first signal is a reference signal.

[0210] In Example 2, the type of the second signal may be predefined or determined based on the type of the first signal. For example, if the second signal is related to the first signal and the first signal is a data signal, then the second signal is implicitly also a data signal, such as a subset of the data signal.

[0211] Example 3: The first indication information indicates the type of the second signal.

[0212] As an example, the type of the second signal may exist in the form of a table, function, text, or character string, such as for storage or transmission. Table 3 below is an example of presenting the type of the second signal in table form.

[0213] Table 3

[0214] Taking Table 3 as an example, the first indication information can be implemented using 1 bit. For example, assuming that the 1 bit is set to "1", it indicates that the second signal is a data signal. For another example, assuming that the 1 bit is set to "0", it indicates that the first signal is a reference signal.

[0215] The above is an example and is not intended to be limiting. For example, the first indication information may also be implemented via a bitmap. For example, a 2-bit bitmap may be used to indicate the type of the first signal, with the 2 bits corresponding to the data signal and the reference signal, respectively. For details, please refer to the descriptions in Examples 1 and 2 above and will not be repeated here.

[0216] In Example 3, the type of the first signal may be predefined or determined based on the type of the second signal. For example, if the second signal is related to the first signal and the second signal is a data signal, the first signal is implicitly also a data signal.

[0217] It should be understood that Tables 1 to 3 above are merely illustrative, and any variations of Tables 1 to 3 are applicable to the embodiments of this application. For example, in Tables 1 and / or 2 above, the reference signal and the custom sequence can be combined. For example, in Table 2, there are two types of first signals: data signal and reference signal. For another example, considering that the signal may change after passing through the channel, the second signal in Tables 1 to 3 can be replaced with the fourth signal.

[0218] Optionally, method 400 further includes: the first device receiving second indication information, the second indication information indicating a pattern of the second signal. Accordingly, the second device sends the second indication information. Based on this, the first device can accurately receive the second signal based on the pattern of the second signal.

[0219] As an example, the pattern of the second signal includes at least one of the following: a mask of the second signal, resources occupied by the second signal, and a sequence number of the second signal. That is, the first device may receive second indication information, where the second indication information indicates at least one of the following: a mask of the second signal, resources occupied by the second signal, and a sequence number of the second signal.

[0220] The mask of the second signal may be used to indicate that positions where the second signal is mapped are set to 1, and other positions are set to 0. For example, the mask of the second signal may be used to indicate that positions where the second signal is mapped in a resource block (RB) or multiple orthogonal frequency division multiplexing (OFDM) symbols (e.g., 14 OFDM symbols) are set to 1, and other positions are set to 0.

[0221] Among them, the resources occupied by the second signal include, for example: the time domain resources and / or frequency domain resources occupied by the second signal. For example, the resources occupied by the second signal include: the subcarrier sequence number and the OFDM symbol sequence number where the second signal is located. For example, assuming that the resources include an RB (the RB includes multiple subcarriers), multiple OFDM symbols (such as 14 OFDM symbols), and the row represents the subcarrier and the column represents the OFDM symbol, then by indicating the row index and the column index, the resources occupied by the second signal can be indicated, that is, the position of the second signal can be indicated.

[0222] The sequence number of the second signal can be used to indicate which locations are the second signal. For example, the first device sends a DMRS to the second device, and the second signal is a subset of the DMRS. The sequence number of the second signal can be used to determine which DMRS locations are the second signal.

[0223] For ease of understanding, several possible processes applicable to method 400 are introduced below in combination with possible forms of the first signal.

[0224] 5 , which is a schematic flow chart of a method 500 according to an embodiment of the present application. The method 500 is applicable to scenarios where the first signal is a custom sequence. The method 500 shown in FIG5 may include the following steps.

[0225] 501. The first device sends configuration information to the second device.

[0226] The configuration information may include at least one of the following: the type of the first signal, the type of the second signal, the pattern of the first signal, and the pattern of the second signal. In this embodiment of the present application, the first signal is a custom sequence, and the second signal is related to the first signal. For information on signal types and patterns, please refer to the description in method 400 and will not be repeated here.

[0227] 502. The first device infers model #1 and obtains the output of model #1.

[0228] Specifically, the data to be transmitted is input into model #1, and the output of model #1, that is, the data signal, is obtained.

[0229] Here, model #1 may be a model located in or deployed in the first device, such as a neural network model.

[0230] 503. The first device sends the first signal and the output of model #1 to the second device.

[0231] For example, the first device maps the first signal and the output of model #1 (i.e., the data signal) onto air interface resources and then transmits them to the second device. Air interface resources include at least one of the following: time domain resources, frequency domain resources, and spatial domain resources. Accordingly, the second device receives the first signal and the data signal after they have been routed through the channel. To distinguish them, the first signal sent by the first device is denoted as A1, and the first signal after it has been routed through the channel is denoted as A3.

[0232] Refer to Figure 6, which is a schematic diagram of signal transmission and feedback applicable to method 500 of an embodiment of the present application. As shown in Figure 6, in forward reasoning, the first device maps the first signal and the data signal to the air interface resource, which can then be transmitted to the second device. When mapping the first signal, the first signal can be mapped to the air interface resource alone, or it can replace the original data signal to be mapped to the air interface resource. For example, assuming that the number of first signals is n and the number of data signals is m, the first signal can be mapped to the air interface resource alone, that is, the number of signals after mapping is (n+m); alternatively, the first signal can also replace the original data signal to be mapped to the air interface resource, that is, the number of signals after mapping is m.

[0233] In method 500 , the second device may determine a first intermediate gradient based on the data signal. For example, the second device may determine a loss function based on the data signal, and then determine the first intermediate gradient based on the loss function, as shown in steps 504 and 505 below.

[0234] 504 , the second device infers model #2, obtains the output of model #2, and calculates the loss function.

[0235] Specifically, the received data signal is input into model #2 to obtain the output of model #2. The second device can calculate the loss function based on the output of model #2. The specific calculation method can refer to existing methods and is not limited to this.

[0236] Here, model #2 can be a model located in or deployed in a second device, such as a neural network model.

[0237] 505. The second device determines a first intermediate gradient according to the loss function.

[0238] As an example, the first intermediate gradient determined by the second device satisfies: Where L represents the loss function, r represents the output of model #2, and Y represents the input of model #2. It means partial derivative.

[0239] 506. The second device sends the first intermediate gradient and the fourth signal to the first device.

[0240] For example, the second device maps the first intermediate gradient and the fourth signal to air interface resources and then transmits them to the first device. Accordingly, the first device receives the fourth signal (i.e., the second signal) after the channelization and the first intermediate gradient. To distinguish them, the fourth signal is denoted as A4, and the fourth signal after the channelization is denoted as the second signal A2.

[0241] As shown in FIG6 , in reverse reasoning, the second device maps the first intermediate gradient and the fourth signal to air interface resources, and then transmits them to the first device.

[0242] The fourth signal is correlated with the third signal, meaning that the fourth signal can be determined based on the third signal. Specifically, after receiving the third signal, the second device determines the fourth signal based on the correlation between the fourth signal and the third signal. The correlation between the fourth signal and the third signal can be predefined or sent by the first device to the second device, without limitation.

[0243] The following describes the correlation between the fourth signal and the third signal. As an example, the fourth signal and the third signal satisfy any of the following: A4 = f(A3), A4 = f(A3 * ). The specific form of function f is not limited. For example, A4 = αA3. For another example, A4 = αA3 * For another example, For another example, For another example, For another example,

[0244] 507 : The first device updates the first intermediate gradient according to the first signal and the received fourth signal to obtain a second intermediate gradient.

[0245] That is, the first device updates the first intermediate gradient based on the first signal and the second signal to obtain the second intermediate gradient. Specifically, the first device calculates a channel reciprocity error value based on the first signal and the second signal, and updates (or corrects, or adjusts, or compensates) the first intermediate gradient based on the channel reciprocity error value.

[0246] For example, the channel reciprocity error E calculated by the first device satisfies: E=A2A1.

[0247] For example, the second intermediate gradient obtained by the first device satisfies: gE * .

[0248] 508. The first device updates the parameters of model #1 according to the second intermediate gradient.

[0249] Specifically, the first device calculates the parameters of model #1 according to the second intermediate gradient and updates the gradient of model #1.

[0250] Optionally, the above steps, such as step 502 to step 508, may be repeated until the value of the loss function is less than a threshold value or meets the target requirement, thereby completing the training of the neural network.

[0251] Through the above method 500, the first device can customize the sequence, i.e., the first signal, and can determine the channel reciprocity error based on the first signal sent by the first device to the second device and the second signal received by the first device. Based on the channel reciprocity error, the intermediate gradient provided by the second device can be updated, and the model parameters can be updated based on the updated intermediate gradient. In this way, the gradient error caused by channel reciprocity can be overcome and the method is applicable to more scenarios.

[0252] 7 , which is a schematic flow chart of a method 700 according to another embodiment of the present application. The method 700 is applicable to scenarios where the first signal is a data signal. The method 700 shown in FIG7 may include the following steps.

[0253] 701. The first device sends configuration information to the second device.

[0254] The configuration information includes the type of the first signal and / or the second signal. In the embodiment of the present application, the first signal is a data signal, and the second signal is related to the first signal.

[0255] 702. The first device infers model #1 and obtains the output of model #1.

[0256] Steps 701-702 may refer to steps 501-502 in method 500 and are not described in detail here.

[0257] 703. The first device sends the output of model #1 to the second device.

[0258] For example, the first device maps the output of model #1 (i.e., the data signal) to air interface resources and then transmits it to the second device. Accordingly, the second device receives the data signal after it has passed through the channel. To distinguish them, the data signal sent by the first device is denoted as A1, and the data signal after it has passed through the channel is denoted as A3.

[0259] See Figure 8, which is a schematic diagram of signal transmission and feedback applicable to method 700 according to an embodiment of the present application. As shown in Figure 8, in forward reasoning, the first device maps the data signal to air interface resources, which can then be transmitted to the second device. It can be seen that, unlike method 500, in method 700, there is no need to send an additional first signal; the data signal can perform the functions of the first signal.

[0260] In method 700 , the second device may determine a first intermediate gradient based on the first signal. For example, the second device may determine a loss function based on the first signal (ie, the data signal), and then determine the first intermediate gradient based on the loss function, as shown in steps 704 and 705 below.

[0261] 704 , the second device infers model #2, obtains the output of model #2, and calculates the loss function.

[0262] 705. The second device determines a first intermediate gradient according to the loss function.

[0263] Steps 704-705 may refer to steps 504-505 in method 500 and are not described in detail here.

[0264] 706. The second device sends a pattern of a fourth signal to the first device.

[0265] The pattern of the fourth signal may refer to the pattern of the second signal in method 400 and will not be described in detail here.

[0266] 707 : The second device sends the first intermediate gradient and the fourth signal to the first device.

[0267] For example, the second device maps the first intermediate gradient and the fourth signal to air interface resources and then transmits them to the first device. Accordingly, the first device receives the fourth signal and the first intermediate gradient after the channelization process. To distinguish them, the fourth signal is denoted as A4, and the fourth signal after the channelization process is denoted as the second signal A2.

[0268] As shown in FIG8 , in reverse reasoning, the second device maps the first intermediate gradient and the fourth signal to air interface resources, and then transmits them to the first device.

[0269] The fourth signal is correlated with data signal A3 and can be determined based on data signal A3. Specifically, after receiving data signal A3, the second device determines the fourth signal based on the correlation between the fourth signal and data signal A3. The specific correlation between the fourth signal and data signal A3 can be found in step 506 of method 500 and is not further described here.

[0270] 708. The first device updates the first intermediate gradient according to the first signal and the received fourth signal to obtain a second intermediate gradient.

[0271] That is, the first device updates the first intermediate gradient according to the first signal and the second signal to obtain the second intermediate gradient.

[0272] 709 : The first device updates the parameters of model # 1 according to the second intermediate gradient.

[0273] Steps 708-709 may refer to steps 507-508 in method 500 and are not described in detail here.

[0274] Optionally, the above steps, such as step 702 to step 709, may be repeated until the value of the loss function is less than a threshold value or meets the target requirement, thereby completing the training of the neural network.

[0275] Through the above-described method 700, the data signal can implement the function of the first signal. Specifically, the first device can determine the channel reciprocity error based on the transmitted data signal and the second signal received by the first device. Based on the channel reciprocity error, the first device can then update the intermediate gradient provided by the second device, and then update the model parameters based on the updated intermediate gradient. In this way, the gradient error caused by channel reciprocity can be overcome.

[0276] 9 is a schematic flow chart of a method 900 according to another embodiment of the present application. The method 900 is applicable to a scenario where the first signal is a reference signal. The method 900 shown in FIG9 may include the following steps.

[0277] 901. The first device sends configuration information to the second device.

[0278] The configuration information includes the type of the first signal and / or the second signal. In the embodiment of the present application, the first signal is a reference signal, and the second signal is related to the first signal.

[0279] 902. The first device infers model #1 and obtains the output of model #1.

[0280] Steps 901-902 may refer to steps 501-502 in method 500 and are not described in detail here.

[0281] 903. The first device sends the output of model #1 and the reference signal to the second device.

[0282] For example, the first device maps the output of model #1 (i.e., the data signal) and the reference signal (i.e., the first signal) to air interface resources and then transmits them to the second device. Accordingly, the second device receives the data signal and reference signal after they have been processed through the channel. For example, the reference signal is a DMRS. To distinguish them, the reference signal sent by the first device is denoted as A1, and the reference signal after they have been processed through the channel is denoted as A3.

[0283] See Figure 10, which is a schematic diagram of signal transmission and feedback applicable to method 900 according to an embodiment of the present application. As shown in Figure 10, in forward reasoning, the first device maps the data signal and reference signal (such as DMRS) to air interface resources, which can then be transmitted to the second device. It can be seen that the difference from method 500 (the first signal is a custom sequence) and method 700 (the first signal is a data signal) is that in method 900, the first signal is a reference signal.

[0284] In method 900, the second device may determine a first intermediate gradient based on the first signal (i.e., the reference signal) and the data signal. For example, the second device may determine a loss function based on the data signal, and then determine the first intermediate gradient based on the loss function, as described in steps 904 and 905 below. Furthermore, the second device may update the first intermediate gradient determined based on the loss function based on the reference signal, as described in step 906.

[0285] 904 , the second device infers model #2, obtains the output of model #2, and calculates the loss function.

[0286] 905. The second device determines a first intermediate gradient according to the loss function.

[0287] Steps 904-905 may refer to steps 504-505 in method 500 and are not described in detail here.

[0288] 906. The second device performs channel estimation according to the received reference signal, and updates the first intermediate gradient according to a result of the channel estimation.

[0289] The second device may perform channel estimation based on the received reference signal A3 and modify the first intermediate gradient (ie, the first intermediate gradient determined according to the data signal, ie, the first intermediate gradient determined in step 905) according to the channel estimation result.

[0290] 907 : The second device sends a pattern of a fourth signal to the first device.

[0291] 908. The second device sends the first intermediate gradient and the fourth signal to the first device.

[0292] For example, the second device maps the first intermediate gradient and the fourth signal to air interface resource #2 and then transmits them to the first device. Accordingly, the first device receives the fourth signal and the first intermediate gradient after the channelization process. To distinguish them, the fourth signal is denoted as A4, and the fourth signal after the channelization process is denoted as second signal A2.

[0293] As shown in FIG. 10 , in reverse reasoning, the second device maps the first intermediate gradient and the fourth signal to air interface resources, and then transmits them to the first device.

[0294] The fourth signal is correlated with reference signal A3 and can be determined based on reference signal A3. Specifically, after receiving reference signal A3, the second device determines the fourth signal based on the correlation between the fourth signal and reference signal A3. The specific correlation between the fourth signal and reference signal A3 can be found in step 506 of method 500 and is not further described here.

[0295] 909 : The first device updates the first intermediate gradient according to the reference signal and the received fourth signal to obtain a second intermediate gradient.

[0296] That is, the first device updates the first intermediate gradient according to the reference signal and the second signal to obtain the second intermediate gradient.

[0297] 910. The first device updates the parameters of model #1 according to the second intermediate gradient.

[0298] Steps 909-910 may refer to steps 507-508 in method 500 and are not described in detail here.

[0299] Optionally, the above steps, such as step 902 to step 910, may be repeated until the value of the loss function is less than a threshold value or meets the target requirement, thereby completing the training of the neural network.

[0300] Through the above-described method 700, the reference signal can implement the function of the first signal. Specifically, the first device can determine the channel reciprocity error based on the reference signal sent and the second signal received by the first device. Based on this channel reciprocity error, the intermediate gradient provided by the second device can be updated, and the model parameters can be updated based on the updated intermediate gradient. In this way, the gradient error caused by channel reciprocity can be overcome.

[0301] The above describes in detail with reference to Figures 4 to 10 that the first signal is correlated with the second signal, and the first device determines the channel reciprocity error based on the first signal and the second signal, and then updates the correlation scheme of the first intermediate gradient. The following introduces another scheme with reference to Figures 11 to 13, that is, the first device and the second device can perform channel estimation separately, and then update the correlation scheme of the first intermediate gradient.

[0302] Referring to Figure 11, Figure 11 is a schematic diagram of a neural network training method 1100 provided in an embodiment of the present application. The method 1100 shown in Figure 11 may include the following steps.

[0303] 1101. A first device sends a first reference signal.

[0304] Correspondingly, the second device receives the first reference signal. It is understood that in actual transmission, the signal may change after passing through the channel. To distinguish, the first reference signal received by the second device is called the third reference signal. That is, the third reference signal is the signal of the first reference signal after passing through the channel.

[0305] The first reference signal may be, for example, DMRS, CSI-RS, SRS, etc.

[0306] 1102. A first device receives a first intermediate gradient and a second reference signal, where the first intermediate gradient is determined based on the first reference signal.

[0307] Accordingly, the second device transmits the first intermediate gradient and the second reference signal. It is understood that in actual transmission, signals may change after passing through the channel. To distinguish them, the second reference signal transmitted by the second device is referred to as the fourth reference signal. That is, the second reference signal is the signal after the fourth reference signal has passed through the channel.

[0308] For the intermediate gradient, please refer to the previous terminology explanation and will not be elaborated here.

[0309] The first intermediate gradient is determined based on the first reference signal, which can be understood as the second device determining the first intermediate gradient based on the first reference signal. Specifically, after receiving the first reference signal, the second device can perform channel estimation based on the first reference signal and determine the first intermediate gradient based on the channel estimation results. For example, based on the channel estimation results, the second device can determine the channel conditions and, based on the channel conditions, compensate the first intermediate gradient determined based on the loss function to obtain the compensated first intermediate gradient, and then send the compensated first intermediate gradient to the first device.

[0310] Considering that the signal may change after passing through the channel, the first intermediate gradient is determined based on the first reference signal. Alternatively, the first intermediate gradient can be determined based on the third reference signal (i.e., the first reference signal after passing through the channel). Specifically, the first device sends the first reference signal to the second device. After passing through the channel, the second device receives the third reference signal. The second device performs channel estimation based on the received third reference signal and determines the first intermediate gradient based on the channel estimation results.

[0311] The second reference signal may be, for example, DMRS, CSI-RS, SRS, etc. The second reference signal and the first reference signal may be independent of each other, that is, the reference signal used for channel estimation of the first device and the reference signal used for channel estimation of the second device may be independent of each other.

[0312] 1103. The first device performs channel estimation based on the second reference signal and updates the first intermediate gradient based on the channel estimation result to obtain a second intermediate gradient, and the second intermediate gradient is used to update the neural network parameters.

[0313] Based on the embodiment of the present application, the second device performs channel estimation based on the reference signal sent by the first device, and then determines the intermediate gradient based on the result of the channel estimation, and sends the intermediate gradient to the first device; the first device performs channel estimation based on the reference signal sent by the second device, and then updates the received intermediate gradient based on the result of the channel estimation, so that the intermediate gradient takes into account the channel reciprocity error, thereby overcoming the gradient error caused by channel reciprocity.

[0314] Optionally, method 1100 further includes: the first device sending first indication information, where the first indication information indicates a type of the first reference signal and / or the second reference signal. Taking Table 1 as an example, in this embodiment of the present application, the first signal and the second signal are of type D. For details about this solution, please refer to the relevant description in method 400 and will not be repeated here.

[0315] Optionally, the method 1100 further includes: the first device receiving second indication information, where the second indication information indicates a pattern of a second reference signal. For this solution, reference may be made to the relevant description in the method 400, which will not be repeated here.

[0316] For ease of understanding, a possible process applicable to method 1100 is introduced below.

[0317] Referring to Figure 12, Figure 12 is a schematic flow chart of a method 1200 provided according to another embodiment of the present application. Method 1200 is applicable to a scenario where both the first signal and the second signal are reference signals. Method 1200 shown in Figure 12 may include the following steps.

[0318] 1201. The first device sends configuration information to the second device.

[0319] The configuration information includes the type of the first signal and / or the second signal. In the embodiment of the present application, the first signal is a first reference signal, and the second signal is a second reference signal.

[0320] 1202. The first device infers model #1 and obtains the output of model #1.

[0321] Steps 1201-1202 may refer to steps 501-502 in method 500 and are not described in detail here.

[0322] 1203. The first device sends the output of model #1 and the first reference signal to the second device.

[0323] For example, the first device maps the output of model #1 (i.e., the data signal) and the first reference signal to air interface resources and then transmits them to the second device. Accordingly, the second device receives the data signal and the first reference signal after undergoing channel processing. For example, the first reference signal is a DMRS. For differentiation, the first reference signal after undergoing channel processing is referred to as the third reference signal.

[0324] 13 is a schematic diagram of signal transmission and feedback applicable to method 1200 of an embodiment of the present application. As shown in FIG13 , in forward reasoning, the first device maps the data signal and the first reference signal to air interface resources, which can then be transmitted to the second device.

[0325] 1204. The second device infers model #2, obtains the output of model #2, and calculates the loss function.

[0326] 1205. The second device determines a first intermediate gradient according to the loss function.

[0327] Steps 1204-1205 may refer to steps 504-505 in method 500 and are not described in detail here.

[0328] 1206. The second device performs channel estimation according to the received first reference signal, and updates the first intermediate gradient according to a result of the channel estimation.

[0329] That is, the second device performs channel estimation according to the third reference signal.

[0330] Step 1206 may refer to step 906 in method 900 and will not be described in detail here.

[0331] 1207. The second device sends the first intermediate gradient and the fourth reference signal to the first device.

[0332] For example, the second device maps the first intermediate gradient and the fourth reference signal to air interface resources and transmits them to the first device. Accordingly, the first device receives the fourth reference signal and the first intermediate gradient after the channel has been processed. For differentiation, the fourth reference signal after the channel has been processed is referred to as the second reference signal.

[0333] As shown in FIG. 13 , in reverse reasoning, the second device maps the first intermediate gradient and the fourth reference signal to air interface resources, and then transmits them to the first device.

[0334] Optionally, before step 1207, method 1200 further includes: the second device sending a pattern of a fourth reference signal to the first device.

[0335] 1208. The first device performs channel estimation according to the received fourth reference signal, and updates the first intermediate gradient according to a result of the channel estimation to obtain a second intermediate gradient.

[0336] That is, the first device performs channel estimation according to the second reference signal.

[0337] 1209. The first device updates the parameters of model #1 according to the second intermediate gradient.

[0338] Optionally, the above steps, such as step 1202 to step 1209, may be repeated until the value of the loss function is less than a threshold value or meets the target requirement, thereby completing the training of the neural network.

[0339] Through the above-mentioned method 1200, the second device performs channel estimation based on the reference signal sent by the first device, and then determines the intermediate gradient based on the result of the channel estimation, and sends the intermediate gradient to the first device; the first device performs channel estimation based on the reference signal sent by the second device, and then updates the received intermediate gradient based on the result of the channel estimation, so that the intermediate gradient takes into account the channel reciprocity error, thereby overcoming the gradient error caused by the channel reciprocity.

[0340] The two solutions proposed in the embodiments of the present application are described above in conjunction with Figures 4 to 12. Through the solutions provided in the embodiments of the present application, model training can be achieved even in scenarios with poor channel reciprocity. Furthermore, using the solutions provided in the embodiments of the present application, the loss function can be reduced at the same rate as the ideal baseline.

[0341] See Figure 14, which is a schematic diagram of the loss function decrease when using the solution provided by an embodiment of the present application. In Figure 14, the horizontal axis represents the number of updates, and the vertical axis represents the loss function. As can be seen from Figure 14, using the solution provided by an embodiment of the present application, the loss function can be reduced at the same rate as the ideal baseline.

[0342] The following describes a signal processing process applicable to the embodiments of the present application.

[0343] After the embodiment of the present application, a signal mapping module and a signal demapping module may be added, wherein the signal mapping module is used to map signals, and the signal demapping module is used to receive signals.

[0344] Optionally, in a scenario where a first device sends a signal to a second device, a first signal mapping module is added to the first device; correspondingly, a first signal demapping module is added to the second device. It is understood that if the first signal is a data signal, the first signal mapping module and the first signal demapping module may not be added.

[0345] For example, on the first device side, after the neural network deployed in the first device outputs the data signal and maps the data signal, the first signal can be mapped through the first signal mapping module, and then the first signal is sent out through the waveform module; correspondingly, on the second device side, after the air interface signal (such as the first signal and the data signal) passes through the waveform decoding module, the first signal is received through the first signal demapping module, and the data signal is received through the data demapping module.

[0346] Optionally, in a scenario where the second device sends a signal to the first device, a second signal mapping module is added on the second device side; correspondingly, a second signal demapping module is added on the first device side.

[0347] For example, on the second device side, a second signal mapping module can be added after the intermediate gradient mapping; correspondingly, on the first device side, the air interface signal (such as the second signal and the intermediate gradient signal) passes through the waveform decoding module, and then the second signal is received through the second signal demapping module, and the intermediate gradient is received through the intermediate gradient demapping module.

[0348] See Figure 15, which is a schematic diagram of signal processing applicable to the embodiments of the present application. For ease of description, the first device sending a signal to the second device is called the downlink phase, and the second device sending a signal to the first device is called the uplink phase. In the following embodiments, unless otherwise specified, z represents the transmitted signal and r represents the received signal. represents the output of the neural network model, that is, the output data signal, A1 represents the first signal, A2 represents the second signal, A3 represents the third signal, and A4 represents the fourth signal. In addition, in the following embodiments, unless otherwise specified, k, n, and l represent sequence numbers (or the sequence numbers of each symbol in the signal). For example, n in z(n) represents the sequence number of the transmitted signal, and k in A1(k) represents the sequence number of the first signal A1. The l in represents the sequence number of the neural network model output, that is, the sequence number of the data signal.

[0349] As shown in FIG15 , the first device may perform the following steps in the downlink phase.

[0350] (1) Obtain training data x. Optionally, the first device further determines a first signal pattern and a first signal.

[0351] (2) Input the training data x into the NN encoder. Assume the output x': x' = nn e (x).

[0352] (3) Data mapping and first signal mapping. Specifically, the output x' in step (2) is mapped to the air interface resource. Taking Table 1 as an example, if the type of the first signal and the second signal belongs to type A, type B, or type D, the transmitted signal z(n) is: z(n) = A1(k) or If the first signal and the second signal are of type B, then

[0353] (5) Sending a waveform (WF). Specifically, the data signal mapped in step (3) and the first signal in step (4) are modulated and mapped into a waveform and sent out.

[0354] The second device may perform the following steps in the downlink phase.

[0355] (6) The air interface signal (such as the third signal A3 (ie, the first signal after passing through the channel) and the data signal) passes through the De-Waveform Function (De-WF) module.

[0356] (7) Demapping the third signal and data. Specifically, if it is the third signal, then A3 = y(n); if it is a data signal, then r(l) = y(n). Where r represents the received data, and y represents the downlink data sent by the first device.

[0357] (8) The data passes through the NN decoder. For example, r'=nn d (r).

[0358] (10) Calculate the loss function.

[0359] The second device may perform the following steps in the uplink phase.

[0360] (11) Calculating the intermediate gradient. Specifically, the second device determines the intermediate gradient based on the loss function calculated in step (10).

[0361] (12) Determine the NN decoder gradient based on the intermediate gradient.

[0362] (13) Mapping of the intermediate gradient (i.e., the first intermediate gradient described above) and the fourth signal. Specifically, the intermediate gradient and the fourth signal are mapped to the air interface resources. As an example, the signal z(n) sent by the second device is: if it is the fourth signal, then z(n) = f(A4(k)); if it is the intermediate gradient, then Wherein, g represents a function, L represents a loss function, and Y represents the input of a neural network model (such as a neural network model deployed in the second device).

[0363] (14) Send waveform.

[0364] The first device may perform the following steps in the uplink phase.

[0365] (15) The air interface signal (such as the second signal (ie, the fourth signal after passing through the channel) and the intermediate gradient signal) passes through the waveform decoding module.

[0366] (16) Demapping the second signal and the intermediate gradient. For example, if it is the second signal, then A2(k) = f(y(n)); if it is the intermediate gradient, then r(l) = g(y(n)). Where f and g represent functions. y represents the uplink data sent by the second device.

[0367] (17) Intermediate gradient update. Specifically, the received intermediate gradient is updated based on the first signal and the second signal. For example, the intermediate gradient correction value e = E(A1, A2). Where E represents a function, i.e., a function used to calculate the channel reciprocity error. As an example, E(A1, A2) = A2A1.

[0368] (18) Calculate the NN encoder update gradient. Specifically, update the NN encoder parameters according to the updated intermediate gradient and calculate the NN encoder gradient.

[0369] It is understood that the above steps are merely examples and are not limiting.

[0370] It can be understood that some optional features in the various embodiments of the present application may not depend on other features in certain scenarios, and may also be combined with other features in certain scenarios, without limitation.

[0371] It is also understood that in some of the above embodiments, sending information is mentioned multiple times. Taking A sending information to B as an example, A sending information to B may include A sending information directly to B or A sending information to B through other devices or network elements, and there is no limitation on this.

[0372] It can also be understood that the solutions in the various embodiments of the present application can be reasonably combined and used, and the explanations or descriptions of the various terms appearing in the embodiments can be referenced or explained with each other in the various embodiments, without limitation to this.

[0373] The method provided in the embodiments of the present application is described in detail above with reference to Figures 4 to 15 . Below, the apparatus provided in the embodiments of the present application is described in detail with reference to Figures 16 to 18 . It should be understood that the description of the apparatus embodiment corresponds to the description of the method embodiment. Therefore, for matters not described in detail, reference can be made to the method embodiment above, and for the sake of brevity, they will not be repeated here.

[0374] Referring to Figure 16 , Figure 16 is a schematic diagram of a communication device 1600 provided in an embodiment of the present application. Device 1600 includes a processing unit 1620. Processing unit 1620 can be used to perform processing, such as updating intermediate gradients. Optionally, device 1600 also includes a transceiver unit 1610. Transceiver unit 1610 can be used to implement corresponding communication functions. Transceiver unit 1610 can also be referred to as a communication interface or communication unit.

[0375] Optionally, the device 1600 may further include a storage unit, which may be used to store instructions and / or data. The processing unit 1620 may read the instructions and / or data in the storage unit so that the device implements the aforementioned method embodiment.

[0376] In a first possible design, the device 1600 may be the communication device in the aforementioned embodiment (such as the first device in Figure 4, Figure 5, Figure 7, or Figure 9), and the device 1600 may implement the steps or processes corresponding to those performed by the communication device in the above method embodiment. The transceiver unit 1610 may be used to perform the transceiver-related operations (such as the operations of sending and / or receiving data or messages) of the communication device in the above method embodiment, and the processing unit 1620 may be used to perform the processing-related operations of the communication device in the above method embodiment, or operations other than transceiver (such as operations other than sending and / or receiving data or messages).

[0377] In one possible implementation, the transceiver unit 1610 is used to send a first signal; the transceiver unit 1610 is also used to receive a first intermediate gradient and a second signal, where the second signal is related to the first signal; and the processing unit 1620 is used to update the first intermediate gradient based on the first signal and the second signal to obtain a second intermediate gradient, where the second intermediate gradient is used to update the neural network parameters.

[0378] Optionally, the first signal is any one of the following: a custom sequence, a reference signal, or a data signal.

[0379] Optionally, the first signal is a custom sequence, and the first intermediate gradient is determined based on the data signal; or, the first signal is a reference signal, and the first intermediate gradient is determined based on the data signal and the reference signal; or, the first signal is a data signal, and the first intermediate gradient is determined based on the first signal.

[0380] Optionally, the processing unit 1620 is specifically configured to determine a channel reciprocity error value according to the first signal and the second signal; and update the first intermediate gradient according to the channel reciprocity error value.

[0381] Optionally, the second signal and the first signal satisfy:

[0382] A2=f(A1), or A2=f(A1 * )

[0383] Among them, A1 represents the first signal, A2 represents the second signal, A1 * Indicates the conjugate operation on A1.

[0384] Optionally, the second signal and the first signal satisfy any of the following:

[0385] A2=αA1,A2=αA1 * , or,

[0386] Among them, A1 represents the first signal, A2 represents the second signal, A1 * represents the conjugate operation on A1, |A1| represents the independent normalization of the power of each symbol in A1, ||A1||2 represents the normalization of the power of some or all symbols in A1, and α is a constant.

[0387] Optionally, the transceiver unit 1610 is further configured to send first indication information, where the first indication information indicates a type of the first signal and / or the second signal.

[0388] Optionally, the transceiver unit 1610 is further configured to receive second indication information, where the second indication information indicates a pattern of the second signal.

[0389] Optionally, the pattern of the second signal includes at least one of the following: a mask of the second signal, resources occupied by the second signal, and a sequence number of the second signal.

[0390] In a second possible design, the device 1600 may be the communication device in the aforementioned embodiment (such as the first device in FIG. 11 or FIG. 12 ), and the device 1600 may implement the steps or processes corresponding to those performed by the communication device in the above method embodiment. The transceiver unit 1610 may be used to perform the transceiver-related operations (such as the operations of sending and / or receiving data or messages) of the communication device in the above method embodiment, and the processing unit 1620 may be used to perform the processing-related operations of the communication device in the above method embodiment, or operations other than transceiver (such as operations other than sending and / or receiving data or messages).

[0391] In one possible implementation, the transceiver unit 1610 is used to send a first reference signal; the transceiver unit 1610 is also used to receive a first intermediate gradient and a second reference signal, where the first intermediate gradient is determined based on the first reference signal; the processing unit 1620 is used to perform channel estimation based on the second reference signal, and update the first intermediate gradient based on the channel estimation result to obtain a second intermediate gradient, where the second intermediate gradient is used to update the neural network parameters.

[0392] Optionally, the transceiver unit 1610 is further configured to send first indication information, where the first indication information indicates a type of the first reference signal and / or the second reference signal.

[0393] Optionally, the transceiver unit 1610 is further configured to receive second indication information, where the second indication information indicates a pattern of a second reference signal.

[0394] Optionally, the pattern of the second reference signal includes at least one of the following: a mask of the second reference signal, resources occupied by the second reference signal, and a sequence number of the second reference signal.

[0395] In a third possible design, the device 1600 may be the communication device in the aforementioned embodiment (such as the second device in Figure 4, Figure 5, Figure 7, or Figure 9), and the device 1600 may implement the steps or processes corresponding to those performed by the communication device in the above method embodiment. The transceiver unit 1610 may be used to perform the transceiver-related operations (such as the operations of sending and / or receiving data or messages) of the communication device in the above method embodiment, and the processing unit 1620 may be used to perform the processing-related operations of the communication device in the above method embodiment, or operations other than transceiver (such as operations other than sending and / or receiving data or messages).

[0396] In one possible implementation, the transceiver unit 1610 is used to receive a first signal; the transceiver unit 1610 is also used to send a first intermediate gradient and a second signal, where the second signal is related to the first signal, the second signal is used to update the first intermediate gradient, and the first intermediate gradient is used to update the neural network parameters.

[0397] Optionally, the first signal is any one of the following: a custom sequence, a reference signal, or a data signal.

[0398] Optionally, the first signal is a custom sequence, and the first intermediate gradient is determined based on the data signal; or, the first signal is a reference signal, and the first intermediate gradient is determined based on the data signal and the reference signal; or, the first signal is a data signal, and the first intermediate gradient is determined based on the first signal.

[0399] Optionally, the second signal and the first signal satisfy:

[0400] A2=f(A1), or A2=f(A1 * )

[0401] Among them, A1 represents the first signal, A2 represents the second signal, A1 * Indicates the conjugate operation on A1.

[0402] Optionally, the second signal and the first signal satisfy any of the following:

[0403] A2=αA1,A2=αA1 * , or,

[0404] Among them, A1 represents the first signal, A2 represents the second signal, A1 * represents the conjugate operation on A1, |A1| represents the independent normalization of the power of each symbol in A1, ||A1||2 represents the normalization of the power of some or all symbols in A1, and α is a constant.

[0405] Optionally, the transceiver unit 1610 is further configured to receive first indication information, where the first indication information indicates a type of the first signal and / or the second signal.

[0406] Optionally, the transceiver unit 1610 is further configured to send second indication information, where the second indication information indicates a pattern of the second signal.

[0407] Optionally, the pattern of the second signal includes at least one of the following: a mask of the second signal, resources occupied by the second signal, and a sequence number of the second signal.

[0408] In a fourth possible design, the device 1600 may be the communication device in the aforementioned embodiment (such as the second device in FIG. 11 or FIG. 12 ), and the device 1600 may implement the steps or processes corresponding to those performed by the communication device in the above method embodiment. The transceiver unit 1610 may be used to perform the transceiver-related operations (such as the operations of sending and / or receiving data or messages) of the communication device in the above method embodiment, and the processing unit 1620 may be used to perform the processing-related operations of the communication device in the above method embodiment, or operations other than transceiver (such as operations other than sending and / or receiving data or messages).

[0409] In one possible implementation, the transceiver unit 1610 is used to receive a first reference signal; the processing unit 1620 is used to perform channel estimation based on the first reference signal and determine a first intermediate gradient based on the result of the channel estimation; the transceiver unit 1610 is also used to send the first intermediate gradient and a second reference signal, the second reference signal is used to update the first intermediate gradient, and the first intermediate gradient is used to update the neural network parameters.

[0410] Optionally, the transceiver unit 1610 is further configured to receive first indication information, where the first indication information indicates a type of the first reference signal and / or the second reference signal.

[0411] Optionally, the transceiver unit 1610 is further configured to send second indication information, where the second indication information indicates a pattern of a second reference signal.

[0412] Optionally, the pattern of the second signal includes at least one of the following: a mask of the second signal, resources occupied by the second signal, and a sequence number of the second signal.

[0413] It should be understood that the specific process of each unit executing the above corresponding steps has been described in detail in the above method embodiment, and for the sake of brevity, it will not be repeated here.

[0414] It should also be understood that the device 1600 here is embodied in the form of a functional unit. The term "unit" here can refer to an application specific integrated circuit (ASIC), an electronic circuit, a processor (such as a shared processor, a dedicated processor or a group processor, etc.) and a memory for executing one or more software or firmware programs, a combined logic circuit and / or other suitable components that support the described functions. In an optional example, those skilled in the art will understand that the device 1600 can be specifically the communication device in the above-mentioned embodiment, and can be used to execute the various processes and / or steps corresponding to the communication device in the above-mentioned method embodiments. To avoid repetition, they will not be described here.

[0415] The apparatus 1600 of each of the above-described solutions has the function of implementing the corresponding steps performed by the communication device in the above-described method. The functions can be implemented by hardware, or by hardware executing corresponding software implementations. The hardware or software includes one or more modules corresponding to the above-described functions; for example, the transceiver unit can be replaced by a transceiver (for example, the transmitting unit in the transceiver unit can be replaced by a transmitter, and the receiving unit in the transceiver unit can be replaced by a receiver), and other units, such as the processing unit, can be replaced by a processor to respectively perform the transceiver operations and related processing operations in each method embodiment.

[0416] In addition, the transceiver unit 1610 may also be a transceiver circuit (for example, may include a receiving circuit and a transmitting circuit), and the processing unit may be a processing circuit.

[0417] It should be noted that the apparatus in FIG16 may be the communication device in the aforementioned embodiment, or may be a chip or chip system, such as a system on chip (SoC). The transceiver unit may be an input / output circuit or a communication interface; the processing unit may be a processor, microprocessor, or integrated circuit integrated on the chip. This is not limited here.

[0418] Referring to FIG. 17 , FIG. 17 is a schematic diagram of another communication device 1700 provided in an embodiment of the present application. The device 1700 includes a processor 1710 coupled to a memory 1720. The memory 1720 is configured to store computer programs or instructions and / or data. The processor 1710 is configured to execute the computer programs or instructions stored in the memory 1720, or read data stored in the memory 1720, to perform the methods described in the above method embodiments.

[0419] Optionally, there are one or more processors 1710 .

[0420] Optionally, the memory 1720 is one or more.

[0421] Optionally, the memory 1720 is integrated with the processor 1710 or provided separately.

[0422] Optionally, as shown in Figure 17, the device 1700 further includes a transceiver 1730, which is used to receive and / or send signals. For example, the processor 1710 is used to control the transceiver 1730 to receive and / or send signals.

[0423] As an example, the processor 1710 may have the function of the processing unit 1620 shown in FIG. 16 , the memory 1720 may have the function of a storage unit, and the transceiver 1730 may have the function of the transceiver unit 1610 shown in FIG. 16 .

[0424] As a solution, the device 1700 is used to implement the operations performed by the communication device in the above various method embodiments.

[0425] For example, the processor 1710 is configured to execute computer programs or instructions stored in the memory 1720 to implement relevant operations of the communication device in the above various method embodiments.

[0426] It should be understood that the processor mentioned in the embodiments of the present application may be a central processing unit (CPU), or may be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc.

[0427] It should also be understood that the memory mentioned in the embodiments of the present application may be a volatile memory and / or a non-volatile memory. Among them, the non-volatile memory may be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory. The volatile memory may be a random access memory (RAM). For example, RAM can be used as an external cache. By way of example and not limitation, RAM includes the following forms: static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM), and direct rambus RAM (DR RAM).

[0428] It should be noted that when the processor is a general-purpose processor, DSP, ASIC, FPGA or other programmable logic device, discrete gate or transistor logic device, discrete hardware component, the memory (storage module) can be integrated into the processor.

[0429] It should also be noted that the memory described herein is intended to include, but is not limited to, these and any other suitable types of memory.

[0430] 18 , which is a schematic diagram of a chip system 1800 according to an embodiment of the present application. The chip system 1800 (or also referred to as a processing system) includes a logic circuit 1810 and an input / output interface 1820 .

[0431] Logic circuit 1810 may be a processing circuit within chip system 1800. Logic circuit 1810 may be coupled to a storage unit and invoke instructions within the storage unit, enabling chip system 1800 to implement the methods and functions of various embodiments of the present application. Input / output interface 1820 may be an input / output circuit within chip system 1800, outputting information processed by chip system 1800 or inputting data or signaling information to be processed into chip system 1800 for processing.

[0432] As a solution, the chip system 1800 is used to implement the operations performed by the communication device (such as the first device, and such as the second device) in the above various method embodiments.

[0433] For example, the logic circuit 1810 is used to implement the processing-related operations performed by the communication device (such as the first device, and also the second device) in the above method embodiments; the input / output interface 1820 is used to implement the sending and / or receiving-related operations performed by the communication device (such as the first device, and also the second device) in the above method embodiments.

[0434] An embodiment of the present application further provides a computer-readable storage medium storing computer instructions for implementing the methods executed by a communication device (such as the first device or the second device) in the above-mentioned method embodiments.

[0435] For example, when the computer program is executed by a computer, the computer can implement the method performed by the communication device (such as the first device, and such as the second device) in each embodiment of the above method.

[0436] An embodiment of the present application further provides a computer program product comprising instructions, which, when executed by a computer, implement the methods performed by a communication device (such as the first device or the second device) in the above-mentioned method embodiments.

[0437] The present application also provides a communication system, which includes the first device and / or second device in each of the above embodiments. For example, the system includes the first device and second device in Figure 4, Figure 5, Figure 7, or Figure 9. For another example, the system includes the first device and second device in Figure 11 or Figure 12.

[0438] The explanation of the relevant contents and beneficial effects of any of the above-mentioned devices can be referred to the corresponding method embodiments provided above, which will not be repeated here.

[0439] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0440] In the above embodiments, it 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 application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. For example, the computer can be a personal computer, a server, or a network device, etc. The computer instructions can 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 can be transmitted from one website, computer, server or data center to another website, 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.) mode. The computer-readable storage medium can 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 integrations. The available medium may be a magnetic medium (e.g., a floppy disk, a hard disk, a magnetic tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid state disk (SSD)). For example, the aforementioned available medium includes, but is not limited to, various media that can store program code, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0441] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

Claims

1. A method for training a neural network, characterized in that: include: sending a first signal; receiving a first intermediate gradient and a second signal, the second signal being related to the first signal; The first intermediate gradient is updated according to the first signal and the second signal to obtain a second intermediate gradient, and the second intermediate gradient is used to update the neural network parameters.

2. The method according to claim 1, characterized in that The first signal is any one of the following: a custom sequence, a reference signal, or a data signal.

3. The method according to claim 2, characterized in that The first signal is the custom sequence, and the first intermediate gradient is determined according to a data signal; or, The first signal is the reference signal, and the first intermediate gradient is determined according to the data signal and the reference signal; or, The first signal is the data signal, and the first intermediate gradient is determined according to the first signal.

4. The method according to any one of claims 1 to 3, characterized in that The updating of the first intermediate gradient according to the first signal and the second signal comprises: Determining a channel reciprocity error value according to the first signal and the second signal; The first intermediate gradient is updated according to the channel reciprocity error value.

5. The method according to any one of claims 1 to 4, characterized in that The second signal and the first signal satisfy: A2 = f(A1), or A2 = f(A1 * ) Wherein, A1 represents the first signal, A2 represents the second signal, and A1 * It means to take the conjugate operation on A1.

6. The method according to any one of claims 1 to 5, characterized in that The second signal and the first signal satisfy any of the following conditions: <h2 style=";text-align:left;direction:ltr">A2 = αA1 = A2 = αA1<h2 style=";text-align:left;direction:ltr"> * <h2 style=";text-align:left;direction:ltr"> ,<h2 style=";text-align:left;direction:ltr"> or, Wherein, A1 represents the first signal, A2 represents the second signal, and A1 * represents a conjugate operation on A1, |A1| represents independent normalization of the power of each symbol in A1, ||A1||2 represents normalization of the power of some or all symbols in A1, and α is a constant.

7. The method according to any one of claims 1 to 6, characterized in that The method further comprises: First indication information is sent, where the first indication information indicates a type of the first signal and / or the second signal.

8. The method according to any one of claims 1 to 7, characterized in that Before receiving the first intermediate gradient and the second signal, the method further includes: Second indication information is received, where the second indication information indicates a pattern of the second signal.

9. The method according to claim 8, characterized in that The pattern of the second signal includes at least one of the following: The mask of the second signal, the resources occupied by the second signal, and the sequence number of the second signal.

10. A method for training a neural network, characterized in that: include: receiving a first signal; A first intermediate gradient and a second signal are sent, wherein the second signal is related to the first signal, the second signal is used to update the first intermediate gradient, and the first intermediate gradient is used to update the neural network parameters.

11. The method according to claim 10, characterized in that The first signal is any one of the following: a custom sequence, a reference signal, or a data signal.

12. The method according to claim 11, characterized in that The first signal is the custom sequence, and the first intermediate gradient is determined according to a data signal; or, The first signal is the reference signal, and the first intermediate gradient is determined according to the data signal and the reference signal; or, The first signal is the data signal, and the first intermediate gradient is determined according to the first signal.

13. The method according to any one of claims 10 to 12, characterized in that The second signal and the first signal satisfy: A2 = f(A1), or A2 = f(A1 * ) Wherein, A1 represents the first signal, A2 represents the second signal, and A1 * It means to take the conjugate operation on A1.

14. The method according to any one of claims 10 to 13, characterized in that The second signal and the first signal satisfy any of the following conditions: <h2 style=";text-align:left;direction:ltr">A2 = αA1 = A2 = αA1<h2 style=";text-align:left;direction:ltr"> * <h2 style=";text-align:left;direction:ltr"> ,<h2 style=";text-align:left;direction:ltr"> or, Wherein, A1 represents the first signal, A2 represents the second signal, and A1 * represents a conjugate operation on A1, |A1| represents independent normalization of the power of each symbol in A1, ||A1||2 represents normalization of the power of some or all symbols in A1, and α is a constant.

15. The method according to any one of claims 10 to 14, characterized in that The method further comprises: First indication information is received, where the first indication information indicates a type of the first signal and / or the second signal.

16. The method according to any one of claims 10 to 15, characterized in that The method further comprises: Second indication information is sent, where the second indication information indicates a pattern of the second signal.

17. The method according to claim 16, characterized in that The pattern of the second signal includes at least one of the following: The mask of the second signal, the resources occupied by the second signal, and the sequence number of the second signal.

18. A method for training a neural network, characterized in that: include: sending a first reference signal; receiving a first intermediate gradient and a second reference signal, wherein the first intermediate gradient is determined based on the first reference signal; Channel estimation is performed according to the second reference signal, and the first intermediate gradient is updated according to the channel estimation result to obtain a second intermediate gradient, where the second intermediate gradient is used to update the neural network parameters.

19. The method according to claim 18, characterized in that The method further comprises: First indication information is sent, where the first indication information indicates a type of the first reference signal and / or the second reference signal.

20. The method according to claim 18 or 19, characterized in that Before receiving the first intermediate gradient and the second reference signal, the method further includes: Second indication information is received, where the second indication information indicates a pattern of the second reference signal.

21. The method according to claim 20, characterized in that The pattern of the second reference signal includes at least one of the following: The mask of the second reference signal, the resources occupied by the second reference signal, and the sequence number of the second reference signal.

22. A method for training a neural network, characterized in that: include: receiving a first reference signal; Performing channel estimation according to the first reference signal, and determining a first intermediate gradient according to a result of the channel estimation; The first intermediate gradient and a second reference signal are sent, wherein the second reference signal is used to update the first intermediate gradient, and the first intermediate gradient is used to update the neural network parameters.

23. The method according to claim 22, characterized in that The method further comprises: First indication information is received, where the first indication information indicates a type of the first reference signal and / or the second reference signal.

24. The method according to claim 22 or 23, characterized in that The method further comprises: Second indication information is sent, where the second indication information indicates a pattern of the second reference signal.

25. The method according to claim 24, characterized in that The pattern of the second reference signal includes at least one of the following: The mask of the second reference signal, the resources occupied by the second reference signal, and the sequence number of the second reference signal.

26. A communication device, characterized in that: include: A unit for performing the method as claimed in any one of claims 1 to 9, or a unit for performing the method as claimed in any one of claims 10 to 17, or a unit for performing the method as claimed in any one of claims 18 to 21, or a unit for performing the method as claimed in any one of claims 22 to 25.

27. A communication device, characterized in that: Comprising a processor, the processor is used to execute a computer program or instruction stored in a memory, so that the communication device performs the method of any one of claims 1 to 9, or the communication device performs the method of any one of claims 10 to 17, or the communication device performs the method of any one of claims 18 to 21, or the communication device performs the method of any one of claims 22 to 25.

28. The device according to claim 27, characterized in that The device further comprises the memory and / or the communication interface, wherein the communication interface is coupled to the processor. The communication interface is used to input and / or output information.

29. The device according to claim 27 or 28, characterized in that The device is any one of the following: a communication device, a chip or a chip system.

30. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program or instructions, and when the computer program or instructions are executed on the communication device, the communication device executes the method described in any one of claims 1 to 9, or the communication device executes the method described in any one of claims 10 to 17, or the communication device executes the method described in any one of claims 18 to 21, or the communication device executes the method described in any one of claims 22 to 25.

31. A computer program product, characterized in that The computer program product comprises a computer program or instructions for executing the method as claimed in any one of claims 1 to 9, or the computer program product comprises a computer program or instructions for executing the method as claimed in any one of claims 10 to 17, or the computer program product comprises a computer program or instructions for executing the method as claimed in any one of claims 18 to 21, or the computer program product comprises a computer program or instructions for executing the method as claimed in any one of claims 22 to 25.

Images