Method and device for obtaining neural network

Abstract

The invention provides a method and device for obtaining a neural network, and aims to reduce air interface overhead caused by downloading neural network structures, parameter information and the like by a transmitter and a receiver. The method comprises the steps of determining whether a constructed target neural network meets a preset output condition or not based on a standardized reference data set, at least one standardized reference neural network and a standardized performance judgment criterion; outputting the target neural network under the condition that an output condition is met; under the condition that the output condition is not met, optimizing the target neural network until the obtained target neural network meets the output condition; wherein the target neural network is a transmitting end target neural network applied to a transmitter or a receiving end target neural network applied to a receiver.

Description

[0001] This application is a divisional application of a Chinese patent application with application number CN202010615552.9 and application name “Method and device for obtaining neural network” filed on June 30, 2020. The entire contents are incorporated into this application by reference. Technical Field

[0002] The present application relates to the field of artificial intelligence, and more specifically, to a method and apparatus for obtaining a neural network. Background Art

[0003] In a communication system, a transmitter is used to send signals, and a receiver is used to receive signals sent by the transmitter. Neural networks can be applied to transmitters and receivers to jointly optimize both ends to improve overall performance. Taking channel state information (CSI) compression and reconstruction as an example, the transmitter can compress CSI based on a neural network to generate air interface information. The receiver can parse the air interface information based on a neural network to reconstruct the channel. In one implementation, a neural network at the joint transmitter and a neural network at the receiver can be used to achieve joint optimization of the algorithm to obtain the optimal compression and reconstruction algorithm.

[0004] However, in actual commercial systems, the receiver and transmitter are often provided by different manufacturers. For example, the transmitter is provided by manufacturer A and the receiver is provided by manufacturer B. In order to solve the joint optimization problem, one implementation method is that the transmitter and receiver download the neural network structure and parameter information that matches the current usage scenario respectively, and then perform processing on both ends based on the downloaded neural network. This will cause huge air interface overhead and is greatly limited in actual use. Summary of the invention

[0005] The present application provides a method and device for obtaining a neural network, in order to reduce the huge air interface overhead caused by the transmitter and receiver downloading the neural network structure and parameter information.

[0006] In a first aspect, a method for obtaining a neural network is provided. The method includes: determining whether a constructed target neural network meets a preset output condition based on a standardized reference data set, one or more reference neural networks, and a performance judgment criterion; if the output condition is met, outputting the target neural network, the outputted target neural network is a transmitting target neural network applied to a transmitter or a receiving target neural network applied to a receiver.

[0007] Based on the above technical solution, by introducing standardized reference data sets, standardized reference neural networks and standardized performance judgment criteria, the acquisition device can evaluate the target neural network based on the existing standardized reference neural network, and output it for use only when the target neural network meets the output conditions. The output target neural network can be applied to the transmitter or receiver. This avoids the overhead of the transmitter or receiver downloading the neural network structure and parameter information through the air interface. At the same time, it can also ensure that products from different manufacturers can be interconnected. While ensuring performance, it can also reflect the differentiated design and competitiveness of products from different manufacturers. Therefore, overall, the feasibility of dual network implementation is greatly improved.

[0008] In combination with the first aspect, in some possible implementations of the first aspect, the method further includes: when the output condition is not met, continuing to optimize the target neural network until the obtained target neural network meets the output condition.

[0009] That is, when the output conditions are not met, the target neural network can continue to be optimized. The optimization may include, but is not limited to, adjusting the neural network structure design, adjusting the reference data set, adjusting the neural network training method, adjusting the definition of the cost function, loss function or objective function, changing the initialization method of the neural network, changing the constraints of the neural network parameters, and changing the definition of the activation function of the neural network.

[0010] By optimizing the target neural network, the optimized target neural network can continue to be evaluated based on the above performance judgment criteria, and this cycle is repeated to guide the optimized target neural network to meet the output conditions and then output it.

[0011] In combination with the first aspect, in some possible implementations of the first aspect, the target neural network is used for constellation modulation, channel coding, precoding, or extraction and compression of channel information; the receiving end target neural network is used for constellation demodulation, channel decoding, detection and reception, or reconstruction of channel information.

[0012] In other words, the target neural network can be applied to all stages of communication. The transmitting end target neural network and the receiving end target neural network can be coupled to each other and work in coordination. For example, the transmitting end target neural network and the receiving end target neural network working in coordination can be used for constellation modulation and demodulation, or channel coding and decoding, or precoding and detection reception, or extraction and compression of channel information and reconstruction of channel information.

[0013] In combination with the first aspect, in some possible implementations of the first aspect, the method further includes: determining a configuration suitable for the current application scenario of the target neural network from at least one set of standardized configurations; each set of standardized configurations includes at least one or more of the following: a standardized reference data set, a standardized reference neural network, a standardized data structure, a standardized performance measurement function, a standardized performance judgment criterion, and a standardized achievable capability measurement value.

[0014] Among them, the reference neural network includes a receiving end reference neural network applied in a receiver and / or a transmitting end reference neural network applied in a transmitter; the data structure is used to generate and / or parse air interface information; the performance measurement function is used to generate a performance measurement value, the performance judgment criterion includes a performance indicator and an output condition of the target neural network, and the performance judgment criterion is used to determine whether the target neural network meets the output condition based on the performance measurement value and the performance indicator; the achievable capability measurement value includes a neural network used to select a matching neural network and / or a matching performance indicator.

[0015] It should be understood that the matched neural network described herein may specifically refer to the neural network being applicable to a transmitter or a receiver under given constraints, and the given constraints may include implementation complexity and / or air interface overhead. The matched performance index may specifically refer to the performance index used to evaluate the neural network in the stage of acquiring the neural network.

[0016] Different standardized configurations can be defined for different application scenarios. For example, in different application scenarios, the performance judgment criteria can be different; or, the reference neural network can be different, etc. Based on such a design, the output target neural network can be matched with different application scenarios, which is conducive to obtaining better system performance.

[0017] Optionally, the performance metric function includes one or more of the following: mean square error, normalized mean square error, mean absolute error, maximum absolute error, correlation coefficient, cross entropy, mutual information, bit error rate, or frame error rate.

[0018] The performance metric functions listed above are only examples and should not constitute any limitation to this application. When the performance metric function is used to evaluate the target neural network, one or more of them can be used to evaluate at the same time. This application does not limit this.

[0019] Optionally, the performance indicator includes one or more performance indicators, and the performance judgment criterion includes determining whether the output condition is met based on a comparison between the performance metric value and the one or more performance indicators.

[0020] It should be understood that the performance indicator may be one or more items, and the upper bound and / or lower bound of the corresponding performance indicator may also be one or more items.

[0021] The following shows several examples of evaluating the target neural network in combination with the above performance judgment criteria to determine whether it meets the output conditions.

[0022] In one possible design, the target neural network is a transmitting target neural network, and the one or more reference neural networks are one or more receiving reference neural networks; and the determination of whether the designed target neural network meets a preset output condition based on a standardized reference data set, one or more reference neural networks and a performance judgment criterion includes: inputting the transmitting data obtained from the reference data set into the transmitting target neural network, the transmitting target neural network being used to process the transmitting data according to a predefined data structure to generate air interface information to be sent to the one or more receiving reference neural networks; obtaining one or more output results from the one or more receiving reference neural networks, the one or more output results being obtained by the one or more receiving reference neural networks based on the received air interface information respectively; using the one or more output results and the transmitting data as inputs of the performance measurement function to obtain one or more performance measurement values; and determining whether the target neural network meets the output condition based on the one or more performance measurement values ​​and the performance judgment criterion.

[0023] That is, the transmitting end target neural network and the receiving end reference neural network are cascaded to evaluate the receiving end target neural network.

[0024] In another possible design, the target neural network is an originating target neural network, and the one or more reference neural networks are one or more originating reference neural networks; and the determination of whether the designed target neural network meets a preset output condition based on a standardized reference data set, one or more reference neural networks and a performance judgment criterion includes: inputting the originating data obtained from the reference data set into the originating target neural network, and the originating target neural network is used to process the originating data according to a predefined data structure to generate target air interface information; inputting the originating data into the one or more originating reference neural networks, and the one or more originating reference neural networks are used to process the originating data according to a predefined data structure to generate one or more reference air interface information; using the one or more reference air interface information and the target air interface information as inputs of the performance measurement function to obtain one or more performance measurement values; and determining whether the originating target neural network meets the output condition based on the one or more performance measurement values ​​and the performance judgment criterion.

[0025] That is, the originating reference neural network is used as a reference to evaluate the originating target neural network.

[0026] In another possible design, the target neural network is a receiving-end target neural network, and the one or more reference neural networks are one or more transmitting-end reference neural networks; and the method based on the standardized reference data set, the one or more reference neural networks and the performance judgment criteria to determine whether the designed target neural network meets the preset output condition includes: inputting the transmitting-end data obtained from the reference data set into the one or more transmitting-end reference neural networks, the one or more transmitting-end reference neural networks being used to process the transmitting-end data according to a predefined data structure to obtain one or more air interface information to be sent to the receiving-end target neural network; obtaining one or more output results from the receiving-end target neural network, the one or more output results being output results respectively generated by the receiving-end target neural network according to the one or more air interface information received; using the one or more output results and the transmitting-end data as inputs of the performance metric function to obtain one or more performance metric values; and determining whether the receiving-end target neural network meets the output condition according to the one or more performance metric values ​​and the performance judgment criteria.

[0027] That is, the receiving end target neural network and the transmitting end reference neural network are cascaded to evaluate the receiving end target neural network.

[0028] In another possible design, the target neural network is a receiving-end target neural network, and the one or more reference neural networks include one or more transmitting-end reference neural networks and one or more receiving-end reference neural networks; and the determination of whether the designed target neural network meets a preset output condition based on the standardized reference data set, the one or more reference neural networks and the performance judgment criteria includes: inputting the transmitting-end data obtained from the reference data set into the one or more transmitting-end reference neural networks, the one or more transmitting-end reference neural networks being used to process the transmitting-end data according to a predefined data structure to obtain one or more air interface information to be sent to the receiving-end target neural network and the one or more receiving-end reference neural networks; and obtaining the output of the one or more air interface information from the receiving-end target neural network and the one or more receiving-end reference neural networks. The target neural network obtains one or more target output results, and the one or more target output results are output results generated by the receiving end target neural network according to the one or more air interface information received; obtains one or more reference output results from the one or more receiving end reference neural networks, and the one or more reference output results are output results generated by the one or more receiving end reference neural networks according to the one or more air interface information received; uses the one or more target output results and the one or more reference output results as inputs of the performance metric function to obtain one or more performance metric values; determines whether the receiving end target neural network meets the output condition according to the one or more performance metric values ​​and the performance judgment criterion.

[0029] That is, the receiving end target neural network and the transmitting end reference neural network are cascaded, and the receiving end reference neural network is used as a reference to evaluate the receiving end target neural network.

[0030] In combination with the first aspect, in some possible implementations of the first aspect, the achievable capability metric value includes one or more of the following: the overhead of air interface information, the amount of computation required for the neural network implementation, the amount of parameters stored for the neural network implementation, and the computational accuracy required for the neural network implementation.

[0031] Therefore, based on the implementation complexity of different applications, a matching neural network can be selected to work, and a matching performance indicator can be selected for evaluation.

[0032] In a second aspect, a communication method is provided, the method comprising: a first device determines a scene mode based on an air interface transmission condition between the first device and a second device, the scene mode being used to determine a neural network suitable for current air interface transmission conditions, the neural network being determined from a plurality of pre-designed neural networks, each of the plurality of neural networks matching one or more scene modes; the first device sends indication information of the scene mode to the second device.

[0033] Based on the above method, the first device can select a suitable scenario mode based on the current air interface transmission conditions between the first device and the second device, and indicate the scenario mode to the second device, so that both the first device and the second device can determine which neural network to use based on the same scenario mode, thereby facilitating performance improvement.

[0034] In combination with the second aspect, in some possible implementations of the second aspect, the method further includes: the first device receiving confirmation information from the second device, the confirmation information being used to indicate successful receipt of the indication information of the scene mode.

[0035] The second device may send confirmation information to the first device based on the successful reception of the above-mentioned indication information, so that the first device can perform subsequent processes according to the reception of the indication information. For example, in the case of successful reception, the neural network matching the scene mode is used; in the case of unsuccessful reception, the indication information of the scene mode is resent, and so on. This application does not limit this.

[0036] In combination with the second aspect, in some possible implementations of the second aspect, the method further includes: the first device communicates with the second device using a neural network matching the scene pattern.

[0037] The first device may work with a neural network matching the scene mode based on the successful reception of the above indication information by the second device. The second device may also determine and work with a neural network matching the indicated scene mode based on the successful reception of the above indication information, thereby enabling two mutually coupled neural networks to work in coordination, which is beneficial to improving system performance.

[0038] According to a third aspect, a communication method is provided, the method comprising: a second device receives indication information of a scene mode from a first device, the scene mode is used to determine a neural network suitable for current air interface transmission conditions, the neural network is determined from a plurality of pre-designed neural networks, each of the plurality of neural networks matches one or more scene modes; the second device determines the scene mode according to the indication information of the scene mode; and determines a neural network suitable for the air interface transmission conditions according to the scene mode.

[0039] Based on the above method, the first device can select a suitable scenario mode based on the current air interface transmission conditions between the first device and the second device, and indicate the scenario mode to the second device, so that both the first device and the second device can determine which neural network to use based on the same scenario mode, thereby facilitating performance improvement.

[0040] In combination with the third aspect, in some possible implementations of the third aspect, the method further includes: the second device sends confirmation information to the first device, where the confirmation information is used to indicate successful reception of the indication information of the scene mode.

[0041] The second device may send confirmation information to the first device based on the successful reception of the above-mentioned indication information, so that the first device can perform subsequent processes according to the reception of the indication information. For example, in the case of successful reception, the neural network matching the scene mode is used; in the case of unsuccessful reception, the indication information of the scene mode is resent, and so on. This application does not limit this.

[0042] In combination with the third aspect, in some possible implementations of the third aspect, the method further includes: the second device communicates with the first device using a neural network matching the scene mode.

[0043] The first device may work with a neural network matching the scene mode based on the successful reception of the above indication information by the second device. The second device may also determine and work with a neural network matching the indicated scene mode based on the successful reception of the above indication information, thereby enabling two mutually coupled neural networks to work in coordination, which is beneficial to improving system performance.

[0044] In combination with the second aspect or the third aspect, in some possible implementations, each of the multiple neural networks is a neural network obtained based on a standardized reference data set, a standardized reference neural network, and a standardized performance judgment criterion.

[0045] That is, the neural network used by the first device and the second device can be a target neural network obtained based on the method described in the first aspect above.

[0046] In combination with the second aspect or the third aspect, in some possible implementations, the first device is a network device, and the second device is a terminal device; or, the first device is a terminal device, and the second device is a network device.

[0047] That is, the determination of the air interface transmission condition and the selection of the scene mode may be performed by the network device or by the terminal device, which is not limited in the present application.

[0048] It should be understood that the methods of the second and third aspects can be used in combination with the method of the first aspect, or can be used alone, which is not limited in the present application.

[0049] It should also be understood that in the above example, the first device and the second device determine the neural network that matches the scene mode through the indication information of the scene mode. However, this should not constitute any limitation to the present application. In another possible implementation method, the determination of the air interface transmission conditions and the selection of the scene mode may not be indicated by signaling interaction, such as the network device and the terminal device may each adapt the neural network based on the preset air interface transmission conditions, the selection of the scene mode and the matching relationship of the neural network. The present application does not limit this either.

[0050] In a fourth aspect, a device for acquiring a neural network is provided, comprising modules or units for executing the method in any possible implementation of the first aspect.

[0051] In a fifth aspect, a device for acquiring a neural network is provided, comprising a processor. The processor is coupled to a memory and can be used to execute instructions or data in the memory to implement the method in any possible implementation of the first aspect. Optionally, the device also includes a memory. Optionally, the device also includes an input / output interface, and the processor is coupled to the input / output interface.

[0052] In a sixth aspect, a communication device is provided, comprising modules or units for executing the method in any possible implementation of the second aspect.

[0053] In a seventh aspect, a communication device is provided, comprising a processor. The processor is coupled to a memory and can be used to execute instructions or data in the memory to implement the method in any possible implementation of the second aspect. Optionally, the communication device also includes a memory. Optionally, the communication device also includes a communication interface, and the processor is coupled to the communication interface.

[0054] In one implementation, the communication device is a terminal device. When the communication device is a terminal device, the communication interface may be a transceiver, or an input / output interface.

[0055] In another implementation, the communication device is a chip configured in a terminal device. When the communication device is a chip configured in a terminal device, the communication interface may be an input / output interface.

[0056] Optionally, the transceiver may be a transceiver circuit. Optionally, the input / output interface may be an input / output circuit.

[0057] In an eighth aspect, a communication device is provided, comprising modules or units for executing the method in any possible implementation of the third aspect.

[0058] In a ninth aspect, a communication device is provided, comprising a processor. The processor is coupled to a memory and can be used to execute instructions or data in the memory to implement the method in any possible implementation of the third aspect. Optionally, the communication device also includes a memory. Optionally, the communication device also includes a communication interface, and the processor is coupled to the communication interface.

[0059] In one implementation, the communication device is a network device. When the communication device is a network device, the communication interface may be a transceiver, or an input / output interface.

[0060] In another implementation, the communication device is a chip configured in a network device. When the communication device is a chip configured in a network device, the communication interface may be an input / output interface.

[0061] Optionally, the transceiver may be a transceiver circuit. Optionally, the input / output interface may be an input / output circuit.

[0062] In a tenth aspect, a processor is provided, comprising: an input circuit, an output circuit, and a processing circuit. The processing circuit is used to receive a signal through the input circuit and transmit a signal through the output circuit, so that the processor executes the method in any possible implementation of the first to third aspects.

[0063] In the specific implementation process, the processor can be one or more chips, the input circuit can be an input pin, the output circuit can be an output pin, and the processing circuit can be a transistor, a gate circuit, a trigger, and various logic circuits. The input signal received by the input circuit can be, for example, but not limited to, received and input by a receiver, and the signal output by the output circuit can be, for example, but not limited to, output to a transmitter and transmitted by the transmitter, and the input circuit and the output circuit can be the same circuit, which is used as an input circuit and an output circuit at different times. The embodiments of the present application do not limit the specific implementation methods of the processor and various circuits.

[0064] In an eleventh aspect, a processing device is provided, comprising a processor and a memory. The processor is used to read instructions stored in the memory, and can receive signals through a receiver and transmit signals through a transmitter to execute the method in any possible implementation of the first to third aspects.

[0065] Optionally, the number of the processors is one or more, and the number of the memories is one or more.

[0066] Optionally, the memory may be integrated with the processor, or the memory may be provided separately from the processor.

[0067] In the specific implementation process, the memory can be a non-transitory memory, such as a read-only memory (ROM), which can be integrated with the processor on the same chip or can be set on different chips respectively. The embodiments of the present application do not limit the type of memory and the setting method of the memory and the processor.

[0068] It should be understood that the relevant data interaction process, such as sending indication information, can be a process of outputting indication information from the processor, and receiving capability information can be a process of receiving input capability information from the processor. Specifically, the data output by the processor can be output to the transmitter, and the input data received by the processor can come from the receiver. Among them, the transmitter and the receiver can be collectively referred to as a transceiver.

[0069] The processing device in the above-mentioned eleventh aspect may be one or more chips. The processor in the processing device may be implemented by hardware or by software. When implemented by hardware, the processor may be a logic circuit, an integrated circuit, etc.; when implemented by software, the processor may be a general-purpose processor implemented by reading software codes stored in a memory, and the memory may be integrated in the processor or located outside the processor and exist independently.

[0070] In the twelfth aspect, a computer program product is provided, which includes: a computer program (also referred to as code, or instruction), which, when executed, enables a computer to execute a method in any possible implementation of the first to third aspects above.

[0071] In the thirteenth aspect, a computer-readable storage medium is provided, which stores a computer program (also referred to as code, or instruction). When the computer-readable storage medium is run on a computer, the computer executes a method in any possible implementation of the first to third aspects above.

[0072] In a fourteenth aspect, a communication system is provided, comprising the aforementioned network device and terminal device, wherein the network device and the terminal device can be respectively configured with the aforementioned target neural network. BRIEF DESCRIPTION OF THE DRAWINGS

[0073] Figure 1 A communication system that can communicate using the neural network provided by the embodiments of the present application;

[0074] Figure 2 It is a schematic diagram of the signal processing process of the transmitter and receiver configured with a neural network;

[0075] Figure 3It is a schematic diagram of a deep neural network (DNN);

[0076] Figure 4 It is a schematic diagram of a convolutional neural network (CNN);

[0077] Figure 5 It is a schematic diagram of a recurrent neural network (RNN);

[0078] Figures 6 to 14 is a schematic flow chart of a method for training a neural network in a communication system provided in an embodiment of the present application;

[0079] Fig.15 and Fig.16 is a schematic flow chart of a communication method provided in an embodiment of the present application;

[0080] Fig.17 and Fig.18 is a schematic block diagram of a device for obtaining a neural network provided in an embodiment of the present application;

[0081] Fig.19 and Fig. 20 is a schematic block diagram of a communication device provided in an embodiment of the present application;

[0082] Fig.21 is a schematic diagram of the structure of a terminal device provided in an embodiment of the present application;

[0083] Fig. 22 It is a structural diagram of the network device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0084] The technical solution in this application will be described below in conjunction with the accompanying drawings.

[0085] The communication method provided in the present application can be applied to various communication systems, for example: Long Term Evolution (LTE) system, LTE frequency division duplex (FDD) system, LTE time division duplex (TDD), universal mobile telecommunication system (UMTS), worldwide interoperability for microwave access (WiMAX) communication system, future fifth generation (5G) mobile communication system or new radio access technology (NR). Among them, the 5G mobile communication system may include non-standalone (NSA) and / or standalone (SA).

[0086] The communication method provided in the present application can also be applied to machine type communication (MTC), long term evolution technology for machine-to-machine communication (LTE-M), device to device (D2D) network, machine to machine (M2M) network, Internet of things (IoT) network or other networks. Among them, the IoT network may include, for example, the Internet of Vehicles. Among them, the communication methods in the Internet of Vehicles system are collectively referred to as vehicle to other devices (vehicle to X, V2X, X can represent anything), for example, the V2X may include: vehicle to vehicle (V2V) communication, vehicle to infrastructure (V2I) communication, vehicle to pedestrian (V2P) communication or vehicle to network (V2N) communication, etc.

[0087] The communication method provided in the present application may also be applied to future communication systems, such as the sixth generation mobile communication system, etc. The present application does not limit this.

[0088] In the embodiment of the present application, the network device may be any device with wireless transceiver function. The network equipment includes, but is not limited to, evolved Node B (eNB), radio network controller (RNC), Node B (NB), base station controller (BSC), base transceiver station (BTS), home base station (e.g., home evolved NodeB, or homeNode B, HNB), baseband unit (BBU), access point (AP) in wireless fidelity (WiFi) system, wireless relay node, wireless backhaul node, transmission point (TP) or transmission and reception point (TRP), etc. It can also be a gNB in ​​a 5G, such as NR, system, or a transmission point (TRP or TP), one or a group of (including multiple antenna panels) antenna panels of a base station in a 5G system, or it can also be a network node constituting a gNB or a transmission point, such as a baseband unit (BBU), or a distributed unit (DU), etc.

[0089] In some deployments, the gNB may include a centralized unit (CU) and a DU. The gNB may also include an active antenna unit (AAU). The CU implements some functions of the gNB, and the DU implements some functions of the gNB, for example, the CU is responsible for processing non-real-time protocols and services, and implementing the functions of the radio resource control (RRC) and packet data convergence protocol (PDCP) layers. The DU is responsible for processing physical layer protocols and real-time services, and implementing the functions of the radio link control (RLC) layer, the medium access control (MAC) layer, and the physical (PHY) layer. The AAU implements some physical layer processing functions, radio frequency processing, and related functions of active antennas. Since the information of the RRC layer will eventually become the information of the PHY layer, or be converted from the information of the PHY layer, therefore, under this architecture, high-level signaling, such as RRC layer signaling, can also be considered to be sent by the DU, or, sent by the DU+AAU. It can be understood that the network device can be a device including one or more of a CU node, a DU node, and an AAU node. In addition, the CU may be classified as a network device in an access network (radio access network, RAN), or the CU may be classified as a network device in a core network (core network, CN), which is not limited in the present application.

[0090] The network equipment provides services for the cell, and the terminal equipment communicates with the cell through the transmission resources (for example, frequency domain resources, or spectrum resources) allocated by the network equipment. The cell can belong to a macro base station (for example, macro eNB or macro gNB, etc.), or it can belong to a base station corresponding to a small cell. The small cell here may include: metro cell, micro cell, pico cell, femto cell, etc. These small cells have the characteristics of small coverage and low transmission power, and are suitable for providing high-speed data transmission services.

[0091] In an embodiment of the present application, the terminal device may also be referred to as user equipment (UE), access terminal, user unit, user station, mobile station, mobile station, remote station, remote terminal, mobile device, user terminal, terminal, wireless communication device, user agent or user device.

[0092] The terminal device can be a device that provides voice / data connectivity to users, for example, a handheld device with wireless connection function, a vehicle-mounted device, etc. At present, some examples of terminals can be: mobile phones, tablet computers (pad), computers with wireless transceiver functions (such as laptops, PDAs, etc.), mobile Internet devices (mobile internet devices, MIDs), virtual reality (virtual reality, VR) devices, augmented reality (augmented reality, AR) devices, wireless terminals in industrial control, wireless terminals in self-driving, wireless terminals in remote medical, wireless terminals in smart grids, wireless terminals in transportation safety, wireless terminals in smart cities, wireless terminals in smart homes, cellular phones, cordless phones, session initiation protocol (session initiation protocol, SIP) phones, wireless local loop (wireless local loop, WLL) stations, personal digital assistants (personal digital assistants) assistant, PDA), handheld devices with wireless communication function, computing devices or other processing devices connected to a wireless modem, vehicle-mounted devices, wearable devices, terminal devices in 5G networks or terminal devices in future evolved public land mobile communication networks (public land mobile network, PLMN), etc.

[0093] Among them, wearable devices can also be called wearable smart devices, which are a general term for the intelligent design and development of wearable devices for daily wear using wearable technology, such as glasses, gloves, watches, clothing and shoes. Wearable devices are portable devices that are worn directly on the body or integrated into the user's clothes or accessories. Wearable devices are not only hardware devices, but also realize powerful functions through software support, data interaction, and cloud interaction. Broadly speaking, wearable smart devices include full-featured, large-sized, and independent of smartphones to achieve complete or partial functions, such as smart watches or smart glasses, as well as those that only focus on a certain type of application function and need to be used in conjunction with other devices such as smartphones, such as various smart bracelets and smart jewelry for vital sign monitoring.

[0094] In addition, the terminal device can also be a terminal device in the Internet of Things (IoT) system. IoT is an important part of the future development of information technology. Its main technical feature is to connect objects to the network through communication technology, thereby realizing an intelligent network of human-machine interconnection and object-to-object interconnection. IoT technology can achieve massive connections, deep coverage, and terminal power saving through narrowband (NB) technology, for example.

[0095] In addition, terminal devices can also include sensors such as smart printers, train detectors, and gas stations. Their main functions include collecting data (part of the terminal equipment), receiving control information and downlink data from network devices, and sending electromagnetic waves to transmit uplink data to network devices.

[0096] To facilitate understanding of the embodiments of the present application, first Figure 1 The communication system applicable to the communication method provided in the embodiment of the present application is described in detail. As shown in the figure, the communication system 100 may include at least one network device, such as Figure 1 The communication system 100 may also include at least one terminal device, such as Figure 1 The terminal devices 102 to 107 shown in the figure. The terminal devices 102 to 107 can be mobile or fixed. The network device 101 and one or more of the terminal devices 102 to 107 can communicate via wireless links. Each network device can provide communication coverage for a specific geographical area and can communicate with terminal devices located in the coverage area. For example, the network device can send configuration information to the terminal device, and the terminal device can send uplink data to the network device based on the configuration information; for another example, the network device can send downlink data to the terminal device. Therefore, Figure 1 The network device 101 and terminal devices 102 to 107 constitute a communication system.

[0097] Optionally, the terminal devices may communicate directly with each other. For example, direct communication between the terminal devices may be achieved using D2D technology. As shown in the figure, the terminal devices 105 and 106, and the terminal devices 105 and 107 may communicate directly using D2D technology. The terminal devices 106 and 107 may communicate with the terminal device 105 individually or simultaneously.

[0098] Terminal devices 105 to 107 may also communicate with network device 101 respectively. For example, they may communicate directly with network device 101, such as terminal devices 105 and 106 in the figure may communicate directly with network device 101; or they may communicate indirectly with network device 101, such as terminal device 107 in the figure communicates with network device 101 via terminal device 105.

[0099] It should be understood that Figure 1 A network device and multiple terminal devices, as well as communication links between the communication devices are exemplarily shown. Optionally, the communication system 100 may include multiple network devices, and each network device may include other number of terminal devices within its coverage, such as more or fewer terminal devices. This application does not limit this.

[0100] The above-mentioned communication equipment, such as Figure 1 The network device 101 and the terminal devices 102 to 107 in the embodiment may be configured with multiple antennas. The multiple antennas may include at least one transmitting antenna for sending signals and at least one receiving antenna for receiving signals. In addition, each communication device also additionally includes a transmitter and a receiver. It can be understood by those skilled in the art that they may include multiple components related to signal transmission and reception (such as processors, modulators, multiplexers, demodulators, demultiplexers or antennas, etc.). Therefore, the network device and the terminal device may communicate via multi-antenna technology.

[0101] Optionally, the wireless communication system 100 may also include other network entities such as a network controller and a mobility management entity, but the embodiments of the present application are not limited thereto.

[0102] It should be understood that Figure 1 In the communication system 100 shown, the network device 101 and any one of the terminal devices 102 to 107 can be applied to the following multiple possible scenarios: modulation and demodulation, encoding and decoding, precoding and detection reception, and extraction and compression of channel information and reconstruction of channel information. It should be understood that the scenarios listed above are only examples and should not constitute any limitation to the present application. The present application includes but is not limited to this.

[0103] Specifically, the transmitter in the network device can be used to modulate the signal, and the receiver in the terminal device can be used to demodulate the modulated signal; the transmitter in the network device can be used to encode the signal, and the receiver in the terminal device can be used to decode the encoded signal; the transmitter in the network device can be used to precode the signal, and the receiver in the terminal device can be used to detect the precoded signal; the transmitter in the terminal device can be used to compress the channel information, and the receiver in the network device can be used to reconstruct the channel based on the compressed channel information.

[0104] In the communication method provided in the embodiment of the present application, both the network device and the terminal device can work using a neural network. The neural network used by the network device and the terminal device can be obtained based on the method for obtaining a neural network provided in the embodiment of the present application. For example, the transmitting neural network obtained by the method for obtaining a neural network provided in the embodiment of the present application is applied to the transmitter of the network device, and the receiving neural network obtained by the method for obtaining a neural network provided in the embodiment of the present application is applied to the receiver of the terminal device; and / or, the transmitting neural network obtained by the method for obtaining a neural network provided in the embodiment of the present application is applied to the receiver of the terminal device, and the receiving neural network obtained by the method for obtaining a neural network provided in the embodiment of the present application is applied to the transmitter of the network device.

[0105] Thus, the transmitter in the network device and the receiver in the terminal device can be jointly optimized based on the neural network, and the transmitter in the terminal device and the receiver in the network device can also be jointly optimized based on the neural network, thereby improving the performance of the communication system.

[0106] Figure 2 The signal processing process of the transmitter and receiver respectively configured with neural networks is shown. Figure 2 The extraction and compression of channel information and the reconstruction of channel information are taken as examples to illustrate the signal processing process of a transmitter using a neural network and a receiver using a neural network.

[0107] like Figure 2 As shown, the transmitting neural network for extracting and compressing channel information may include a pre-processing module, an information extraction module and a post-processing module. However, this should not constitute any limitation to the present application. The transmitting neural network for extracting and compressing channel information may not include a pre-processing module, but include an information extraction module and a post-processing module, or may not include a pre-processing module and a post-processing module, but include an information extraction module. It should be understood that the modules specifically included in the transmitting neural network can be defined based on different designs, and the present application does not limit this.

[0108] The receiving-end neural network for reconstructing channel information may include an inverse pre-processing module, an inverse information extraction module, and an inverse post-processing module. However, this should not constitute any limitation to the present application. The receiving-end neural network for reconstructing channel information may also not include an inverse pre-processing module, but may include an inverse information extraction module and an inverse post-processing module, or may also not include an inverse pre-processing module and an inverse post-processing module, but include an inverse information extraction module. It should be understood that the modules specifically included in the receiving-end neural network may be defined based on different designs, and the present application does not limit this.

[0109] Several possible structures of the transmitting neural network and the receiving neural network are listed below.

[0110] A first possible structure of a transmitting neural network for extracting and compressing channel information includes a pre-processing module, an information extraction module, and a post-processing module. The pre-processing module can be used to pre-process the original data input to the transmitting target neural network based on a standardized data structure. The information extraction module can be used to extract information from the pre-processed data so as to compress the pre-processed data to obtain compressed data. The post-processing module can be used to quantize and bit-map the compressed data based on a standardized data structure to generate air interface information.

[0111] The second possible structure of the transmitting neural network for extracting and compressing channel information is to include an information extraction module and a post-processing module. In other words, the transmitting target neural network may not include a pre-processing module. The pre-processing module can be configured in the transmitter and can be used to pre-process the original data input to the transmitter based on a standardized data structure. The pre-processed data is input into the transmitting neural network. The information extraction module in the transmitting neural network can be used to extract information from the pre-processed data so as to compress the pre-processed data to obtain compressed data. The post-processing module can be used to quantize and bit map the compressed data based on a standardized data structure to generate air interface information.

[0112] The third possible structure of the transmitting neural network for extracting and compressing channel information is to include an information extraction module. In other words, the transmitting neural network may not include a pre-processing module and a post-processing module. The pre-processing module and the post-processing module may be configured in a transmitter. The pre-processing module can be used to pre-process the original data input to the transmitter based on a standardized data structure. The pre-processed data is input into the transmitting neural network. The information extraction module in the transmitting neural network is used to extract information from the pre-processed data so as to compress the pre-processed data to obtain compressed data. The compressed data can be input into the post-processing module, and the post-processing module can be used to quantize and bit map the compressed data based on a standardized data structure to generate air interface information. The first possible structure of the receiving neural network for reconstructing channel information is to include an inverse post-processing module, an inverse information extraction module and an inverse pre-processing module. The inverse post-processing module can be used to parse the air interface information input to the receiving target neural network based on a standardized data structure to obtain compressed data. The inverse information extraction module can be used to process the compressed data to obtain the data before compression. The inverse pre-processing module can be used to process the data before compression to obtain the original data.

[0113] A second possible structure of a receiving neural network for reconstructing channel information is to include an inverse information extraction module and an inverse post-processing module. In other words, the receiving neural network may not include an inverse pre-processing module. The inverse pre-processing module may be configured in a receiver. The inverse post-processing module in the receiving neural network can be used to parse the air interface information input to the receiving target neural network based on a standardized data structure to obtain compressed data. The inverse information extraction module can be used to process the compressed data to obtain pre-compression data. The pre-compression data can be input into the inverse pre-processing module, and the inverse pre-processing module can be used to process the pre-compression data to obtain the original data.

[0114] The third possible structure of the receiving end target neural network for reconstructing channel information is to include an inverse information extraction module. In other words, the receiving end neural network may not include an inverse pre-processing module and an inverse post-processing module. The inverse pre-processing module and the inverse post-processing module may be configured in a receiver. The inverse post-processing module can be used to parse the air interface information input to the receiver based on a standardized data structure to obtain compressed data. The compressed data can be input into the receiving end neural network. The inverse information extraction module in the receiving end neural network can be used to process the compressed data to obtain the data before compression. The compressed data can be obtained by processing the received air interface information based on a standardized data structure. The data before compression can be input into the inverse pre-processing module, and the inverse pre-processing module can be used to process the data before compression to obtain the original data.

[0115] It should be understood that the several possible structures listed above are only examples and should not constitute any limitation to the present application.

[0116] It should be understood that the above-mentioned pre-processing, information processing and post-processing need to be performed based on standardized data structures. Correspondingly, inverse post-processing, inverse information extraction and inverse pre-processing also need to be performed based on standardized data structures consistent with the transmitting end. The purpose is to ensure that the receiving end provides reliable analysis of air interface information.

[0117] Among them, the inverse pre-processing module can be regarded as the inverse processing module of the pre-processing module, the inverse information extraction module can be regarded as the inverse processing module of the information extraction module, and the inverse post-processing module can be regarded as the inverse processing module of the post-processing module. In other words, the behaviors of each module in the transmitter and each module in the receiver are inverse to each other.

[0118] It should also be understood that the modules included in the transmitting neural network and the receiving neural network listed above are named only for ease of understanding and should not constitute any limitation to the present application. For example, the pre-processing module can also be called a pre-processing module, the post-processing module can also be called a quantization module, and so on.

[0119] Figure 2 It is a schematic diagram of the extraction and compression of channel information and the reconstruction of channel information. Figure 2 The channel information shown is channel H. After the transmitter obtains the channel H, it can first input the channel H into the pre-processing module, and the pre-processing module pre-processes the channel H to obtain the channel characteristic information V after pre-processing. For example, singular value decomposition is performed on the channel H to obtain a characteristic vector, or the channel H is transformed into the space-frequency domain to obtain the angle delay spectrum. V can be understood as the channel after pre-processing. The channel characteristic information V after pre-processing is input into the information extraction module, and the information extraction module can extract information from the channel characteristic information V based on a pre-designed algorithm to obtain a compressed channel V'. For example, based on the function f en (V, θ en ) to obtain compressed channel characteristic information V'. The compressed channel characteristic information V' is input to a post-processing module, which can quantize and bit-map the compressed channel characteristic information V' based on a predefined data structure to generate air interface information S.

[0120] After the receiver receives the air interface information S, the air interface information S can be input into the inverse post-processing module in the receiver. The inverse post-processing module can process the air interface information S based on a predefined data structure to restore the compressed channel characteristic information. Compressed channel characteristic information The inverse information extraction module can extract the channel characteristic information based on the algorithm corresponding to the information extraction module in the transmitter. Processing is performed to restore the channel characteristic information before compression For example, based on the function f de (f en (V,Θ en ),Θ de ) Recover channel characteristic information Channel characteristic information It is input to the inverse pre-processing module, which processes the channel characteristic information After processing, the channel can be restored It should be understood that the channel thus recovered is the estimated value of the channel H input to the transmitter as above. As shown in the figure, the processing of the input data by the transmitting neural network and the receiving neural network can be represented by the following function: where Θ = {Θ en ,Θ de}.

[0121] It should be noted that in the above process, the pre-processing module and the inverse pre-processing module need to work based on consistent rules, the information extraction module and the inverse information extraction module need to work based on consistent rules, and the post-processing module and the inverse post-processing module need to work based on consistent rules to ensure that the neural network at the transmitting end and the neural network at the receiving end can work together. The consistency here means that the two can be designed based on consistent rules, but it does not mean that both parties use the same algorithm, the same data structure, etc. to process data. Working based on consistent rules is intended to emphasize that the algorithms, data structures, etc. used by both parties when processing data are corresponding.

[0122] For example, the post-processing module in the transmitter quantizes and bit-maps the compressed channel V' based on a certain data structure to generate air interface information S; the inverse post-processing module in the receiver needs to restore the channel characteristic information to the greatest extent based on the corresponding data structure according to the air interface information S.

[0123] For example, the information extraction module in the transmitter compresses the channel characteristic information V based on a certain algorithm to obtain the compressed channel characteristic information V'; the inverse information extraction module in the receiver needs to process the compressed channel characteristic information V' based on a corresponding algorithm to restore the channel characteristic information before compression to the greatest extent possible.

[0124] It should be understood that Figure 2 The channel H used as an example of channel information is only for ease of understanding and should not constitute any limitation to the present application. The channel H and channel characteristic information V listed above can all be used as examples of channel information. Due to the different designs of the modules included in the neural network, the channel information input to the transmitting neural network may be channel H or channel characteristic information V; the output of the transmitting neural network may be compressed channel characteristic information V' or air interface information S. The input of the receiving neural network may be air interface information S or compressed channel characteristic information restored based on air interface information S. The channel information output from the receiving neural network may be the recovered channel feature information It may also be the recovered channel Moreover, since the structures of the receiving neural network and the transmitting neural network are not necessarily completely symmetrical, the definition of channel information may also be different. For example, the channel information input to the transmitting neural network may be the channel H, while the channel information output by the receiving neural network may be the recovered channel characteristic information

[0125] Based on the difference in channel information, the above Figure 2 Function described Corresponding deformations can be made. For example, it can be expressed as, Y = f(X, Θ) = f de (f en (X, Θ en ),Θ de ). Where X represents the channel information input to the transmitting neural network, and Y represents the channel information output by the receiving neural network. For example, X can be any one of H or V, and Y can be or Any one of .

[0126] It should also be understood that the above Figure 2 The described transmitting neural network is a neural network used for extracting and compressing channel information, and the modules included therein are related to the process of extracting and compressing channel information. Figure 2 The receiving neural network described is a neural network used to reconstruct channel information, and the modules included therein are also related to the reconstruction process of channel information. When used in other signal processing stages, such as encoding and decoding, constellation modulation and demodulation, precoding and detection reception, the modules included in the transmitting neural network and the receiving neural network may be different. For the sake of brevity, examples are not given here one by one.

[0127] In actual commercial use, the transmitter and receiver may be provided by different manufacturers. To meet the above requirements, the transmitter or receiver may need to download the neural network structure and parameter information that matches the current scenario from the other end, or the transmitter and receiver may need to download the neural network structure and parameter information that matches the current scenario from a certain network virtual processing node. For example, download the rules for generating and parsing air interface information; for example, download the parameters and structural description of the neural network that the transmitter uses to extract and compress information from channel H. This may result in a large air interface signaling overhead.

[0128] In view of this, the present application provides a neural network acquisition method in a communication system, in order to reduce the huge air interface overhead caused by the transmitter and receiver downloading the neural network structure and parameter information.

[0129] In order to better understand the embodiments of the present application, the terms involved in this document are briefly explained below.

[0130] 1. Neural network (NN): As an important branch of artificial intelligence, it is a network structure that imitates the behavioral characteristics of animal neural networks to process information. The structure of a neural network is composed of a large number of nodes (or neurons) connected to each other. It processes information by learning and training input information based on a specific operation model. A neural network includes an input layer, a hidden layer, and an output layer. The input layer is responsible for receiving input signals, the output layer is responsible for outputting the calculation results of the neural network, and the hidden layer is responsible for complex functions such as feature expression. The function of the hidden layer is represented by a weight matrix and a corresponding activation function.

[0131] Deep neural networks (DNNs) are generally multi-layer structures. Increasing the depth and width of a neural network can improve its expressive power and provide more powerful information extraction and abstract modeling capabilities for complex systems.

[0132] DNN can be constructed in many ways, including but not limited to recurrent neural network (RNN), convolutional neural network (CNN) and fully connected neural network.

[0133] Figure 3 Schematic diagram of a feedforward neural network (FNN). As shown in the figure, different boxes represent different layers. For example, from left to right, the first layer is the input layer, the second layer is the hidden layer, and the third layer is the output layer. DNN generally also contains multiple hidden layers, whose function is to extract feature information to varying degrees. The output layer can be used to map the required output information from the extracted feature information. Each black circle in the figure can represent a neuron, and neurons in different boxes represent neurons in different layers. The way neurons in each layer are connected and the activation function used will determine the expression function of the neural network.

[0134] 2. Convolutional Neural Network (CNN): A multi-layer neural network, each layer consists of multiple two-dimensional planes, and each plane consists of multiple independent neurons. Multiple neurons in each plane share weights. Weight sharing can reduce the number of parameters in the neural network. At present, in convolutional neural networks, the processor usually performs convolution operations by converting the convolution of input signal features and weights into matrix multiplication operations between signal matrices and weight matrices. In the specific matrix multiplication operation, the signal matrix and weight matrix are processed in blocks to obtain multiple fractal signal matrices and fractal weight matrices, and then matrix multiplication and accumulation operations are performed on the multiple fractal signal matrices and fractal weight matrices.

[0135] Figure 4 It is a schematic diagram of the hidden layer in CNN. The black boxes in the figure are channels of each layer, or feature maps. Different channels can correspond to the same or different convolution kernels. For CNN, not all neurons in the upper and lower layers can be directly connected, but through the "convolution kernel" as an intermediary. In this network, the information to be processed can still retain its original position relationship after the convolution operation. As shown in the figure, the information of the m-1th layer on the left is convolved to obtain the mth layer on the right.

[0136] It should be understood that Figure 4 This is just an example for ease of understanding. For a more detailed description of CNN, please refer to the prior art. For the sake of brevity, it will not be described in detail here.

[0137] 3. Recursive Neural Network (RNN): An artificial neural network with a tree-like hierarchical structure and network nodes that recursively respond to input information in order. In RNN, the output of a neuron can directly act on itself at the next moment, that is, the input of the i-th layer of neurons at time m includes not only the output of the (i-1) layer of neurons at that moment, but also its own output at time (m-1).

[0138] Figure 5 This is a schematic diagram of RNN. Each black circle in the figure represents a neuron. It can be seen that the input of the neurons in hidden layer 2 includes not only the output of the neurons in hidden layer 1 at this moment (as shown by the straight line structure in the figure), but also its own output at the previous moment (as shown by the semicircular arrow in the figure).

[0139] It should be understood that Figure 5 This is just an example for ease of understanding. For a more detailed description of RNN, please refer to the prior art. For the sake of brevity, it will not be described in detail here.

[0140] 4. Performance judgment criteria: In the embodiment of the present application, in order to obtain a target neural network with better performance, the pre-constructed target neural network can be evaluated by using performance judgment criteria.

[0141] In a possible design, the performance judgment criteria may include performance indicators and output conditions. The evaluation of the target neural network may, for example, obtain a performance metric value through a standardized performance metric function, and input the performance metric value into the performance judgment criteria, and compare it with one or more performance indicators to determine whether it meets the output condition.

[0142] As an example and not limitation, the performance metric function may include but is not limited to one or more of the following: mean square error (MSE), normalized mean square error (NMSE), mean absolute error (MAE), maximum absolute error (also known as absolute error bound), correlation coefficient, cross entropy, mutual information, bit error rate, or bit error rate, etc.

[0143] Among them, MSE, NMSE, MAE, maximum absolute error, bit error rate, bit error rate, etc. can be used to characterize the degree of difference, and correlation coefficient, cross entropy, mutual information, etc. can be used to characterize the degree of similarity. However, those skilled in the art can understand that when the difference between two values ​​is large, the similarity between the two values ​​is small; when the difference between two values ​​is small, the similarity between the two values ​​is large. Therefore, the similarity and difference are corresponding.

[0144] The performance indicator may include one or more upper bounds of the performance indicator and / or one or more lower bounds of the performance indicator. The performance indicator may be characterized by, for example, one or more of the multiple items listed in the performance metric function above. Different performance indicators may be defined for different performance metric functions.

[0145] For example, corresponding to the performance measurement function used to characterize the degree of similarity, the higher the degree of similarity, the better the performance. Taking the correlation coefficient as an example, the upper bound of the performance index can be defined as 0.99999 and the lower bound of the performance index can be defined as 0.9.

[0146] For another example, corresponding to the performance measurement function used to characterize the degree of difference, the smaller the degree of difference, the better the performance. Taking MSE as an example, the upper bound of the performance index can be defined as 0.001 and the lower bound of the performance index can be defined as 0.1.

[0147] It should be understood that the performance index is an index set to determine whether the target neural network meets the output conditions. In other words, the target neural network that achieves the performance index can reach an acceptable range of performance after being applied to a transmitter or receiver. For example, taking system throughput as an example, if the similarity coefficient is less than 0.9, the performance can be considered too poor; if the correlation coefficient is greater than or equal to 0.9, the performance can be considered acceptable; if the correlation coefficient is greater than 0.99999, it can be considered that there is no greater gain in performance, or the gain is very small. Therefore, the acceptable range of the correlation coefficient is between [0.9, 0.99999]. Therefore, for the correlation coefficient, reaching the lower limit of the performance index can refer to being greater than or equal to 0.9, and not exceeding the upper limit of the performance index can refer to being less than or equal to 0.99999.

[0148] For another example, taking the bit error rate as an example, if the MSE is greater than 0.1, the performance can be considered too poor; if the MSE is less than or equal to 0.1, the performance can be considered acceptable; if the MSE is less than 0.001, it can be considered that there is no greater performance gain, so the acceptable range of MSE can be between [0.001, 0.1]. Therefore, for MSE, reaching the lower limit of the performance index can mean less than or equal to 0.1, and not exceeding the upper limit of the performance index can mean greater than or equal to 0.001.

[0149] From the above examples, it can be seen that in the embodiments of the present application, if the performance metric function is used to characterize the degree of similarity, the higher the performance metric value output by the performance metric function, the higher the degree of similarity, and the better the performance of the target neural network, so the lower bound of the performance indicator is smaller than the upper bound; if the performance metric function is used to characterize the degree of difference, the lower the performance metric value output by the performance metric function, the smaller the degree of difference, and the better the performance of the target neural network, so the upper bound of the performance indicator is smaller than the lower bound of the performance indicator. In other words, the upper and lower bounds of the performance indicator are defined based on the good and bad performance, so for different performance indicators, the size relationship between the upper or lower bound and the specific value is not fixed.

[0150] In addition, in the embodiments of the present application, when it comes to the relevant description of "reaching the lower limit of the performance indicator", it may refer to the degree of similarity being greater than or equal to a certain threshold (for the convenience of distinction and explanation, for example, recorded as the first threshold); it may also refer to the degree of difference being less than or equal to a certain threshold (for the convenience of distinction and explanation, for example, recorded as the second threshold). When it comes to the relevant description of "not exceeding the upper limit of the performance indicator", it may refer to the degree of similarity being less than or equal to a certain threshold (for the convenience of distinction and explanation, for example, recorded as the third threshold); it may also refer to the degree of difference being greater than or equal to a certain threshold (for the convenience of distinction and explanation, for example, recorded as the fourth threshold). Among them, for the degree of similarity, the corresponding first threshold is less than the third threshold; for the degree of difference, the corresponding second threshold is greater than the fourth threshold.

[0151] When the performance judgment criteria are used to judge whether a target neural network meets the output conditions, the judgment can be made based on one or more performance indicators. The one or more performance indicators may include, for example, one or more performance indicators given for a performance metric function, such as an upper bound and / or a lower bound of the performance indicator; the one or more performance indicators may also include one or more performance indicators given for each of a plurality of different performance metric functions. This application does not limit this.

[0152] In one example, whether the target neural network meets the output condition is determined based on the correlation coefficient. The protocol can define the upper and lower bounds of the correlation coefficient, and can define the output condition as follows: the performance metric values ​​all reach the lower bound of the performance indicator and do not exceed the upper bound of the performance indicator. As in the previous example, the lower bound is 0.9 and the upper bound is 0.99999. The results output by the target neural network and the reference neural network are input into the performance metric function, and one or more performance metric values ​​can be obtained. The one or more performance metric values ​​are compared with the upper and lower bounds to see if they all fall within the range of [0.9, 0.99999]. If so, the output condition is met; if not, the output condition is not met.

[0153] Another example, still based on the correlation coefficient, determines whether the target neural network meets the output condition. The protocol can define the upper and lower bounds of the correlation coefficient, and can define the output condition as: the performance metric values ​​all reach the lower bound of the performance index, and 90% of the performance metric values ​​do not exceed the upper bound of the performance index. As in the previous example, the lower bound is 0.9 and the upper bound is 0.99999. Then one or more performance metric values ​​output by the performance metric function can be compared with the upper and lower bounds to see whether they all reach the lower bound of 0.9, and whether more than 90% of the performance metric values ​​do not exceed the upper bound of 0.99999. If so, the output condition is met; if not, the output condition is not met.

[0154] Another example is to determine whether the target neural network meets the output condition based on the correlation coefficient and the bit error rate. The protocol can define the upper bound of the performance indicator to include the upper and lower bounds of the correlation coefficient and the lower bound of the bit error rate; and the output condition can be defined as: the performance metric value reaches the lower bound of the correlation coefficient and the lower bound of the bit error rate, respectively, and does not exceed the upper bound of the correlation coefficient. As in the previous example, the lower bound of the correlation coefficient is 0.9, the upper bound is 0.99999, and the lower bound of the bit error rate is 0.1. Then one or more performance metric values ​​representing the correlation coefficient output by the performance metric function can be compared with the upper and lower bounds of the correlation coefficient to see whether they all reach the lower bound of 0.9 and do not exceed the upper bound of 0.99999; and whether one or more performance metric values ​​representing the bit error rate output by the performance metric function all reach the lower bound.

[0155] In addition, the performance judgment criteria may also include upper limits and / or lower limits of performance indicators set for different service priorities, as well as output conditions.

[0156] For example, for services such as voice calls and video calls, the corresponding performance indicators may be higher, or the output conditions may be more stringent; for services such as short messages, the corresponding performance indicators may be lower, or the output conditions may be more relaxed.

[0157] For example, the lower bound of the correlation coefficient is 0.9, which has a higher performance index than the lower bound of the correlation coefficient is 0.8. For another example, the lower bound of the MSE is 0.1, which has a higher performance index than the lower bound of the MSE is 0.01. For another example, the output condition that all performance metric values ​​reach the lower bound of the performance indicator is more stringent than that some performance metric values ​​reach the lower bound of the performance indicator.

[0158] When evaluating the target neural network, it can be evaluated using corresponding performance indicators and output conditions according to the priority of the business to which it is applied.

[0159] It should be understood that the above text describes the performance judgment criteria in detail in combination with examples of multiple performance indicators and output conditions, but it should be understood that these examples are only provided for ease of understanding and should not constitute any limitation to the present application.

[0160] 5. Dual network (or dual architecture): used to describe the neural network structure of joint optimization of the transmitting and receiving ends (such as the base station side and the terminal side) in the communication network. The dual network can specifically include a transmitting neural network and a receiving neural network. The dual network includes but is not limited to a transmitting and receiving neural network based on an autoencoder structure or various variations of an autoencoder structure, or a combination of other neural network structures. Under the autoencoder structure, the transmitting neural network is an encoder neural network, and the receiving neural network is a decoder neural network. The two constrain each other and work together.

[0161] It should be noted that the encoder and decoder described here are different from the encoder and decoder used for channel encoding and decoding described above.

[0162] In the embodiment of the present application, the transmitting neural network and the receiving neural network can be a one-to-one, one-to-many, or many-to-one combination to adapt to various system performance and implementation complexity requirements. A pair of mutually coupled neural networks, that is, a transmitting neural network and a receiving neural network that can work together.

[0163] The method for obtaining a neural network provided by an embodiment of the present application will be described in detail below in conjunction with the accompanying drawings. It should be understood that the method may include a process for obtaining a transmitting neural network applied in a transmitter and a process for obtaining a receiving neural network applied in a receiver. The process of obtaining a transmitting neural network and a receiving neural network is described below in conjunction with different embodiments. Among them, for the convenience of distinction and description, the neural network to be obtained is referred to as a target neural network below. The target neural network applied to the transmitter is referred to as a transmitting target neural network, and the target neural network applied to the receiver is referred to as a receiving target neural network.

[0164] It should also be understood that the acquisition of the transmitting end target neural network and the receiving end target neural network can be completed offline, for example. The transmitting end target neural network and the receiving end target neural network can be obtained from the same device, or from different devices, and this application does not limit this. The device used to obtain the target neural network can be, for example, a communication device, or a computing device different from the communication device. This application does not limit this. For the convenience of explanation below, the device used to obtain the target neural network is referred to as an acquisition device.

[0165] Figure 6 600 is a schematic flow chart of a method 600 for obtaining a neural network provided in an embodiment of the present application. Figure 6 As shown, the method 600 includes: step 610 to step 630. Each step in the method 600 is described in detail below.

[0166] In step 610, based on the standardized reference data set, the standardized one or more reference neural networks and the standardized performance judgment criteria, it is determined whether the constructed target neural network meets the preset output conditions.

[0167] The target neural network can be a dedicated neural network for a certain application. The target neural network can be applied to various signal processing stages in the communication system. For example, encoding and decoding, constellation modulation and demodulation, precoding and detection reception, extraction and compression of channel information and reconstruction of channel information. Specifically, the transmitting neural network can be used for encoding, and the receiving neural network can be used for decoding; the transmitting neural network can be used for constellation modulation, and the receiving neural network can be used for demodulation; the transmitting neural network can be used for precoding, and the receiving neural network can be used for detection reception; the transmitting neural network can be used to extract and compress channel information, and the receiving end can be used to reconstruct channel information.

[0168] It should be understood that the target neural network obtained based on the method provided in the embodiments of the present application can also be used for other functions, and the present application does not limit this.

[0169] The target neural network can be built on an existing neural network structure, for example, by expanding the design on basic network structures such as CNN and RNN; or a new network structure can be designed and trained based on a standardized reference data set and / or other data sets related to the application to obtain the target neural network.

[0170] Among them, the relevant description of the neural network structure has been combined above Figures 3 to 5 A detailed description has been given and will not be repeated here for the sake of brevity.

[0171] In order to make the output of the neural network as close as possible to the value that is really wanted to be predicted, the acquisition device can train the target neural network. For example, the weight vector of each layer of the neural network can be updated according to the difference between the predicted value of the current network and the target value that is really wanted by comparing the predicted value of the current network. For example, if the predicted value of the network is high, the weight vector can be adjusted to make it predict lower, and through continuous adjustment, until the neural network can predict the target value that is really wanted. Therefore, it is necessary to predefine "how to compare the difference between the predicted value and the target value", that is, the loss function, cost function or objective function, which are important equations for measuring the difference between the predicted value and the target value. Among them, taking the loss function as an example, the higher the output value (loss) of the loss function, the greater the difference, then the training of the neural network becomes a process of minimizing this output value as much as possible.

[0172] The above training of the target neural network can be implemented based on a standardized reference data set and / or other data sets related to the application.

[0173] Exemplarily, the standardized reference data set may specifically include: a reference data set for training a target neural network, a reference data set for evaluating the performance of a target neural network, etc. Standardization may refer to protocol definition. Each manufacturer may construct a target neural network based on the above method, and evaluate the target neural network based on a standardized reference data set.

[0174] Since the target neural network is a neural network used in a transmitter or a receiver, in order to obtain better performance in a communication system, the target neural network can be evaluated based on a standardized reference data set, a standardized reference neural network and a standardized performance judgment criterion to determine whether it meets the output condition.

[0175] Reference data sets, reference neural networks, and performance judgment criteria can all be standardized, or in other words, predefined by the protocol. Corresponding to different application scenarios, the protocol can predefine multiple sets of different standardized configurations. When evaluating the target neural network based on different application scenarios, a set of standardized configurations that matches the applied scenarios can be used for evaluation. Specifically, in different application scenarios, matching reference data sets, reference neural networks, performance measurement functions, and performance judgment criteria can be used to evaluate the target neural network to obtain a target neural network for application in different scenarios.

[0176] The application scenarios mentioned here may specifically refer to scenarios distinguished based on different air interface transmission conditions. For example, the above application scenarios may include, but are not limited to, different channel scenarios such as outdoor dense urban areas, outdoor ordinary urban areas, outdoor rural areas, outdoor mountainous areas, indoor offices, indoor factories, etc., single interference sources, multiple interference sources, and multiple interference signal types with different strengths, different signal-to-noise ratio conditions such as high signal-to-noise ratio at near and medium points and low signal-to-noise ratio at far points, and different mobile conditions such as walking at low speed, medium speed in urban vehicles, and ultra-high speed on high-speed railways. Because the air interface transmission conditions are different in different application scenarios, and different application scenarios have different requirements for transmission reliability. The standardized configuration matching the application scenario may include one or more of the following: standardized reference data sets, standardized reference neural networks, standardized data structures, standardized performance measurement functions, and standardized performance judgment criteria, etc.

[0177] The reference neural network may include a receiving end reference neural network and / or a transmitting end reference neural network. Furthermore, when the reference neural network includes a receiving end reference neural network and a transmitting end reference neural network, the receiving end reference neural network and the transmitting end reference neural network are one-to-one coupled.

[0178] The data structure can be used to generate and / or parse air interface information. More specifically, the data structure can be used by the transmitting neural network to generate air interface information, and can also be used by the receiving neural network to parse air interface information. The data structure can also be understood as a rule for generating air interface information, or a rule for interpreting air interface information.

[0179] It should be understood that the data structure is a high-level summary that can be used to guide the processing of data at different stages of data processing.

[0180] Take the compression and extraction of the above-mentioned channel information and the reconstruction of the channel information as an example. The transmitting end can pre-process the input raw data based on a certain data structure (for example, recorded as data structure 1), such as rearranging the input raw data; and can extract information from the rearranged data based on another data structure (for example, recorded as data structure 2), such as determining what kind of tensor expression to use; and can also generate air interface information based on another data structure (for example, recorded as data structure 3), such as determining how to perform quantization and bit mapping. The receiving end can parse the air interface information based on the data structure corresponding to data structure 1, and can process the parsed data based on the data structure corresponding to data structure 2, and can also obtain the original data based on the data structure corresponding to data structure 3.

[0181] This application does not limit the specific content of the data structure.

[0182] The standardized achievable capability metric can be understood as the capability parameter required by the neural network to implement a certain application, and can be used to select a matching neural network and / or a matching performance indicator. The matching neural network mentioned here may specifically refer to the neural network being able to be applied in a transmitter or a receiver under given constraints, and the given constraints may include implementation complexity and / or air interface overhead. The matching performance indicator may specifically refer to the performance indicator used to evaluate the neural network at the stage of acquiring the neural network.

[0183] The achievable capability metric value may include, but is not limited to, one or more of the following: the overhead of air interface information, the amount of computation required for the neural network implementation, the amount of parameters stored for the neural network implementation, and the computational accuracy required for the neural network implementation.

[0184] Exemplarily, the overhead of air interface information may refer to bit overhead, for example; the amount of computation required for neural network implementation may refer to floating-point operations per second (Flops), for example; the computational accuracy required for neural network implementation may be, for example, 2-bit fixed point, 4-bit fixed point, 8-bit fixed point, 16-bit fixed point, 16-bit floating point, 32-bit floating point, or 64-bit floating point, etc. This application is not limited to this.

[0185] Each neural network can be characterized by the corresponding parameters listed above. Since the implementation complexity of different applications may be different, the requirements for the neural network capabilities are also different. The acquisition device can select a matching neural network based on the current application's support capability for the implementation complexity.

[0186] It should be noted that, based on the achievable capability metrics of different neural networks, the upper and / or lower limits of the performance that can be achieved by different neural networks may be different. Therefore, the same performance indicator may have one or more performance indicator thresholds (such as upper and / or lower limits), corresponding to one or more different capabilities. When evaluating the neural network in the future, the corresponding performance indicator can be selected for evaluation based on the selected achievable capability metrics.

[0187] It should be understood that the standardized configurations matching different application scenarios may be different from each other or partially the same. For example, the standardized configurations matching different application scenarios may share the same reference data set or the same data structure, etc. This application does not limit this.

[0188] The evaluation process of the target neural network will be described in detail below in combination with the specific implementation method.

[0189] In one implementation, each manufacturer can cascade the constructed target neural network with one or more standardized reference neural networks based on a standardized reference data set, and evaluate the output results after cascading, and determine whether the target neural network meets the output conditions according to standardized performance judgment criteria. If the output conditions are met, step 620 can be executed to output the target neural network; if the output conditions are not met, step 630 can be executed to optimize the target neural network until the obtained target neural network meets the output conditions and is output.

[0190] In another implementation, each manufacturer can input the same or multiple data from a standardized reference data set into the constructed target neural network and the standardized or multiple reference neural networks, compare the output result of the target neural network with the output result of the reference neural network, and determine whether the target neural network meets the output condition according to the standardized performance judgment criteria. If the output condition is met, step 620 can be executed to output the target neural network; if the output condition is not met, step 630 can be executed to optimize the target neural network until the obtained target neural network meets the output condition and is output.

[0191] The following will be combined Figures 7 to 14 The above two implementations are described in more detail with reference to specific examples. In other words, the multiple examples shown below are more detailed descriptions of step 610. For ease of understanding and description, in the following text, unless otherwise specified, the transmitting neural network may include a pre-processing module, an information extraction module, and a post-processing module, and the receiving neural network may include an inverse post-processing module, an inverse information processing module, and an inverse pre-processing module.

[0192] Figure 7 and Figure 8 It is a schematic diagram of a method for obtaining a neural network provided in an embodiment of the present application. Figure 7 and Figure 8 The method for obtaining a neural network shown is an example of cascading a target neural network and a standardized reference neural network, and evaluating the output result of the cascade. Figure 7 The target neural network shown in can be applied to a transmitter, and the target neural network is a transmitting end reference neural network; the reference neural network can be applied to a receiver, and the reference neural network is a receiving end reference neural network. Figure 8 The target neural network shown in can be applied to a receiver, and the target neural network is a receiving end target neural network. The reference neural network can be applied to a transmitter, and the reference neural network is a transmitting end reference neural network.

[0193] like Figure 7 As shown, the acquisition device can obtain one or more transmitting data from the reference data set. For example, one of the transmitting data is recorded as X, and X is input into the transmitting target neural network. After the transmitting target neural network performs pre-processing and information extraction on the transmitting data X, the compressed data X' can be obtained. The transmitting target neural network can further process the compressed data X' according to the standardized data structure to generate air interface information S. Thereafter, the air interface information S can be input into the receiving reference neural network. After the receiving neural network parses the air interface information S according to the standardized data structure, the compressed data can be restored. The receiving end can refer to the neural network and further analyze the compressed data based on the compressed data. The output result Y is obtained.

[0194] As mentioned above, the transmitting target neural network can be applied to different stages of signal processing. Take the extraction and compression of channel information and the reconstruction of channel information as an example. The transmitting data X can be, for example, the channel H. The transmitting neural network can be based on Figure 2 The operation flow of the transmitting neural network listed in the figure performs pre-processing, information extraction and post-processing on the channel H to generate air interface information S.

[0195] The receiving end reference neural network can reconstruct the channel H based on the received air interface information S. The receiving end reference neural network can reconstruct the channel H based on the received air interface information S. Figure 2 The operation process of the receiving end neural network listed in the figure performs inverse post-processing, inverse information extraction and inverse pre-processing on the air interface information S to recover the channel The receiving end refers to the output of the neural network is the estimated value of the channel H It should be understood that the estimated value is also the output result.

[0196] Thereafter, the acquisition device may input the output result Y and the inputted transmission data X into the performance metric function to obtain the performance metric value. It should be understood that the acquisition device may input the output result Y and the transmission data X into one or more performance metric functions to obtain one or more performance metric values.

[0197] It should also be understood that the acquisition device can also process one or more transmission end data obtained from the reference data set according to the above process to obtain multiple pairs of output results and transmission end data. Each pair of output results and transmission end data can be input into one or more performance metric functions to obtain one or more performance metric values. Multiple pairs of output results and transmission end data can obtain multiple performance metric values.

[0198] The acquisition device can further compare the output performance measurement value with the standardized performance index according to the standardized performance judgment criteria, and then determine whether the originating target neural network meets the preset output conditions according to the comparison result.

[0199] In one example, the output condition may be that the performance metric values ​​all reach the lower bound of the performance indicator. The performance metric function may be, for example, MSE, and the lower bound of the performance indicator may be, for example, the lower bound of MSE. One or more pairs of output results and the transmitting end data may be input into the performance metric function to obtain one or more MSE values. For example, the output result The MSE of the two can be obtained by inputting the transmitting data H into the performance measurement function, that is, The acquisition and merging can further determine whether one or more performance metric values ​​obtained reach the lower bound of the standardized MSE. If all of them are reached, the output condition is met; if some or all of them are not reached, the output condition is not met.

[0200] It should be understood that if the receiving end reference neural network is a matching neural network selected based on an achievable capability metric, the above performance indicators can be evaluated based on the performance indicators corresponding to the achievable capability metric.

[0201] It should also be understood that, as mentioned above, based on the different designs of the modules included in the neural network, the input and output of the transmitting neural network and the input and output of the receiving neural network are not necessarily the same as the above examples. For example, the input of the transmitting neural network may be the channel H, while the output of the receiving neural network may be a feature vector (it should be understood that the feature vector can be understood as an example of the channel feature information V described above). Here, the feature vector output by the receiving end is the recovered feature vector, for example, denoted as However, this does not affect the use of the performance judgment criteria. Due to the relationship between the eigenvector and the channel, U = eig (H) (eig () represents the function used to solve the eigenvector), the eigenvector Further recovery of the channel Alternatively, the channel H can be used to further determine the feature vector U, and then determine the MSE of the two. That is, It may also appear as Or, and The following text involves the description of performance measurement functions in many places. For the sake of brevity, the description of the same or similar situations is omitted in the following text.

[0202] In another example, the output condition may be that 90% of the performance metric values ​​reach the lower limit of the performance indicator. The performance metric function may be, for example, a calculation function of a correlation coefficient, and the lower limit of the performance indicator may be, for example, the lower limit of the correlation coefficient. One or more correlation coefficients may be obtained by inputting one or more pairs of output results and the originating data into the performance metric function. For example, the output result The correlation coefficient with the originating data H can be expressed as or Corr() represents a function for finding the correlation coefficient. The acquisition device can further determine whether the obtained one or more performance metric values ​​reach the lower bound of the standardized correlation coefficient. If more than 90% of the performance metric values ​​reach the lower bound of the correlation coefficient, the output condition is met; if the number of performance metric values ​​that reach the lower bound of the correlation coefficient is less than 90%, the output condition is not met.

[0203] It should be understood that MSE and correlation coefficient are only examples, and the present application does not limit the performance indicators and related performance measurement functions used by the acquisition device to determine whether the target neural network meets the output conditions. It should also be understood that reaching the lower bound of the performance indicator can be used as an example of the output condition of the target neural network, but it should not constitute any limitation to the present application.

[0204] Since the performance judgment criteria have been described in detail above with reference to multiple examples, for the sake of brevity, more output conditions will not be listed here, and different output conditions will be described in detail with examples one by one.

[0205] like Figure 8 The acquisition device shown can obtain one or more transmitting data from the reference data set. For example, one of the transmitting data is recorded as X, and X is input into the transmitting reference neural network. After the transmitting reference neural network performs pre-processing and information extraction on the transmitting data X, the compressed data X' can be obtained. The transmitting reference neural network can further process the compressed data X' according to the standardized data structure to generate air interface information S. Thereafter, the air interface information S can be input into the receiving target neural network. The receiving target neural network can parse the air interface information S based on the standardized data structure to recover the compressed data. The receiving end target neural network can further be based on the compressed data The output result Y is obtained.

[0206] Still taking the extraction and compression of channel information and the reconstruction of channel information as an example, the transmitting data X can be the channel H. The transmitting reference neural network can be based on Figure 2 The operation flow of the transmitting neural network listed in the figure performs pre-processing, information extraction and post-processing on the channel H to generate air interface information S.

[0207] The receiving end target neural network can reconstruct the channel H based on the received air interface information S. The receiving end target neural network can reconstruct the channel H based on the received air interface information S. Figure 2 The operation process of the receiving end neural network listed in the above is to perform inverse post-processing, inverse information extraction and inverse pre-processing on the air interface information line to recover the channel The output of the receiving target neural network is the estimated value of the channel H It should be understood that the estimated value is also the output value.

[0208] Thereafter, the acquisition device may input the output result Y and the inputted transmission data X into the performance metric function to obtain the performance metric value. It should be understood that the acquisition device may input the output result Y and the transmission data X into one or more performance metric functions to obtain one or more performance metric values.

[0209] It should also be understood that the acquisition device can also process one or more transmission end data obtained from the reference data set according to the above process to obtain multiple pairs of output results and transmission end data. Each pair of output results and transmission end data can be input into one or more performance metric functions to obtain one or more performance metric values. Multiple pairs of output results and transmission end data can obtain multiple performance metric values.

[0210] The acquisition device can further compare the output performance measurement value with the standardized performance index according to the standardized performance judgment criteria, and then determine whether the originating target neural network meets the preset output conditions according to the comparison result.

[0211] Since the performance judgment criteria have been described in detail above with reference to multiple examples, for the sake of brevity, more output conditions will not be listed here, and different output conditions will be described in detail with examples one by one.

[0212] Fig. 9 and Fig.10 It is a schematic diagram of a method for acquiring a neural network provided in an embodiment of the present application. Fig. 9 and Fig.10 The neural network training method shown uses a standardized reference neural network as a reference to evaluate the trained target neural network. Fig. 9 The target neural network and reference neural network shown in the figure can be applied to a transmitter, where the target neural network is a transmitting target neural network and the reference neural network is a transmitting reference neural network. Fig.10 The target neural network and reference neural network shown can be applied to a receiver, wherein the target neural network is a receiving end target neural network, and the reference neural network is a receiving end reference neural network.

[0213] like Fig. 9 As shown, the acquisition device can acquire one or more transmission data from the reference data set. Each transmission data can be input into the transmission target neural network and the transmission reference neural network to obtain the target air interface information output from the transmission target neural network and the reference air interface information generated from the transmission reference neural network.

[0214] Take one of the transmitting data X as an example. The transmitting data X is input into the transmitting target neural network and the transmitting reference neural network. The transmitting reference neural network can perform pre-processing and information extraction on the transmitting data X based on a pre-designed algorithm to obtain compressed data X0'. The transmitting reference neural network can process the compressed data X0' based on a standardized data structure to obtain reference air interface information S0 generated by the transmitting reference neural network. The transmitting target neural network can perform pre-processing and information extraction on the transmitting data X based on an algorithm corresponding to the receiving neural network to obtain compressed data X1'. The transmitting target neural network can process the compressed data X1' based on a standardized data structure to obtain target air interface information S1 generated by the transmitting target neural network.

[0215] The pre-processing module for pre-processing the transmission data X may be a module shared by the transmission reference neural network and the transmission target neural network, that is, the process of pre-processing the transmission data X is performed by the same pre-processing module. Alternatively, the pre-processing module for pre-processing the transmission data X may be a module respectively included in the transmission reference neural network and the transmission target neural network, that is, the process of pre-processing the transmission data X is respectively performed by the transmission reference neural network and the transmission target neural network.

[0216] Of course, the pre-processing module may not be included in the transmitting reference neural network and the transmitting target neural network, and may exist alone in the transmitter as a pre-processing module. The pre-processing module may input the pre-processed data into the transmitting reference neural network and the transmitting target neural network respectively. In this case, the transmitting data X is input into the transmitting target neural network and the transmitting reference neural network after being processed by the pre-processing module.

[0217] Still taking the extraction and compression of channel information and the reconstruction of channel information as an example, the transmitting data X may be channel H. In one implementation, channel H may be input into the transmitting reference neural network and the transmitting target neural network respectively, and the transmitting reference neural network and the transmitting target neural network may be respectively input into the transmitting reference neural network and the transmitting target neural network as follows: Figure 2 According to the operation flow of the transmitting neural network listed in , the channel H is pre-processed, information extracted and post-processed in sequence to obtain the reference air interface information S0 and the target air interface information S1 respectively.

[0218] Thereafter, the acquisition device may input the reference air interface information S0 and the target air interface information S1 into a performance metric function to obtain a performance metric value. It should be understood that the acquisition device may input the reference air interface information S0 and the target air interface information S1 into one or more performance metric functions to obtain one or more performance metric values.

[0219] It should also be understood that the acquisition device can also process one or more transmitting end data obtained from the reference data set according to the above process to obtain multiple pairs of reference air interface information and target air interface information. A pair of reference air interface information and target air interface information obtained from the same transmitting end data can be input into one or more performance metric functions to obtain one or more performance metric values. Multiple pairs of reference air interface information and target air interface information can obtain multiple performance metric values.

[0220] The acquisition device can further compare the output performance measurement value with the standardized performance index according to the standardized performance judgment criteria, and then determine whether the originating target neural network meets the preset output conditions according to the comparison result.

[0221] Since the performance judgment criteria have been described in detail above with reference to multiple examples, for the sake of brevity, more output conditions will not be listed here, and different output conditions will be described in detail with examples one by one.

[0222] like Fig.10 As shown, the acquisition device can input the same one or more air interface information into the receiving end reference neural network and the receiving end target neural network. The one or more air interface information can be directly obtained from the reference data set, or it can be the air interface information obtained from a transmitter after the transmitting end data obtained from the reference data set is input into a transmitter. This application does not limit this. As long as the air interface information input into the receiving end reference neural network and the receiving end target neural network each time is the same air interface information.

[0223] In one possible design, the air interface information input to the receiving end reference neural network and the receiving end target neural network can be obtained in the following manner: the transmitting end data obtained from the reference data set is input into a transmitting end reference neural network, and the transmitting end reference neural network can process the input transmitting end data according to a predefined data structure to obtain the air interface information to be sent to the receiving end target neural network and the receiving end reference neural network.

[0224] The air interface information output by the transmitting end reference neural network can be input to the receiving end reference neural network coupled thereto. In other words, in this embodiment, the receiving end target neural network can be evaluated by a pair of mutually coupled transmitting end reference neural networks and receiving end reference neural networks.

[0225] Take air interface information S as an example. The air interface information S can be input into the receiving end target neural network and the receiving end reference neural network. The receiving end reference neural network can parse the air interface information S based on a standardized data structure, and can further process the parsing result based on a pre-designed algorithm to obtain a reference output result Y0 output by the receiving end reference neural network. The receiving end target neural network can parse the air interface information S based on a standardized data structure, and can further process the parsing result based on the corresponding algorithm of the receiving end reference neural network to obtain a target output result Y1 output by the receiving end target neural network. Receiving end reference neural network and receiving end target neural network

[0226] Still taking the extraction and compression of channel information and the reconstruction of channel information as an example, the air interface information input to the receiving end target neural network and the receiving end reference neural network is S. After receiving the air interface information S, the receiving end reference neural network can Figure 2 The operation process of the receiving end neural network listed in the figure performs inverse post-processing, inverse information extraction, and inverse pre-processing on the air interface information S to obtain the estimated value of the channel H After receiving the air interface information S, the target neural network at the receiving end can Figure 2 The operation process of the receiving end neural network listed in the figure performs inverse post-processing, inverse information extraction, and inverse pre-processing on the air interface information S to obtain the estimated value of the channel H It should be understood that the estimated value of the reference neural network output That is, the reference output result. The estimated value of the target neural network output That is the target output result.

[0227] After that, the acquisition device can output the reference result Output results with target Input into the performance metric function to obtain the performance metric value. It should be understood that the acquisition device can output the result of the reference Output results with target Input into one or more performance measurement functions to obtain one or more performance measurement values.

[0228] It should also be understood that the acquisition device can process one or more air interface information according to the above process to obtain multiple target output results output from the receiving end target neural network and multiple reference output results output from the reference neural network. A pair of target output results and reference output results obtained from the same air interface information can be input into one or more performance measurement functions to obtain one or more performance measurement values. Multiple pairs of target output results and reference output results can obtain multiple performance measurement values.

[0229] The acquisition device can further compare the output performance measurement value with the standardized performance index according to the standardized performance judgment criteria, and then determine whether the originating target neural network meets the preset output conditions according to the comparison result.

[0230] Since the performance judgment criteria have been described in detail above with reference to multiple examples, for the sake of brevity, more output conditions will not be listed here, and different output conditions will be described in detail with examples one by one.

[0231] Combination of the above Figures 7 to 10 The specific process of evaluating the target neural network by combining the acquisition device with a standardized transmitting end reference neural network and / or receiving end reference neural network is mainly described. However, it should be understood that the examples listed above should not constitute any limitation to the present application.

[0232] For example, the acquisition device can also use two different reference neural networks to evaluate a target neural network. The two reference neural networks can generate two different output results based on the same input. The two output results can be the output results of two reference neural networks designed from the perspective of obtaining different lower limits of performance, or the two output results can be the output results of two reference neural networks designed from the perspective of obtaining the upper and lower limits of performance. Therefore, the two output results can be used to determine whether the target neural network meets the output condition.

[0233] More specifically, if the two output results are the output results of two different reference neural networks designed based on obtaining different performance lower limits, the two output results can be respectively input into different performance metric functions with the source data to obtain different performance metric values, which can be compared with the lower limit of the corresponding performance index. When both output results exceed the lower limit of the performance index, it is determined that the target neural network meets the output condition. Thereby, it can be ensured that the performance of the target neural network is not lower than the lower limit when in use.

[0234] If the two output results are the output results of two reference neural networks designed based on the upper and lower limits of the performance, then of the two output results, the output result of the reference neural network designed based on the upper limit of the performance can be input into the performance measurement function together with the transmitting data to obtain a performance measurement value, which can be used to compare with the upper limit of the performance indicator, thereby avoiding the performance of the target neural network exceeding the upper limit when in use; the output result of the reference neural network designed based on the lower limit of the performance can be input into the performance measurement function together with the transmitting data to obtain another performance measurement value, which can be used to compare with the lower limit of the performance indicator, thereby ensuring that the performance of the target neural network is not lower than the lower limit when in use.

[0235] Below Figures 11 to 14 The specific process of the acquisition device evaluating the target neural network by combining multiple standardized transmitting-end reference neural networks and / or receiving-end reference neural networks is shown.

[0236] Fig.11 The specific process of the acquisition device combining two standardized reference neural networks to determine whether the target neural network meets the output conditions is shown. Fig.11 In the embodiment, the target neural network can be cascaded with the reference neural network. The target neural network can be used in a transmitter as a transmitting target neural network; the reference neural network can be used in a receiver as a receiving reference neural network.

[0237] Fig.11 One of the two receiving end reference neural networks shown (e.g., receiving end reference neural network 1 in the figure) can be a neural network designed to obtain a lower limit of a certain performance (e.g., MSE), and the other (e.g., receiving end reference neural network 2 in the figure) can be a neural network designed to obtain a lower limit of another performance (e.g., correlation coefficient). The air interface information S output by the transmitting end target neural network is input to the receiving end reference neural network 1 and the receiving end reference neural network 2, respectively. The receiving end reference neural network 1 and the receiving end reference neural network 2 can be respectively based on Figure 2 The operation flow of the receiving end neural network listed in the above is used to obtain the output result. For example, the receiving end reference neural network 1 can obtain the output result Y 01 , the output result Y can be obtained by referring to neural network 2 at the receiving end 02 Based on the method described above, the acquisition device can compare the transmitting data X input to the transmitting target neural network with the output result Y generated by the receiving reference neural network. 01 and Y 02 The input is input into the performance measurement function to obtain the performance measurement value, and further based on the performance judgment criteria and the performance measurement value, it is determined whether the originating target neural network meets the output condition.

[0238] Fig.11 The performance metric functions shown include MSE and correlation coefficient. Fig.11 The source data X and the output result Y 01 The performance metric MSE(X, Y 01 ); from the source data X and the output result Y 02 The performance metric correlation coefficient (X, Y 02 ). The acquisition device further combines the lower bound of the standardized MSE and the lower bound of the correlation coefficient to determine whether the originating target neural network meets the output condition.

[0239] Optionally, the output condition is: the output result Y generated by the receiving end reference neural network 101 The performance metric value of the data at the sending end reaches the lower bound of MSE, and the output result Y generated by the receiving end reference neural network 2 02 The performance metric value of the transmitting end data reaches the lower bound of the correlation coefficient. Since the two receiving end reference neural networks that output the above two output results are obtained based on different performance lower limits, the obtained performance metric values ​​can be compared with the lower bounds of different performance indicators.

[0240] In this embodiment, if MSE(X, Y 01 ) reaches the lower bound of the performance index, the correlation coefficient (X, Y 02 ) reaches the lower bound of the correlation coefficient, then the originating target neural network meets the output condition.

[0241] If one of the above conditions is not met, the target neural network does not meet the output condition. For example, for output condition 1, if MSE(X, Y 01 ) does not reach the lower bound of MSE, or the correlation coefficient (X, Y 02 ) does not reach the lower bound of the correlation coefficient, it can be determined that the originating target neural network does not meet the output condition.

[0242] It should be understood that the lower bounds of the MSE and the correlation coefficient above can be lower bounds that match the capabilities satisfied by the target neural network. For example, the acquisition device can select the corresponding performance indicator based on the achievable capability metric satisfied by the selected target neural network to evaluate the target neural network.

[0243] It should also be understood that the above two output results Y 01 and Y 02 It can be obtained by processing the same air interface information by two different reference neural networks, for example, one is a high-complexity deep convolutional network and the other is a low-complexity light convolutional network, which is not limited in this application.

[0244] It should be understood that the number of receiving-end reference neural networks used for cascading with the transmitting-end target neural network may be two or more, and this application does not limit this.

[0245] For example, more receiving-end reference neural networks are cascaded with the transmitting-end target neural network. A portion of these receiving-end reference neural networks (e.g., recorded as receiving-end reference neural network set 1) can be a neural network designed from the perspective of obtaining the lower limit of performance, and another portion (e.g., recorded as receiving-end reference neural network set 2) can be a neural network designed from the perspective of obtaining the upper limit of performance. In this way, the output results of more receiving-end neural networks can be obtained. These output results can also be respectively input into the performance measurement function with the transmitting-end data input into the transmitting-end target neural network to obtain multiple performance measurement values.

[0246] In this case, the output condition may be, for example: the performance measurement values ​​of the result output by the receiving end reference neural network set 1 and the transmitting end data both reach the lower limit of the performance index, and the performance measurement values ​​of the result output by the receiving end reference neural network set 2 and the transmitting end data both do not exceed the upper limit of the performance index. Alternatively, the output condition may also be: the performance measurement values ​​of the result output by the receiving end reference neural network set 1 and the transmitting end data both reach the lower limit of the performance index, and more than 90% of the performance measurement values ​​of the result output by the receiving end reference neural network set 2 and the transmitting end data do not exceed the upper limit of the performance index, and so on. For the sake of brevity, they are not listed here. It should be understood that the present application does not limit the specific content of the output conditions.

[0247] Fig.12 The specific process of the acquisition device combining two standardized reference neural networks to determine whether the target neural network meets the output conditions is shown. Fig.12 The target neural network in the method can be cascaded with the reference neural network. The target neural network is applied to the receiver and is the receiving end target neural network; the reference neural network is applied to the transmitter and is the transmitting end reference neural network. Fig.12 One of the two reference neural networks shown (e.g., reference neural network 1 in the figure) may be a neural network set up to obtain a lower limit of a certain performance (e.g., MSE), and the other (e.g., reference neural network 2 in the figure) may be a neural network set up to obtain a lower limit of another performance (e.g., correlation coefficient). The air interface information S can be obtained by the transmitting end reference neural network 1. 01 , the air interface information S can be obtained by referring to the neural network 2 at the transmitting end 02 The receiving end target neural network can be based on Figure 2 The operation flow of the receiving end neural network shown in FIG. 1 is based on the air interface information S 01 and S 02 Based on the method described above, the acquisition device can compare the transmitting data X input to the transmitting reference neural network with the receiving target neural network based on the air interface information S 01 and S 02 The output results Y1 and Y2 respectively obtained are input into the performance measurement function to obtain the performance measurement value, and further according to the performance judgment criteria and the performance measurement value, it is determined whether the target neural network meets the output condition.

[0248] Fig.12The performance measurement function shown includes MSE and correlation coefficient. The performance measurement value correlation coefficient (X, Y2) can be obtained from the transmitting data X and the output result Y2. The acquisition device further combines the lower bound of the standardized MSE and the lower bound of the correlation coefficient to determine whether the receiving target neural network meets the output condition. Since the above has already described the judgment process in detail in combination with the performance judgment criteria, it will not be repeated here for the sake of brevity.

[0249] Judging whether the target neural network meets the output condition by the upper and lower bounds of the performance index is not limited to the above-listed method of cascading the target neural network with a reference neural network. The training device can also input the same or multiple source data into the source target neural network and the source reference neural network, and compare the output result of the source target neural network with the output result of the source reference neural network to determine whether the source target neural network meets the output condition.

[0250] Fig.13 The specific process of combining the acquisition device with two standardized reference neural networks to judge whether the target neural network meets the output condition is shown. Both the target neural network and the reference neural network can be applied to the transmitter, namely the transmitting end target neural network and the transmitting end reference neural network.

[0251] Fig.13 One of the two transmitting reference neural networks shown (e.g., transmitting reference neural network 1 in the figure) can be a neural network set up to obtain the lower limit of a certain performance (e.g., MSE), and the other (e.g., transmitting reference neural network 2 in the figure) can be a neural network set up to obtain the lower limit of another performance (e.g., correlation coefficient). The same transmitting data X is input to the transmitting reference neural network 1, the transmitting reference neural network 2, and the transmitting target neural network, respectively. The transmitting reference neural network 1, the transmitting reference neural network 2, and the transmitting target neural network can process the transmitting data X, respectively, for example, according to Figure 2 The operation flow of the transmitting end neural network listed in the above process processes the transmitting end data X to generate different air interface information. The transmitting end reference neural network 1 can generate reference air interface information S 01 , the reference neural network 2 at the transmitting end can generate reference air interface information S 02 , the target air interface information S1 can be generated by the transmitting target neural network. Based on the method described above, the acquisition device obtains the target air interface information S1 and the reference air interface information S 01 and S 02 Input into the performance measurement function to obtain the performance measurement value, and further determine whether the target neural network meets the output condition according to the performance judgment criterion and the performance measurement value. Since the above has already described the judgment process in detail in combination with the performance judgment criterion, it will not be repeated here for the sake of brevity.

[0252] Fig.14 The specific process of the acquisition device combining two standardized reference neural networks to determine whether the target neural network meets the output conditions is shown. Fig.14 The reference neural network includes a transmitting end reference neural network applied in a transmitter and a receiving end reference neural network applied in a receiver; the target neural network is applied in a receiver and is a receiving end target neural network.

[0253] The acquisition device can input the transmitting end data obtained from the reference data set into the transmitting end reference neural network. The transmitting end reference neural network can generate air interface information S based on a standardized data interface. The air interface information S can be input into the receiving end target neural network and two receiving end reference neural networks. Among them, one of the two receiving end reference neural networks (for example, the receiving end reference neural network 1 in the figure) can be a neural network designed from the perspective of obtaining the lower limit of a certain performance (such as MSE), and the other (for example, the receiving end reference neural network 2 in the figure) can be a neural network designed from the perspective of obtaining the lower limit of another performance (such as the correlation coefficient). The receiving end reference neural network 1, the receiving end reference neural network 2 and the transmitting end target neural network can respectively parse the air interface information S based on the standardized data structure to obtain different output results. The reference output result Y can be obtained from the receiving end reference neural network 1. 01 , the reference output result Y can be obtained by referring to the neural network 2 at the receiving end 02 , the target output result Y1 can be obtained by the target neural network at the receiving end. Based on the method described above, the acquisition device can obtain the target output result Y1 and the reference output result Y 01 and Y 02 Input into the performance measurement function to obtain the performance measurement value, and further determine whether the receiving end target neural network meets the output condition according to the performance judgment criterion and the performance measurement value. Since the above has already described the judgment process in detail in combination with the performance judgment criterion, it will not be repeated here for the sake of brevity.

[0254] It should be understood that the above is only for ease of understanding. Figures 7 to 14 The specific implementation process of the above step 610 is described in detail. However, these examples are only shown for ease of understanding and should not constitute any limitation to the present application. Based on the same concept, those skilled in the art can perform equivalent replacement or simple deformation on part or all of the above steps to achieve the same effect. Therefore, these equivalent replacements or simple deformations should all fall within the protection scope of the present application.

[0255] It should also be understood that Figures 7 to 14 Mainly Figure 2The specific process of step 610 is described by taking the modules included in the transmitting neural network and the receiving neural network as examples. However, this should not constitute any limitation to the present application. In the case where the modules included in the transmitting neural network and the receiving neural network are designed differently, the input and output of the transmitting neural network may be different from those shown in the figure, and the input and output of the receiving neural network may also be different from those shown in the figure.

[0256] Thereafter, the acquisition device may execute one of step 620 or step 630 based on whether the target neural network meets the output condition. If the target neural network meets the output condition, step 620 may be executed to output the target neural network; if the target neural network does not meet the output condition, step 630 may be executed to continue optimizing the target neural network.

[0257] The process of obtaining the device to optimize the target neural network may specifically include, but is not limited to, adjusting the neural network structure design, adjusting the reference data set, adjusting the neural network training method, adjusting the definition of the cost function, loss function or objective function, changing the initialization method of the neural network, changing the constraints of the neural network parameters, and changing the definition of the activation function of the neural network, etc.

[0258] By executing one or more of the above-listed steps, the acquisition device can obtain an optimized target neural network. The acquisition device can further evaluate the optimized target neural network to determine whether it meets the output condition. The specific process of the acquisition device evaluating the optimized target neural network can refer to the relevant description in step 610 above, and will not be repeated here for the sake of brevity.

[0259] Based on the above technical solution, by introducing standardized reference data sets, standardized reference neural networks and standardized performance judgment criteria, the acquisition device can evaluate the target neural network based on the existing standardized reference neural network. If the output conditions are not met, the target neural network can be optimized until the target neural network meets the output conditions. This avoids the overhead of downloading the neural network structure and parameter information through the air interface. At the same time, it can also ensure that products from different manufacturers can be interconnected. While ensuring performance, it can also reflect the differentiated design and competitiveness of products from different manufacturers. Therefore, overall, the feasibility of dual network implementation is greatly improved.

[0260] It should be understood that the above is only for the sake of understanding, and the above method is further described in detail by taking the extraction and compression of channel information and the reconstruction of channel information as examples. However, this should not constitute any limitation to the present application. In other processing stages of the information listed above, the method for obtaining a neural network provided in the embodiment of the present application is still applicable. For the sake of brevity, examples are not given here one by one.

[0261] The target neural network obtained based on the method described above can be applied to a communication network. For example, the transmitting end target neural network can be applied to a network device, and the receiving end target neural network can be applied to a terminal device. The network device may include a transmitter as described above, and the terminal device may include a receiver as described above. For another example, the transmitting end target neural network can be applied to a terminal device, and the receiving end target neural network can be applied to a network device. The terminal device may include a transmitter as described above, and the network device may include a receiver as described above. This application does not limit this.

[0262] Since the air interface transmission conditions between the network device and the terminal device vary depending on the environment, when selecting the target neural network to be applied to the network device and the terminal device, a matching scenario mode can be selected according to the current air interface transmission conditions. The air interface transmission conditions may include, but are not limited to, channel status, interference status, etc.

[0263] One of the network device and the terminal device can determine a scenario mode based on the air interface transmission condition so as to determine a target neural network suitable for the current air interface transmission condition.

[0264] Fig.15 It is a schematic flow chart of the communication method provided in an embodiment of the present application. Fig.15 The method 700 shown may include steps 710 to 750. Each step in the method 700 is described in detail below.

[0265] In step 710, the terminal device sends the capability information of the terminal device. Correspondingly, the network device receives the capability information of the terminal device.

[0266] Specifically, the capability information may be used to indicate an achievable capability metric value. The achievable capability metric value may specifically refer to an achievable capability metric value of a neural network that can be adopted by the terminal device. The network device may select a matching neural network based on the achievable capability metric value reported by the terminal device and a scenario mode determined by the air interface transmission conditions between the terminal device and the network device.

[0267] The specific content of the achievable capability metric value has been described in detail in the above method 600, and will not be repeated here for the sake of brevity.

[0268] In step 720, the network device determines a scenario mode based on air interface transmission conditions between the network device and the terminal device.

[0269] The network device may, for example, continuously or periodically obtain the air interface transmission conditions between the network device and the terminal device, for example, by measuring the channel, measuring the interference, etc. to obtain the air interface transmission information. The specific duration of the period may be a predefined value. The present application does not limit the actual behavior of the network device in obtaining the air interface transmission conditions.

[0270] The network device can determine the current scene mode based on the air interface transmission conditions between the terminal device, such as channel status, interference status, etc. The scene mode may include, for example, but is not limited to: different channel scenes such as outdoor dense urban areas, outdoor ordinary urban areas, outdoor rural areas, outdoor mountainous areas, indoor offices, indoor factories, etc., single interference sources, multiple interference sources, multiple interference signal types with different strengths, different signal-to-noise ratio conditions such as high signal-to-noise ratio at near points and low signal-to-noise ratio at far points, walking at low speed, medium speed in urban vehicles, high-speed rail and other different mobile conditions, etc. Since different scenes may have different effects on air interface transmission conditions, different scene modes can be matched with different neural networks to apply the neural network matching the current air interface transmission conditions to the network device and the terminal device. In step 730, the network device sends the indication information of the scene mode to the terminal device. Accordingly, the terminal device receives the indication information of the scene mode from the network device.

[0271] After determining the scene mode, the network device can indicate the indication information of the scene mode to the terminal device through signaling, so that the terminal device can select a neural network that is one-to-one coupled with the network device to work in collaboration.

[0272] There are many specific ways for the network device to indicate the scene mode. For example, the corresponding relationship between multiple indication bits and multiple scene modes can be predefined, and different scene modes can be indicated by different indication bits.

[0273] Optionally, the indication information of the scenario mode is carried in a high-layer signaling, such as a radio resource control (RRC) message or a medium access control (MAC) control element (CE).

[0274] That is, the scene mode may be statically configured or semi-statically configured.

[0275] Optionally, the indication information of the scenario mode is carried in physical layer signaling. The physical layer signaling is, for example, downlink control information (DCI).

[0276] That is, the scene mode may be dynamically configured.

[0277] In step 740, the network device receives confirmation information from the terminal device, where the confirmation information is used to indicate successful reception of the indication information of the scene mode.

[0278] Based on the successful reception of the above indication information, the terminal device can send confirmation information to the network device.

[0279] For example, the terminal device may send the confirmation information in the case of successful reception, and not send the confirmation information in the case of unsuccessful reception. The network device may start a timer after sending the indication information of the above-mentioned scene mode, and the duration of the timer may be a predefined value. If the confirmation information is received before the timer times out, it can be determined that the terminal device has successfully received the indication information of the scene mode; otherwise, it is considered that the terminal device has not successfully received the indication information of the scene mode. In other words, step 740 is an optional step. If the terminal device has not successfully received the indication information, the network device may not be able to receive the confirmation information from the terminal device.

[0280] For another example, the confirmation information may be an indication bit, such as "0" or "null" indicating unsuccessful reception, and "1" indicating successful reception.

[0281] It should be understood that the specific form of the confirmation information can be many, and this application does not limit this.

[0282] In step 750, the network device communicates with the terminal device using a neural network that matches the scenario mode and the capability information of the terminal device. Correspondingly, the terminal device also communicates with the network device using a neural network that matches the scenario mode and the capability information of the terminal device.

[0283] The network device can determine that the terminal device has successfully received the above-mentioned indication information based on the confirmation information. The network device and the terminal device can then work using the transmitting neural network and the receiving neural network that match the scene mode and the capability information of the terminal device.

[0284] For example, the network device may work with a transmitting neural network that matches the scenario mode and the capability information of the terminal device, and the terminal device may work with a receiving neural network that matches the scenario mode and the capability information of the terminal device; for another example, the network device may work with a receiving neural network that matches the scenario mode and the capability information of the terminal device, and the terminal device may work with a transmitting neural network that matches the scenario mode and the capability information of the terminal device; for another example, the network device may work with a transmitting neural network that matches a certain scenario mode (such as scenario mode 1) and the capability information of the terminal device, and may also work with a receiving neural network that matches another scenario mode (such as scenario mode 2) and the capability information of the terminal device, and the terminal device may work with a receiving neural network that matches scenario mode 1 and the capability information of the terminal device, and may also work with a transmitting neural network that matches scenario mode 2 and the capability information of the terminal device. This application does not limit this.

[0285] It should be understood that in the same scenario mode, there can be a transmitting neural network and a receiving neural network matching it. It should be understood that each transmitting neural network has at least one receiving neural network that can work with it; each receiving neural network has at least one transmitting neural network that can work with it.

[0286] It should also be understood that both the transmitting end neural network and the receiving end neural network can be obtained based on the method in method 600 above.

[0287] If the network device determines that the terminal device has not successfully received the above indication information, the network device may resend the indication information, for example, re-execute the above process until receiving confirmation information from the terminal device.

[0288] It should also be understood that the network device can redetermine the scene mode when it finds that the transmission air interface condition has changed, and when the scene mode has changed, indicate the redetermined scene mode to the terminal device, so that the network device and the terminal device can work with a matching neural network based on the newly determined scene mode. That is, the above steps 720 to 750 can be repeatedly performed.

[0289] Based on the above method, the network device can select a suitable scenario mode based on the current air interface transmission conditions between the network device and the terminal device, and indicate the scenario mode to the terminal device, so that both the network device and the terminal device can determine which neural network to use based on the same scenario mode and the capability information of the terminal device, thereby facilitating performance improvement.

[0290] Fig.16 It is a schematic flowchart of a communication method provided by another embodiment of the present application. Fig.16The method 800 shown may include steps 810 to 850. Each step in the method 800 is described in detail below.

[0291] In step 810, the terminal device reports the capability information of the terminal device. Accordingly, the network device receives the capability information of the terminal device. In step 820, the terminal device determines the scene mode based on the air interface transmission condition between the terminal device and the network device.

[0292] In step 830, the terminal device sends the indication information of the scene mode to the network device. Correspondingly, the network device receives the indication information of the scene mode of the terminal device.

[0293] In step 840, the terminal device receives confirmation information from the network device, where the confirmation information is used to indicate successful reception of the indication information of the scene mode.

[0294] In step 850, the terminal device communicates with the network device using a neural network that matches the scenario mode and the capability information of the terminal device. Accordingly, the network device communicates with the terminal device using a neural network that matches the scenario mode and the capability information of the terminal device.

[0295] It should be understood that the specific process of step 820 to step 850 is similar to the specific process of step 720 to step 750 in the above method 700, except that in method 800, the terminal device determines the scene mode according to the air interface transmission condition and indicates the scene mode to the network device. The network device replies with confirmation information based on the successful reception of the indication information. Thereafter, both the network device and the terminal device can work with a neural network that matches the scene mode and the capability information of the terminal device.

[0296] Since the specific process of step 720 to step 750 has been described in detail in the above method 700, it will not be described in detail here for the sake of brevity.

[0297] It should be understood that the terminal device can redetermine the scene mode when it finds that the transmission air interface condition has changed, and when the scene mode has changed, indicate the redetermined scene mode to the network device, so that the network device and the terminal device can work with a matching neural network based on the newly determined scene mode. That is, the above steps 810 to 840 can also be repeatedly performed.

[0298] Based on the above method, the terminal device can select a suitable scenario mode based on the current air interface transmission conditions between the terminal device and the network device, and indicate the scenario mode to the network device, so that both the network device and the terminal device can determine which neural network to use based on the same scenario mode and the capability information of the terminal device, thereby facilitating performance improvement.

[0299] Above, combined Figures 6 to 16 The training method of the neural network in the communication network provided by the embodiment of the present application is described in detail. Figures 17 to 20 A training device for a neural network in a communication network provided in an embodiment of the present application is described in detail.

[0300] Fig.17 1 is a schematic block diagram of an apparatus 1000 for acquiring a neural network provided in an embodiment of the present application. Fig.17 As shown, the training device 1000 includes a processing unit 1100 and an input-output unit 1200. The processing unit 1100 can be used to determine whether the constructed target neural network meets the preset output condition based on the standardized reference data set, one or more reference neural networks and performance judgment criteria; the input-output unit 1200 can be used to output the target neural network when the output condition is met, and the output target neural network is a transmitting neural network applied to a transmitter or a receiving neural network applied to a receiver.

[0301] It should be understood that the device 1000 may correspond to Figures 6 to 14 The acquisition device in the embodiment shown may include a device for executing Figures 6 to 14 The processing unit 1100 can be used to execute steps 610 and 620 in the above method 600, and the input-output unit 1200 can be used to execute step 630 in the above method 600.

[0302] The steps performed by the processing unit 1100 may be implemented by one or more processors executing corresponding programs. For example, the training of the target neural network may be implemented by a processor dedicated to training the neural network. The processing unit 1100 may correspond to Fig.18 Processor 2010 shown in .

[0303] The steps performed by the input / output unit 1200 may be implemented by, for example, an input / output interface, a circuit, etc. The input / output unit 1200 may correspond to, for example Fig.18 The input and output interface 2020 shown in FIG.

[0304] In one possible design, the device 1000 may be deployed on a chip.

[0305] Fig.18 2 is a schematic block diagram of an apparatus 2000 for acquiring a neural network provided in an embodiment of the present application. Fig.18As shown, the device 2000 includes a processor 2100, an input / output interface 2200, and a memory 2030. The processor 2010, the input / output interface 2020, and the memory 2030 communicate with each other through an internal connection path, the memory 2030 is used to store instructions, and the processor 2010 is used to execute the instructions stored in the memory 2030 to control the input / output interface 2020 to send and / or receive signals.

[0306] It should be understood that the device 2000 may correspond to Figures 6 to 14 The acquisition device in the embodiment shown may include a device for executing Figures 6 to 14 The method 600 in the embodiment of the acquisition device is a unit of the method executed by the acquisition device. Optionally, the memory 2030 may include a read-only memory and a random access memory, and provide instructions and data to the processor. A portion of the memory may also include a non-volatile random access memory. The memory 2030 may be a separate device or may be integrated in the processor 2010. The processor 2010 may be used to execute the instructions stored in the memory 2030, and when the processor 2010 executes the instructions stored in the memory, the processor 2010 is used to execute the various steps and / or processes of the above-mentioned method embodiment corresponding to the acquisition device.

[0307] Fig.19 is a schematic block diagram of a communication device provided in an embodiment of the present application. Fig.19 As shown, the communication device 3000 may include a processing unit 3100 and a transceiver unit 3200 .

[0308] Optionally, the communication device 3000 may correspond to the terminal device in the above method embodiment, for example, it may be a terminal device, or a component configured in the terminal device (such as a circuit, a chip or a chip system, etc.).

[0309] It should be understood that the communication device 3000 may correspond to the terminal device in the method 700 or the method 800 according to the embodiment of the present application, and the communication device 3000 may include a method for executing Fig.15 Method 700 or Fig.16 The units of the method performed by the terminal device in the method 800 in the communication device 3000 are respectively for implementing Fig.15 Method 700 or Fig.16 The corresponding process of method 800 in .

[0310] Wherein, when the communication device 3000 is used to perform Fig.15When the method 700 in the embodiment is performed, the processing unit 3100 can be used to execute step 750 in the method 700, and the transceiver unit 3200 can be used to execute step 710, step 730 to step 750 in the method 700. 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.

[0311] When the communication device 3000 is used to perform Fig.16 When the method 800 in the embodiment is performed, the processing unit 3100 can be used to execute step 820 and step 850 in the method 800, and the transceiver unit 3200 can be used to execute step 810, step 830 to step 850 in the method 800. 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.

[0312] It should also be understood that when the communication device 3000 is a terminal device, the transceiver unit 3200 in the communication device 3000 can be implemented by a transceiver, for example, corresponding to Fig. 20 The transceiver 4020 or Fig.21 The transceiver 5020 in the terminal device 5000 shown in FIG. 3 , the processing unit 3100 in the communication device 3000 may be implemented by at least one processor, for example, corresponding to Fig. 20 The processor 4010 in the communication device 4000 shown in Fig.21 The processor 5010 in the terminal device 5000 is shown.

[0313] It should also be understood that when the communication device 3000 is a chip or a chip system configured in a terminal device, the transceiver unit 3200 in the communication device 3000 can be implemented through an input / output interface, circuit, etc., and the processing unit 3100 in the communication device 3000 can be implemented through a processor, microprocessor or integrated circuit integrated on the chip or chip system.

[0314] Optionally, the communication device 3000 may correspond to the network device in the above method embodiment, for example, it may be a network device, or a component configured in the network device (such as a circuit, a chip or a chip system, etc.).

[0315] It should be understood that the communication device 3000 may correspond to the network device in the method 700 or the method 800 according to the embodiment of the present application, and the communication device 3000 may include a method for executing Fig.15 Method 700 or Fig.16 The units of the method performed by the network device in the method 800 in the communication device 3000 are respectively for implementing Fig.15Method 700 or Fig.16 The corresponding process of method 800 in .

[0316] Wherein, when the communication device 3000 is used to perform Fig.15 When the method 700 is performed in the embodiment, the processing unit 3100 can be used to execute steps 720 and 750 in the method 700, and the transceiver unit 3200 can be used to execute steps 710, 730 to 750 in the method 700. 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.

[0317] When the communication device 3000 is used to perform Fig.16 When the method 800 in the embodiment is executed, the processing unit 3100 can be used to execute step 850 in the method 800, and the transceiver unit 3200 can be used to execute step 810, step 830 to step 850 in the method 800. 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.

[0318] It should also be understood that when the communication device 3000 is a network device, the transceiver unit 3200 in the communication device 3000 can be implemented by a transceiver, for example, corresponding to Fig. 20 The transceiver 4020 or Fig. 22 The RRU 6100 in the base station 6000 shown in FIG. 6A , the processing unit 3100 in the communication device 3000 may be implemented by at least one processor, for example, corresponding to Fig. 20 The processor 4010 in the communication device 4000 shown in Fig. 22 The processing unit 6200 or processor 6202 in the base station 6000 is shown.

[0319] It should also be understood that when the communication device 3000 is a chip or a chip system configured in a network device, the transceiver unit 3200 in the communication device 3000 can be implemented through an input / output interface, circuit, etc., and the processing unit 3100 in the communication device 3000 can be implemented through a processor, microprocessor or integrated circuit integrated on the chip or chip system.

[0320] Fig. 20 4 is another schematic block diagram of a communication device 4000 provided in an embodiment of the present application. Figure 6As shown, the communication device 4000 includes a processor 2010, a transceiver 4020 and a memory 4030. The processor 4010, the transceiver 4020 and the memory 4030 communicate with each other through an internal connection path, the memory 4030 is used to store instructions, and the processor 4010 is used to execute the instructions stored in the memory 4030 to control the transceiver 4020 to send signals and / or receive signals.

[0321] It should be understood that the communication device 4000 may correspond to the terminal device in the above method embodiment, and may be used to execute the various steps and / or processes executed by the network device or the terminal device in the above method embodiment. Optionally, the memory 4030 may include a read-only memory and a random access memory, and provide instructions and data to the processor. A portion of the memory may also include a non-volatile random access memory. The memory 4030 may be a separate device or may be integrated in the processor 4010. The processor 4010 may be used to execute the instructions stored in the memory 4030, and when the processor 4010 executes the instructions stored in the memory, the processor 4010 is used to execute the various steps and / or processes of the above method embodiment corresponding to the network device or the terminal device.

[0322] Optionally, the communication device 4000 is the terminal device in the previous embodiment.

[0323] Optionally, the communication device 4000 is the network device in the previous embodiment.

[0324] Among them, the transceiver 4020 may include a transmitter and a receiver. The transceiver 4020 may further include an antenna, and the number of antennas may be one or more. The processor 4010 and the memory 4030 and the transceiver 4020 may be devices integrated on different chips. For example, the processor 4010 and the memory 4030 may be integrated in a baseband chip, and the transceiver 4020 may be integrated in a radio frequency chip. The processor 4010 and the memory 4030 and the transceiver 4020 may also be devices integrated on the same chip. This application does not limit this.

[0325] Optionally, the communication device 4000 is a component configured in a terminal device, such as a circuit, a chip, a chip system, etc.

[0326] Optionally, the communication device 4000 is a component configured in a network device, such as a circuit, a chip, a chip system, etc.

[0327] The transceiver 4020 may also be a communication interface, such as an input / output interface, a circuit, etc. The transceiver 4020, the processor 4010, and the memory 4020 may be integrated into the same chip, such as a baseband chip.

[0328] Fig.21 is a schematic diagram of the structure of a terminal device 5000 provided in an embodiment of the present application. The terminal device 5000 can be applied to Figure 1 In the system shown, the functions of the terminal device in the above method embodiment are performed. As shown in the figure, the terminal device 5000 includes a processor 5010 and a transceiver 5020. Optionally, the terminal device 5000 also includes a memory 5030. Among them, the processor 5010, the transceiver 5020 and the memory 5030 can communicate with each other through an internal connection path to transmit control and / or data signals. The memory 5030 is used to store a computer program, and the processor 5010 is used to call and run the computer program from the memory 5030 to control the transceiver 5020 to send and receive signals. Optionally, the terminal device 5000 may also include an antenna 3040 for sending the uplink data or uplink control signaling output by the transceiver 5020 through a wireless signal.

[0329] The processor 5010 and the memory 5030 may be combined into a processing device, and the processor 5010 is used to execute the program code stored in the memory 5030 to implement the above functions. In specific implementation, the memory 5030 may also be integrated into the processor 5010, or independent of the processor 5010. Fig.19 The processing unit 3100 or Fig. 20 Corresponding to the processor 4010 in.

[0330] The transceiver 5020 can be used with Fig.19 The transceiver unit 3200 or Fig. 20 The transceiver 5020 may include a receiver (or receiver, receiving circuit) and a transmitter (or transmitter, transmitting circuit). The receiver is used to receive signals, and the transmitter is used to transmit signals.

[0331] It should be understood that Fig.21 The terminal device 5000 shown can realize Fig.15 or Fig.16 The method embodiment shown involves various processes of the terminal device. The operations and / or functions of each module in the terminal device 5000 are respectively to implement the corresponding processes in the above method embodiment. For details, please refer to the description in the above method embodiment. To avoid repetition, the detailed description is appropriately omitted here.

[0332] The processor 5010 can be used to execute the actions implemented by the terminal device in the previous method embodiment, and the transceiver 5020 can be used to execute the actions of the terminal device sending to or receiving from the network device described in the previous method embodiment. Please refer to the description in the previous method embodiment for details, which will not be repeated here.

[0333] Optionally, the terminal device 5000 may further include a power supply 5050 for providing power to various devices or circuits in the terminal device.

[0334] In addition, in order to make the functions of the terminal device more complete, the terminal device 5000 may also include one or more of an input unit 5060, a display unit 5070, an audio circuit 5080, a camera 5090 and a sensor 5100, and the audio circuit may also include a speaker 5082, a microphone 5084, etc.

[0335] Fig. 22 6000 is a schematic diagram of the structure of a network device provided in an embodiment of the present application, for example, a schematic diagram of the structure of a base station. The base station 6000 can be applied to Figure 1 In the system shown in the figure, the functions of the network device in the above method embodiment are performed. As shown in the figure, the base station 6000 may include one or more radio frequency units, such as a remote radio unit (RRU) 6100 and one or more baseband units (BBU) (also referred to as distributed units (DU)) 6200. The RRU 6100 may be referred to as a transceiver unit and may be connected to the base station 6000. Fig.19 The transceiver unit 3200 or Fig. 20 6100 corresponds to the transceiver 4020 in . Optionally, the RRU 6100 can also be called a transceiver, a transceiver circuit, or a transceiver, etc., which may include at least one antenna 6101 and a radio frequency unit 6102. Optionally, the RRU 6100 may include a receiving unit and a sending unit, the receiving unit may correspond to a receiver (or receiver, receiving circuit), and the sending unit may correspond to a transmitter (or transmitter, transmitting circuit). The RRU 6100 part is mainly used for receiving and sending radio frequency signals and converting radio frequency signals into baseband signals, for example, for sending indication information to terminal devices. The BBU 6200 part is mainly used for baseband processing, controlling the base station, etc. The RRU 6100 and the BBU 6200 may be physically arranged together or physically separated, that is, a distributed base station.

[0336] The BBU 6200 is the control center of the base station, which can also be called a processing unit. Fig.19 The processing unit 3100 or Fig. 20 The processor 4010 in the embodiment corresponds to the baseband processing function, such as channel coding, multiplexing, modulation, spread spectrum, etc. For example, the BBU (processing unit) can be used to control the base station to execute the operation process of the network device in the above method embodiment, for example, to generate the above indication information, etc.

[0337] In one example, the BBU 6200 may be composed of one or more single boards, and multiple single boards may jointly support a wireless access network of a single access standard (such as an LTE network), or may respectively support wireless access networks of different access standards (such as an LTE network, a 5G network, or other networks). The BBU 6200 also includes a memory 6201 and a processor 6202. The memory 6201 is used to store necessary instructions and data. The processor 6202 is used to control the base station to perform necessary actions, for example, to control the base station to execute the operation process of the network device in the above method embodiment. The memory 6201 and the processor 6202 can serve one or more single boards. In other words, a memory and a processor may be separately set on each single board. It is also possible that multiple single boards share the same memory and processor. In addition, necessary circuits may be set on each single board.

[0338] It should be understood that Fig. 22 The base station 6000 shown can achieve Fig.15 or Fig.16 The illustrated method embodiment involves various processes of the network device. The operations and / or functions of various modules in the base station 6000 are respectively to implement the corresponding processes in the above method embodiment. For details, please refer to the description in the above method embodiment. To avoid repetition, the detailed description is appropriately omitted here.

[0339] The BBU 6200 can be used to perform the actions implemented by the network device described in the previous method embodiment, and the RRU 6100 can be used to perform the actions of the network device sending to or receiving from the terminal device described in the previous method embodiment. Please refer to the description in the previous method embodiment for details, which will not be repeated here.

[0340] It should be understood that Fig. 22 The base station 6000 shown is only one possible form of network equipment and should not constitute any limitation to the present application. The method provided in the present application can be applied to network equipment in other forms. For example, it includes AAU, and may also include CU and / or DU, or include BBU and adaptive radio unit (adaptive radio unit, ARU), or BBU; it may also be customer premises equipment (customer premises equipment, CPE), and may also be in other forms. The present application does not limit the specific form of the network equipment.

[0341] Among them, the CU and / or DU can be used to perform the actions implemented by the network device in the previous method embodiment, and the AAU can be used to perform the actions of the network device sending to or receiving from the terminal device described in the previous method embodiment. Please refer to the description in the previous method embodiment for details, which will not be repeated here.

[0342] The present application also provides a processing device, comprising at least one processor, wherein the at least one processor is used to execute a computer program stored in a memory, so that the processing device executes the method executed by the acquisition device, the method executed by the network device, or the method executed by the terminal device in the above method embodiment.

[0343] The embodiment of the present application also provides a processing device, including a processor and an input / output interface. The input / output interface is coupled to the processor. The input / output interface is used to input and / or output information. The information includes at least one of an instruction and data. The processor is used to execute a computer program so that the processing device executes the method executed by the acquisition device, the method executed by the network device, or the method executed by the terminal device in the above method embodiment.

[0344] The embodiment of the present application also provides a processing device, including a processor and a memory. The memory is used to store a computer program, and the processor is used to call and run the computer program from the memory, so that the processing device executes the method executed by the training device in the above method embodiment.

[0345] It should be understood that the above-mentioned processing device can be one or more chips. For example, the processing device can be a field programmable gate array (FPGA), an application specific integrated circuit (ASIC), a system on chip (SoC), a central processor unit (CPU), a network processor (NP), a digital signal processor (DSP), a microcontroller unit (MCU), a programmable logic device (PLD) or other integrated chips.

[0346] In the implementation process, each step of the above method can be completed by an integrated logic circuit of hardware in a processor or an instruction in the form of software. The steps of the method disclosed in conjunction with the embodiment of the present application can be directly embodied as a hardware processor for execution, or a combination of hardware and software modules in a processor for execution. The software module can be located in a storage medium mature in the art such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory or an electrically erasable programmable memory, a register, etc. The storage medium is located in a memory, and the processor reads the information in the memory and completes the steps of the above method in conjunction with its hardware. To avoid repetition, it is not described in detail here.

[0347] It should be noted that the processor in the embodiment of the present application can be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method embodiment can be completed by an integrated logic circuit of hardware in the processor or an instruction in the form of software. The above processor can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The methods, steps and logic block diagrams disclosed in the embodiments of the present application can be implemented or executed. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc. The steps of the method disclosed in the embodiment of the present application can be directly embodied as a hardware decoding processor to perform, or the hardware and software modules in the decoding processor can be combined and performed. The software module can be located in a mature storage medium in the field such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory or an electrically erasable programmable memory, a register, etc. The storage medium is located in a memory, and the processor reads the information in the memory and completes the steps of the above method in combination with its hardware.

[0348] It can be understood that the memory in the embodiments of the present application can be a volatile memory or a non-volatile memory, or can include both volatile and non-volatile memories. Among them, the non-volatile memory can 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 can be a random access memory (RAM), which is used as an external cache. By way of example and not limitation, many forms of RAM are available, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), and direct RAM (DR RAM). It should be noted that the memory of the systems and methods described herein is intended to include, but is not limited to, these and any other suitable types of memory.

[0349] According to the method provided in the embodiment of the present application, the present application also provides a computer program product, which includes: a computer program code, when the computer program code is run on a computer, the computer executes Figure 6 In the embodiment shown, the method for obtaining the execution of the device, Fig.15 or Fig.16 The method executed by the terminal device or the method executed by the network device in the illustrated embodiment.

[0350] According to the method provided in the embodiment of the present application, the present application also provides a computer-readable storage medium, which stores a program code, and when the program code is run on a computer, the computer executes Figure 6 In the embodiment shown, the method for obtaining the execution of the device, Fig.15 or Fig.16 The method executed by the terminal device or the method executed by the network device in the illustrated embodiment.

[0351] According to the method provided in an embodiment of the present application, the present application also provides a system, which includes the aforementioned one or more terminal devices and one or more network devices, and the one or more terminal devices and the one or more network devices are respectively configured with a target neural network acquired by the aforementioned acquisition device.

[0352] The network devices in the above-mentioned various device embodiments completely correspond to the network devices or terminal devices in the terminal devices and method embodiments, and the corresponding modules or units perform the corresponding steps. For example, the communication unit (transceiver) performs the steps of receiving or sending in the method embodiment, and other steps except sending and receiving can be performed by the processing unit (processor). The functions of the specific units can refer to the corresponding method embodiments. Among them, the processor can be one or more.

[0353] In the above embodiments, the terminal device can be used as an example of a receiving device, and the network device can be used as an example of a sending device. However, this should not constitute any limitation to the present application. For example, the sending device and the receiving device can also be terminal devices, etc. The present application does not limit the specific types of the sending device and the receiving device.

[0354] The terms "component", "module", "system", etc. used in this specification are used to represent computer-related entities, hardware, firmware, a combination of hardware and software, software, or software in execution. For example, a component can be, but is not limited to, a process running on a processor, a processor, an object, an executable file, an execution thread, a program and / or a computer. By way of illustration, both applications running on a computing device and a computing device can be components. One or more components can reside in a process and / or an execution thread, and a component can be located on a computer and / or distributed between two or more computers. In addition, these components can be executed from various computer-readable media having various data structures stored thereon. Components can, for example, communicate through local and / or remote processes according to signals having one or more data packets (e.g., data from two components interacting with another component between a local system, a distributed system and / or a network, such as the Internet interacting with other systems through signals).

[0355] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0356] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0357] In the several embodiments provided in the present application, it should be understood that the disclosed systems, 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. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that 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.

[0358] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0359] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0360] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in the various embodiments of the present application. The aforementioned storage media include: various media that can store program codes, such as USB flash drives, mobile hard disks, read-only memories (ROM), random access memories (RAM), magnetic disks or optical disks.

[0361] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A method for obtaining a neural network, characterized in that: include: Based on a standardized reference data set, one or more reference neural networks and a performance judgment criterion, determining that the constructed target neural network meets a preset output condition; Output indication information of the target neural network, where the target neural network is applied to a transmitting end target neural network or to a receiving end target neural network.

2. The method according to claim 1, characterized in that The target neural network at the transmitting end is used for constellation modulation, channel coding, precoding, or extraction and compression of channel information; the target neural network at the receiving end is used for constellation demodulation, channel decoding, detection reception, or reconstruction of channel information.

3. The method according to any one of claims 1 or 2, characterized in that The method further comprises: Determining a configuration suitable for the current application scenario of the target neural network from at least one set of standardized configurations; each set of standardized configurations includes one or more of the following: a standardized reference data set, a standardized reference neural network, a standardized data structure, a standardized performance measurement function, a standardized performance judgment criterion, and a standardized achievable capability measurement value; Among them, the reference neural network includes a reference neural network applied to a receiving end and / or a reference neural network applied to a transmitting end; the data structure is used to generate and / or parse air interface information; the performance measurement function is used to generate a performance measurement value, the performance judgment criterion includes a performance indicator and an output condition of the target neural network, and the performance judgment criterion is used to determine that the target neural network meets the output condition based on the performance measurement value and the performance indicator; the achievable capability measurement value is used to select a matching neural network and / or a matching performance indicator.

4. The method according to claim 3, characterized in that The target neural network is a transmitting end target neural network, and the one or more reference neural networks are one or more receiving end reference neural networks; as well as The step of determining whether the designed target neural network meets the preset output condition based on the standardized reference data set, one or more reference neural networks and performance judgment criteria includes: Inputting the sender data obtained from the reference data set into the sender target neural network, the sender target neural network is used to process the sender data according to a predefined data structure to generate air interface information to be sent to the one or more receiving reference neural networks; Acquire one or more output results from the one or more receiving-end reference neural networks, wherein the one or more output results are obtained by the one or more receiving-end reference neural networks based on the received air interface information respectively; Using the one or more output results and the sending end data as inputs of the performance metric function to obtain one or more performance metric values; According to the one or more performance metrics and the performance judgment criterion, it is determined that the target neural network meets the output condition.

5. The method according to claim 3, characterized in that The target neural network is a transmitting end target neural network, and the one or more reference neural networks are one or more transmitting end reference neural networks; as well as The step of determining whether the designed target neural network meets the preset output condition based on the standardized reference data set, one or more reference neural networks and performance judgment criteria includes: Inputting the sender data obtained from the reference data set into the sender target neural network, wherein the sender target neural network is used to process the sender data according to a predefined data structure to generate target air interface information; Inputting the transmitting end data into the one or more transmitting end reference neural networks, wherein the one or more transmitting end reference neural networks are used to process the transmitting end data according to a predefined data structure to generate one or more reference air interface information; Using the one or more reference air interface information and the target air interface information as inputs of the performance metric function to obtain one or more performance metric values; According to the one or more performance metrics and the performance judgment criterion, it is determined that the target neural network meets the output condition.

6. The method according to claim 3, characterized in that The target neural network is a receiving end target neural network, and the one or more reference neural networks are one or more sending end reference neural networks; as well as The step of determining whether the designed target neural network meets the preset output condition based on the standardized reference data set, one or more reference neural networks and performance judgment criteria includes: Inputting the sender data obtained from the reference data set into the one or more sender reference neural networks, wherein the one or more sender reference neural networks are used to process the sender data according to a predefined data structure to obtain one or more air interface information to be sent to the receiving end target neural network; Acquire one or more output results from the receiving end target neural network, where the one or more output results are output results respectively generated by the receiving end target neural network according to the one or more air interface information received; Using the one or more output results and the sending end data as inputs of the performance metric function to obtain one or more performance metric values; According to the one or more performance metrics and the performance judgment criterion, it is determined that the target neural network meets the output condition.

7. The method according to claim 3, characterized in that The target neural network is a receiving end target neural network, and the one or more reference neural networks include one or more sending end reference neural networks and one or more receiving end reference neural networks; as well as The step of determining whether the designed target neural network meets the preset output condition based on the standardized reference data set, one or more reference neural networks and performance judgment criteria includes: Inputting the sender data obtained from the reference data set into the one or more sender reference neural networks, wherein the one or more sender reference neural networks are used to process the sender data according to a predefined data structure to obtain one or more air interface information to be sent to the receiving end target neural network and the one or more receiving end reference neural networks; Acquire one or more target output results from the receiving end target neural network, where the one or more target output results are output results respectively generated by the receiving end target neural network according to the one or more air interface information received; Acquire one or more reference output results from the one or more receiving-end reference neural networks, where the one or more reference output results are output results generated by the one or more receiving-end reference neural networks respectively according to the one or more received air interface information; Using the one or more target output results and the one or more reference output results as inputs of the performance metric function to obtain one or more performance metric values; According to the one or more performance metrics and the performance judgment criterion, it is determined that the target neural network meets the output condition.

8. The method according to any one of claims 3 to 7, characterized in that The performance metric function includes one or more of the following: mean square error, normalized mean square error, mean absolute error, maximum absolute error, correlation coefficient, cross entropy, mutual information, bit error rate, or frame error rate.

9. The method according to any one of claims 3 to 8, characterized in that The achievable capability metric value includes one or more of the following: overhead of air interface information, amount of computation required for neural network implementation, amount of parameters stored for neural network implementation, and computational accuracy required for neural network implementation.

10. The method according to any one of claims 3 to 9, characterized in that The performance indicator includes one or more performance indicators, and the performance judgment criterion includes determining whether the output condition is satisfied based on a comparison between the performance metric value and the one or more performance indicators.

11. A device for obtaining a neural network, characterized in that: include: A processing unit, configured to determine whether the designed target neural network meets a preset output condition based on a standardized reference data set, one or more reference neural networks and a performance judgment criterion; The input-output unit is used to output the target neural network, and the output target neural network is applied to the sending end target neural network or the receiving end target neural network.

12. The device according to claim 11, characterized in that The target neural network at the transmitting end is used for constellation modulation, channel coding, precoding, or extraction and compression of channel information; the target neural network at the receiving end is used for constellation demodulation, channel decoding, detection reception, or reconstruction of channel information.

13. The device according to any one of claims 11 or 12, characterized in that The processing unit is further used to determine a configuration suitable for the current application scenario of the target neural network from at least one set of standardized configurations; each set of standardized configurations includes one or more of the following: a standardized reference data set, a standardized reference neural network, a standardized data structure, a standardized performance measurement function, a standardized performance judgment criterion, and a standardized achievable capability measurement value; Among them, the reference neural network includes a reference neural network applied to a receiving end and / or a reference neural network applied to a transmitting end; the data structure is used to generate and / or parse air interface information; the performance measurement function is used to generate a performance measurement value, the performance judgment criterion includes a performance indicator and an output condition of the target neural network, and the performance judgment criterion is used to determine that the target neural network meets the output condition based on the performance measurement value and the performance indicator; the achievable capability measurement value is used to select a matching neural network and / or a matching performance indicator.

14. The device according to claim 13, characterized in that The target neural network is a receiving end target neural network, and the one or more reference neural networks include one or more sending end reference neural networks and one or more receiving end reference neural networks; The input-output unit is also used for: Inputting the sender data obtained from the reference data set into the sender target neural network, the sender target neural network is used to process the sender data according to a predefined data structure to generate air interface information to be sent to the one or more receiving reference neural networks; Acquire one or more output results from the one or more receiving-end reference neural networks, wherein the one or more output results are obtained by the one or more receiving-end reference neural networks based on the received air interface information respectively; Using the one or more output results and the sending end data as inputs of the performance metric function to obtain one or more performance metric values; The processing unit is specifically used for: According to the one or more performance metrics and the performance judgment criterion, it is determined that the target neural network meets the output condition.

15. The device according to claim 13, characterized in that The target neural network is a transmitting end target neural network, and the one or more reference neural networks are one or more transmitting end reference neural networks; The input-output unit is also used for: Inputting the sender data obtained from the reference data set into the sender target neural network, wherein the sender target neural network is used to process the sender data according to a predefined data structure to generate target air interface information; Inputting the transmitting end data into the one or more transmitting end reference neural networks, wherein the one or more transmitting end reference neural networks are used to process the transmitting end data according to a predefined data structure to generate one or more reference air interface information; Using the one or more reference air interface information and the target air interface information as inputs of the performance metric function to obtain one or more performance metric values; The processing unit is specifically used for: According to the one or more performance metrics and the performance judgment criterion, it is determined that the target neural network meets the output condition.

16. The device according to claim 13, characterized in that The target neural network is a receiving end target neural network, and the one or more reference neural networks are one or more sending end reference neural networks; The input-output unit is also used for: Inputting the sender data obtained from the reference data set into the one or more sender reference neural networks, wherein the one or more sender reference neural networks are used to process the sender data according to a predefined data structure to obtain one or more air interface information to be sent to the receiving end target neural network; Acquire one or more output results from the receiving end target neural network, where the one or more output results are output results respectively generated by the receiving end target neural network according to the one or more air interface information received; Using the one or more output results and the sending end data as inputs of the performance metric function to obtain one or more performance metric values; The processing unit is specifically used for: According to the one or more performance metrics and the performance judgment criterion, it is determined that the target neural network meets the output condition.

17. The device according to claim 13, characterized in that The target neural network is a receiving end target neural network, and the one or more reference neural networks include one or more sending end reference neural networks and one or more receiving end reference neural networks; The input-output unit is also used for: Inputting the sender data obtained from the reference data set into the one or more sender reference neural networks, wherein the one or more sender reference neural networks are used to process the sender data according to a predefined data structure to obtain one or more air interface information to be sent to the receiving end target neural network and the one or more receiving end reference neural networks; Acquire one or more target output results from the receiving end target neural network, where the one or more target output results are output results respectively generated by the receiving end target neural network according to the one or more air interface information received; Acquire one or more reference output results from the one or more receiving-end reference neural networks, where the one or more reference output results are output results generated by the one or more receiving-end reference neural networks respectively according to the one or more received air interface information; Using the one or more target output results and the one or more reference output results as inputs of the performance metric function to obtain one or more performance metric values; The processing unit is specifically used for: According to the one or more performance metrics and the performance judgment criterion, it is determined that the target neural network meets the output condition.

18. The device according to any one of claims 13 to 17, characterized in that The performance metric function includes one or more of the following: mean square error, normalized mean square error, mean absolute error, maximum absolute error, correlation coefficient, cross entropy, mutual information, bit error rate, or frame error rate.

19. The device according to any one of claims 13 to 18, characterized in that The achievable capability metric value includes one or more of the following: overhead of air interface information, amount of computation required for neural network implementation, amount of parameters stored for neural network implementation, and computational accuracy required for neural network implementation.

20. The device according to any one of claims 13 to 19, characterized in that The performance indicator includes one or more performance indicators, and the performance judgment criterion includes determining whether the output condition is satisfied based on a comparison between the performance metric value and the one or more performance indicators.

21. A computer-readable storage medium, characterized in that: The invention comprises a computer program, which, when being run on a computer, causes the computer to execute the method according to any one of claims 1 to 10.

Images