Training method and communication device

Abstract

The embodiment of the invention provides a training method and a communication device so as to improve the alignment efficiency among multiple module models. The method comprises the following steps: receiving training data, and training a first module group according to the training data; wherein the training data is used for training a first module group, the training data is obtained based on a second module group in a second model, the first module group comprises all modules in the first model or a part of cascaded modules, and the second module group comprises all modules in the second model or a part of cascaded modules; the first module group comprises modules corresponding to the modules in the second module group, the first module group comprises target modules, and the target modules are modules, which are not aligned with the corresponding modules in the second module group, in the first model.

Description

Technical Field

[0001] This application relates to the field of communication technologies, and in particular, to a training method and a communication device. Background Art

[0002] Transmitters and receivers based on neural networks, such as encoders and decoders, can be trained by actual data driving, and can be optimized according to scenarios to achieve better transmit signal design and receive performance. Neural network transceivers can be used for physical layer signal processing, such as symbol modulation-demodulation, channel coding-decoding, pilot-channel estimation, etc.; they can also be used for data processing, such as channel state information compression-reconstruction.

[0003] The neural network models of transmitters and receivers can be jointly optimized to achieve the best performance. For example, between communication devices, the transmitters / receivers trained by one communication device can be adapted to the receivers / transmitters trained by the peer communication device to achieve the best end-to-end performance.

[0004] The neural network models of transmitters and receivers generally include multiple neural network modules. For example, they include neural network modules for symbol modulation / demodulation, neural network modules for channel coding / decoding, neural network modules for channel state information compression / reconstruction, etc. Currently, for the neural network models of transmitters and receivers including multiple neural network modules, there is a problem of low joint optimization efficiency. Summary of the Invention

[0005] This application provides a training method and a communication device to improve the joint optimization efficiency of multi-module models.

[0006] In a first aspect, a training method is provided. This method is executed by a first communication device (the first communication device can be a terminal device or a network device), or, this method is executed by some components in the first communication device (such as a processor, a chip, or a chip system, etc.), or this method can also be implemented by a logic module or software that can implement all or part of the functions of the first communication device.

[0007] The method includes receiving training data and then training a first module group according to the training data. The training data is used to train the first module group and is obtained based on a second module group in a second model. The first module group includes all modules in the first model or some cascaded modules. The second module group includes all modules in the second model or some cascaded modules. The first module group includes modules corresponding to those in the second module group. The first module group includes target modules, where the target modules are the modules in the first model that are not aligned with the corresponding modules in the second module group. For the unaligned target modules in the first model, by training the first module group including the target modules in units of module groups, the training efficiency can be improved and the communication overhead during training can be reduced.

[0008] Among them, the first model is the model of the first communication device, and the second model is the model of the second communication device. The first model is used to implement at least some functions in the transmitter or receiver of the first communication device, for example. The second model is used to implement at least some functions in the transmitter or receiver of the second communication device, for example. The first model and the second model are related models. In one implementation, the first model and the second model are used to implement reciprocal functions. For example, the second model is used to perform inverse processing on the output of the first model, that is, the first model is used to implement at least some functions of the transmitter, and the second model is used to implement at least some functions of the receiver. Or the first model is used to perform inverse processing on the output of the second model, that is, the first model is used to implement at least some functions of the receiver, and the second model is used to implement at least some functions of the transmitter. In another implementation, the first model and the second model are used to implement the same function. For example, both the first model and the second model are used to implement at least some functions in the transmitter or receiver.

[0009] The first model includes multiple modules, and the second model includes multiple modules. Each module is used to implement one or more functions in the transmitter or receiver. The module can be a neural network model. There is a corresponding relationship between the multiple modules in the first model and the multiple modules in the second model. In one implementation, when the first model and the second model are used to implement reciprocal functions, a pair of corresponding modules (one belonging to the first model and the other belonging to the second model) in the first model and the second model can be two modules used to implement reciprocal functions. In another implementation, when the first model and the second model are used to implement the same function, a pair of corresponding modules in the first model and the second model can be two modules used for the same function.

[0010] Multiple modules of the first model are cascaded, that is, the output of one module can be used as the input of at least one other module, or the input of one module comes from the output of at least one other module. Multiple modules of the second model are cascaded, that is, the output of one module can be used as the input of at least one other module, or the input of one module comes from the output of at least one other module.

[0011] Alignment means that the performance of a pair of modules (module pair) meets the requirements. For example, when a pair of modules jointly process a set of data and the value of the loss function corresponding to the obtained processing result is less than the loss threshold, the two modules in the module pair are aligned; otherwise, the two modules in the module pair are not aligned. For example, a pair of modules is used to implement reciprocal functions. That is, one of the modules (module x) is used to implement the function in the transmitter, and the other module (module y) is used to implement the function in the receiver. Then, if the value of the loss function between the output data of module y and the input data of module x is less than the loss threshold, it can be considered that module x and the second module y are aligned; conversely, it can be considered that module x and module y are not aligned (unaligned). Another example is that a pair of modules is used to implement the same function. Then, for the same input data, if the value of the loss function between the output data of the two modules is less than the loss threshold, it can be considered that the two modules in the module pair are aligned; conversely, it can be considered that the two modules in the module pair are not aligned (unaligned).

[0012] In a possible implementation, the first module group includes an adaptation module, and the adaptation module and the target module are different modules. The adaptation module is a module used to make the performance of the first module group meet the requirements. The adaptation module includes a neural network model, and the parameters in the adaptation module can be trained and adjusted to make the performance of the first module group meet the requirements, that is, to align the first module group with the second module group, or to make the first module group adapt to the second module group. Training the first module group according to the training data includes: adjusting the parameters in the adaptation module according to the training data to align the first module group with the second module group. When training the first module group, taking the performance of the first module group as the goal and adjusting the parameters of the adaptation module without adjusting the parameters of the target module can ensure the generality of the target module. Optionally, since the target module is used to implement one or more functions in the receiver or transmitter, and the adaptation module is used to align the first module group with the second module group, the parameters in the adaptation module are usually fewer than those in the target module, which can reduce the training data required in the training process, reduce the training complexity, and improve the training efficiency.

[0013] In a possible implementation, the adaptation module can be adjacent to the target module. By setting the adaptation module adjacent to the target module, the adaptation module can be regarded as a detachable network layer of the target module. The parameters of the adaptation module can be adjusted without adjusting the parameters of the target module, so as to retain the generality of the target module and improve the training efficiency. Alternatively, the adaptation module is disposed on the output side or the input side of the first module group. Alternatively, the adaptation module can be disposed between any two modules in the first module group.

[0014] In a possible implementation, the first module group includes N target modules and N adaptation modules. The N target modules correspond to the N adaptation modules one by one, and each adaptation module is adjacent to the corresponding target module; N is an integer greater than or equal to 1.

[0015] In a possible implementation, the first module group includes M target modules and 1 adaptation module; M is an integer greater than or equal to 2. That is, when there are multiple misaligned modules, taking the performance of the first module group as the goal, the alignment of the first module group and the second module group can be achieved through 1 adaptation module, without independently training and adapting each target module, which can improve the alignment efficiency between the first model and the second model and reduce the communication overhead between the first communication device and the second communication device during the training and adaptation process.

[0016] In a possible implementation, training the first module group according to the training data includes: adjusting the parameters in the target module according to the training data. That is, the alignment between the first module group and the second module group can be achieved by adjusting the parameters in the target module. The first module group can include multiple target modules. Taking the first module group as a unit for alignment, multiple target modules can be optimized and adapted simultaneously, thereby improving the alignment efficiency between the first module group and the second module group.

[0017] In a possible implementation, the training data includes at least one of the following: the input data of the second module group and the output data of the second module group; the gradient obtained based on the input data of the first module group, the output data of the first module group, and the output data of the second module group; the second module group; the second model. That is, the training data is provided in units of module groups, and the first module group is trained with the performance of the module group as the goal, thereby improving the alignment efficiency between the first module group and the second module group.

[0018] In a possible implementation, the method further includes: sending or receiving verification data, where the verification data is used to determine the target module. By verifying the module pairs in the first model and the second model, the misaligned target modules can be determined, and then the first module group can be determined based on the target modules, and training and adaptation can be performed based on the first module group, thereby improving the alignment efficiency between the first module group and the second module group.

[0019] In a possible implementation, the verification data includes first verification data, and the first verification data is used to verify whether the third module group and the fourth module group are aligned. The third module group includes all the modules or some cascaded modules in the first model, and the fourth module group includes the modules corresponding to the modules in the third module group in the second model. Verifying whether module groups are aligned in units of module groups can improve the verification efficiency.

[0020] In a possible implementation, the verification data includes second verification data, and the second verification data is used to verify whether the fifth module group and the sixth module group are aligned. The fifth module group is a subset of the third module group, and the sixth module group is a subset of the fourth module group.

[0021] In a possible implementation, the verification data includes third verification data, and the third verification data is used to verify whether the seventh module group and the eighth module group are aligned. The third module group is a subset of the seventh module group, and the fourth module group is a subset of the eighth module group.

[0022] In a possible implementation, the verification data includes fourth verification data, and the fourth verification data is used to verify whether the ninth module group and the tenth module group are aligned. The third module group, the fourth module group, the ninth module group, and the tenth module group all include at least two modules. The modules in the third module group do not overlap with the modules in the ninth module group, and the modules in the fourth module group do not overlap with the modules in the tenth module group.

[0023] In a second aspect, a training method is provided. This method is executed by a second communication device (the second communication device can be a terminal device or a network device), or this method is executed by some components in the second communication device (such as a processor, a chip, or a chip system, etc.), or this method can also be implemented by a logic module or software that can implement all or part of the functions of the second communication device.

[0024] The method includes: sending training data, where the training data is used to train the first module group. The training data is obtained based on the second module group in the second model. The first module group includes all the modules or some cascaded modules in the first model, the second module group includes all the modules or some cascaded modules in the second model, the first module group includes the modules corresponding to the modules in the second module group, and the first module group includes a target module, where the target module is the module in the first model that is not aligned with the corresponding module in the second module group.

[0025] In a possible implementation, the training data includes at least one of the following: the input data of the second module group and the output data of the second module group; the gradient obtained based on the input data of the first module group, the output data of the first module group, and the output data of the second module group; the second module group; the second model.

[0026] In a possible implementation, the method further includes: sending or receiving verification data for determining a target module.

[0027] In a possible implementation, the verification data includes first verification data for verifying whether a third module group is aligned with a fourth module group. The third module group includes all modules or some cascaded modules in a first model, and the fourth module group includes modules corresponding to the modules in the third module group in a second model.

[0028] In a possible implementation, the verification data includes second verification data for verifying whether a fifth module group is aligned with a sixth module group. The fifth module group is a subset of the third module group, and the sixth module group is a subset of the fourth module group.

[0029] In a possible implementation, the verification data includes third verification data for verifying whether a seventh module group is aligned with an eighth module group. The third module group is a subset of the seventh module group, and the fourth module group is a subset of the eighth module group.

[0030] A third aspect provides a communication device. The communication device has the functions to implement the behaviors in the method examples of the first aspect above. The beneficial effects can be referred to the description of the first aspect and will not be elaborated here. The communication device can be the first communication device in the first aspect, or the communication device can be a device such as a chip or a chip system that can support the first communication device in the first aspect to implement the functions required by the method provided in the first aspect.

[0031] In a possible design, the communication device includes corresponding means or modules for executing the method of the first aspect. For example, the communication device includes a processing unit (sometimes also referred to as a processing module) and / or a transceiver unit (sometimes also referred to as a transceiver module). These units (modules) can execute the corresponding functions in the method examples of the first aspect above. For specific details, refer to the detailed description in the method examples and will not be elaborated here.

[0032] A fourth aspect provides a communication device. The communication device has the functions to implement the behaviors in the method examples of the second aspect above. The beneficial effects can be referred to the description of the second aspect and will not be elaborated here. The communication device can be the second communication device in the second aspect, or the communication device can be a device such as a chip or a chip system that can support the second communication device in the second aspect to implement the functions required by the method provided in the second aspect.

[0033] In a possible design, the communication device includes corresponding means or modules for performing the method of the second aspect. For example, the communication device includes a processing unit (sometimes also referred to as a processing module) and / or a transceiver unit (sometimes also referred to as a transceiver module). These units (modules) can perform the corresponding functions in the method examples of the second aspect above. For specific details, refer to the detailed description in the method examples and will not be elaborated here.

[0034] In a fifth aspect, an embodiment of the present application provides a communication device. The communication device can be the communication device in the third aspect or the fourth aspect in the above embodiments, or a chip or a chip system disposed in the communication device in the third aspect or the fourth aspect. The communication device includes a communication interface and a processor. Optionally, it further includes a memory. The memory is used to store computer programs or instructions or data. The processor is coupled to the memory and the communication interface. When the processor reads the computer programs or instructions or data, the communication device executes the methods performed by the terminal device or the network device in the above method embodiments.

[0035] In a sixth aspect, an embodiment of the present application provides a communication device. The communication device includes at least one processor. Optionally, it further includes a memory. The at least one processor is coupled to the memory. The at least one processor is used to execute the methods described in the first aspect or the second aspect.

[0036] In a seventh aspect, an embodiment of the present application provides a chip system. The chip system includes a processor and may further include a memory and / or a communication interface for implementing the methods described in the first aspect or the second aspect. In a possible implementation, the chip system further includes a memory for storing program instructions and / or data. The chip system can be composed of chips or can include chips and other discrete devices.

[0037] In an eighth aspect, an embodiment of the present application provides a communication system. The communication system includes a communication device for performing the method described in the first aspect and a communication device for performing the method described in the second aspect. Among them, the communication device for performing the method described in the first aspect is, for example, the first communication device described in the first aspect, and the communication device for performing the method described in the second aspect is, for example, the second communication device described in the second aspect.

[0038] In a ninth aspect, the present application provides a computer-readable storage medium. The computer-readable storage medium stores a computer program. When the computer program is run, it implements the method in any one of the first aspect to the second aspect above.

[0039] In a tenth aspect, a computer program product is provided, which includes computer program code that, when run, causes the method in any one of the first to second aspects to be executed.

[0040] Among them, for the technical effects brought by any one of the design manners in the second to tenth aspects, reference can be made to the technical effects brought by different design manners in the first aspect above, which will not be elaborated here. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1a A schematic diagram of a communication system provided by the present application;

[0042] Figure 1b A schematic diagram of another communication system provided by the present application;

[0043] Figure 2 A schematic flowchart of a training method provided by the present application;

[0044] Figure 3 A schematic diagram of a model verification method provided by the present application;

[0045] Figure 4 A schematic diagram of another model verification method provided by the present application;

[0046] Figure 5 A schematic diagram of the structure of a first module group provided by the present application;

[0047] Figure 6 A schematic diagram of the implementation of an adaptation module provided by the present application;

[0048] Figure 7 A schematic diagram of the implementation of another adaptation module provided by the present application;

[0049] Figure 8 A schematic diagram of a communication device provided by the present application;

[0050] Figure 9 A schematic diagram of another communication device provided by the present application;

[0051] Figure 10 A schematic diagram of another communication device provided by the present application;

[0052] Figure 11 A schematic diagram of another communication device provided by the present application;

[0053] Figure 12 A schematic diagram of another communication device provided by the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0054] First, some terms in the embodiments of the present application are explained to facilitate understanding by those skilled in the art.

[0055] (1) Terminal device: It can be a wireless terminal device capable of receiving scheduling and indication information from a network device. The wireless terminal device can be a device that provides voice and / or data connectivity to a user, or a handheld device with a wireless connection function, or other processing devices connected to a wireless modem.

[0056] The terminal device can communicate with one or more core networks or the Internet via a radio access network (RAN). The terminal device can be a mobile terminal device, such as a mobile phone (or a "cellular" phone, mobile phone), a computer, and a data card. For example, it can be a portable, pocket-sized, handheld, computer-integrated, or vehicle-mounted mobile device that exchanges voice and / or data with the radio access network. For example, personal communication service (PCS) phones, cordless phones, session initiation protocol (SIP) phones, wireless local loop (WLL) stations, personal digital assistants (PDAs), tablets (Pads), computers with wireless transceiver functions, and other devices. The wireless terminal device can also be referred to as a system, a subscriber unit, a subscriber station, a mobile station, a mobile station (MS), a remote station, an access point (AP), a remote terminal device, an access terminal device, a user terminal device, a user agent, a subscriber station (SS), a customer premises equipment (CPE), a terminal, a user equipment (UE), a mobile terminal (MT), etc.

[0057] By way of example and not limitation, in the embodiments of the present application, the terminal device may also be a wearable device. Wearable devices, also known as wearable intelligent devices or smart wearable devices, etc., are the general term for devices developed by applying wearable technology to the intelligent design of daily wear, such as glasses, gloves, watches, clothing, and shoes. Wearable devices are portable devices that are directly worn on the body or integrated into the user's clothes or accessories. Wearable devices are not only a hardware device, but also achieve powerful functions through software support, data interaction, and cloud interaction. Broadly speaking, wearable intelligent devices include those with complete functions and large sizes that can achieve complete or partial functions without relying on a smartphone, such as smart watches or smart glasses, etc., and those that only focus on a certain type of application function and need to cooperate with other devices such as smartphones, such as various smart bracelets for physical sign monitoring, smart helmets, and smart jewelry.

[0058] The terminal may also be a drone, a robot, a terminal in device-to-device (D2D) communication, a terminal in vehicle-to-everything (V2X) communication, a virtual reality (VR) terminal device, an augmented reality (AR) terminal device, a wireless terminal in industrial control, a wireless terminal in self-driving, a wireless terminal in remote medical, a wireless terminal in smart grid, a wireless terminal in transportation safety, a wireless terminal in smart city, a wireless terminal in smart home, etc.

[0059] In addition, the terminal device may also be a terminal device in a communication system evolved after the fifth-generation (5G) communication system (such as the sixth-generation (6G) communication system, etc.) or a terminal device in a future-evolved public land mobile network (PLMN). Exemplarily, the 6G network can further expand the form and function of 5G communication terminals, and 6G terminals include but are not limited to vehicles, cellular network terminals (integrating satellite terminal functions), drones, and Internet of Things (IoT) devices.

[0060] In the embodiments of the present application, the above terminal device may also be a device with an AI model, which can process the data to be sent or the received signal based on the AI model.

[0061] (2) Network device: It can be a device in a wireless network. For example, the network device can be a RAN node (or device) that connects a terminal device to a wireless network, and can also be called a base station. Currently, some examples of RAN devices are: base station, evolved NodeB (eNodeB), gNB (gNodeB) in a 5G communication system, transmission reception point (TRP), evolved Node B (eNB), radio network controller (RNC), Node B (NB), home base station (e.g., home evolved Node B, or home Node B, HNB), base band unit (BBU), or wireless fidelity (Wi-Fi) access point AP, etc. Additionally, in a network structure, the network device can include a centralized unit (CU) node, or a distributed unit (DU) node, or a RAN device including a CU node and a DU node.

[0062] Optionally, the RAN node can also be a macro base station, a micro base station or an indoor station, a relay node or a donor node, or a radio controller in a cloud radio access network (CRAN) scenario. The RAN node can also be a server, a wearable device, a vehicle or an in-vehicle device, etc. For example, the access network device in vehicle to everything (V2X) technology can be a road side unit (RSU).

[0063] In another possible scenario, multiple RAN nodes cooperate to assist a terminal in achieving wireless access, and different RAN nodes respectively implement partial functions of a base station. For example, the RAN node can be a central unit (CU), a distributed unit (DU), a CU-control plane (CP), a CU-user plane (UP), or a radio unit (RU), etc. The CU and the DU can be set separately, or can also be included in the same network element, such as a baseband unit (BBU). The RU can be included in a radio frequency device or a radio frequency unit, such as included in a remote radio unit (RRU), an active antenna unit (AAU), or a remote radio head (RRH).

[0064] In different systems, the CU (or CU-CP and CU-UP), DU, or RU may also have different names, but those skilled in the art can understand their meanings. For example, in an open RAN (O-RAN or ORAN) system, the CU can also be called an O-CU (open CU), the DU can also be called an O-DU, the CU-CP can also be called an O-CU-CP, the CU-UP can also be called an O-CU-UP, and the RU can also be called an O-RU. For the convenience of description, in this application, the CU, CU-CP, CU-UP, DU, and RU are used as examples for description. Any one of the CU (or CU-CP, CU-UP), DU, and RU in this application can be implemented by a software module, a hardware module, or a combination of a software module and a hardware module.

[0065] The communication between the access network device and the terminal device follows a certain protocol layer structure. The protocol layer may include a control plane protocol layer and a user plane protocol layer. The control plane protocol layer may include at least one of the following: radio resource control (RRC) layer, packet data convergence protocol (PDCP) layer, radio link control (RLC) layer, media access control (MAC) layer, or physical (PHY) layer, etc. The user plane protocol layer may include at least one of the following: service data adaptation protocol (SDAP) layer, PDCP layer, RLC layer, MAC layer, or physical layer, etc.

[0066] For the correspondence between the network elements in the ORAN system and their achievable protocol layer functions, refer to Table 1 below.

[0067] Table 1

[0068] ORAN Network Element 3GPP Protocol Layer Function O-CU-CP RRC+PCDP - Control Plane (PDCP-C) O-CU-UP SDAP+PCDP - User Plane (PDCP-U) O-DU RLC+MAC+PHY-high O-RU PHY-low

[0069] The network device may be other devices that provide wireless communication functions for the terminal device. The specific technologies and specific device forms adopted by the network device are not limited in the embodiments of the present application. For ease of description, the embodiments of the present application do not limit.

[0070] The network device may further include core network devices, such as the mobility management entity (MME), home subscriber server (HSS), serving gateway (S-GW), policy and charging rules function (PCRF), and public data network gateway (PDN gateway, P-GW) in the 4th generation (4G) network; network elements such as the access and mobility management function (AMF), user plane function (UPF), or session management function (SMF) in the 5G network. In addition, the core network device may further include other core network devices in the 5G network and the next-generation network of the 5G network.

[0071] In the embodiments of this application, the above network device may further have a network node with an AI model, which can process the data to be sent or the received signal based on the AI model.

[0072] In the embodiments of this application, the device for implementing the functions of the network device may be the network device or a device capable of supporting the network device to implement such functions, such as a chip system, and this device may be installed in the network device. In the technical solutions provided in the embodiments of this application, the device for implementing the functions of the network device is taken as an example of the network device to describe the technical solutions provided in the embodiments of this application.

[0073] (3) The AI model, that is, the AI algorithm (or AI operator), is a general term for mathematical algorithms constructed based on the principles of artificial intelligence and is also the basis for using AI to solve specific problems. According to the different specific methods and / or technologies for implementing artificial intelligence, the AI model may specifically be referred to as a machine learning model, a deep learning model, or a reinforcement learning model. Machine learning is a method for implementing artificial intelligence, and the goal of this method is to design and analyze some algorithms (that is, models) that allow a computer to automatically "learn", and the designed algorithms are called machine learning models. A machine learning model is a type of algorithm that automatically analyzes and obtains rules from data and uses the rules to predict unknown data.

[0074] The typical structure of current deep learning models is a deep neural network. A neural network is a mathematical model or computational model that mimics the structure and function of a biological neural network (the central nervous system of an animal, especially the brain). A neural network performs calculations through a large number of neuron connections. A neural network can include multiple neural network layers with different functions, and each layer includes parameters and calculation rules. Depending on the different calculation formulas or functions, different layers in a neural network have different names. For example, the layer that performs convolutional calculations is called a convolutional layer, and convolutional layers are often used for feature extraction of input signals. A neural network can also be composed of multiple sub-neural networks combined. Neural networks with different structures can be applicable to different scenarios (e.g., classification, recognition) or provide different effects when used in the same scenario. The differences in the structure of a neural network specifically include one or more of the following: the number of network layers in the neural network is different, the order of each network layer is different, the weights, parameters, or calculation formulas in each network layer are different. There are already many different neural networks with high accuracy for application scenarios such as recognition or classification in the industry. Some neural networks can be trained with specific datasets and used alone to complete a task or combined with other neural networks (or other functional modules) to complete a task.

[0075] (4) The terms "system" and "network" in the embodiments of the present application can be used interchangeably. "Multiple" means two or more. "And / or" describes the association relationship of associated objects and indicates that three relationships can exist. For example, A and / or B can represent: A exists alone, A and B exist simultaneously, or B exists alone, where A and B can be singular or plural. The character " / " generally represents an "or" relationship between the associated objects before and after. "At least one (item)" or its similar expression refers to any combination of these items, including any combination of single items (items) or multiple items (items). For example, "at least one of A, B, and C" includes A, B, C, AB, AC, BC, or ABC. Also, unless otherwise specified, the ordinal numbers such as "first" and "second" mentioned in the embodiments of the present application are used to distinguish multiple objects and do not limit the order, time sequence, priority, or importance of multiple objects.

[0076] (5) In the embodiments of the present application, "send" and "receive" represent the direction of signal transmission. For example, "send information to XX" can be understood as the destination of this information is XX, which can include directly sending through the air interface, and also includes indirectly sending through the air interface by other units or modules. "Receive information from YY" can be understood as the source of this information is YY, which can include directly receiving from YY through the air interface, and can also include indirectly receiving from YY through the air interface by other units or modules. "Send" can also be understood as the "output" of the chip interface, and "receive" can also be understood as the "input" of the chip interface.

[0077] In other words, the sending and receiving can be carried out between devices, for example, between a network device and a terminal device, or can be carried out within a device, for example, sending or receiving between components, modules, chips, software modules or hardware modules within a device through a bus, trace or interface.

[0078] It can be understood that the information may be subjected to necessary processing, such as encoding, modulation, etc. between the source end and the destination end of the information sending, but the destination end can understand the valid information from the source end. Similar expressions in this application can be understood similarly and will not be elaborated here.

[0079] (6) In the embodiments of this application, "indication" may include direct indication and indirect indication, and may also include explicit indication and implicit indication. Regarding the information indicated by a certain piece of information (such as the indication information described below) as the information to be indicated, in the specific implementation process, there are many ways to indicate the information to be indicated. For example, but not limited to, the information to be indicated can be directly indicated, such as the information to be indicated itself or the index of the information to be indicated, etc. It is also possible to indirectly indicate the information to be indicated by indicating other information, where there is an association relationship between the other information and the information to be indicated; it is also possible to only indicate a part of the information to be indicated, while the other parts of the information to be indicated are known or pre-agreed. For example, the arrangement order of each piece of information pre-agreed (such as protocol predefined) can be used to implement the indication of specific information, thereby reducing the indication overhead to a certain extent. This application does not limit the specific manner of indication. It can be understood that for the sender of the indication information, the indication information can be used to indicate the information to be indicated, and for the receiver of the indication information, the indication information can be used to determine the information to be indicated.

[0080] In this application, unless otherwise specified, the same or similar parts between various embodiments can be referred to each other. In various embodiments of this application, as well as in each method / design / implementation manner in each embodiment, if there is no special description and logical conflict, the terms and / or descriptions between different embodiments, as well as between each method / design / implementation manner in each embodiment, are consistent and can be mutually referred to. The technical features in different embodiments, as well as in each method / design / implementation manner in each embodiment, can be combined according to their internal logical relationships to form new embodiments, methods, or implementation manners. The embodiments of this application described below do not constitute a limitation on the protection scope of this application.

[0081] This application can be applied to a long term evolution (LTE) system, a new radio (NR) system, or a communication system evolved after 5G (such as 6G, etc.). Among them, the communication system includes at least one network device and / or at least one terminal device.

[0082] Please refer to Figure 1a , which is a schematic diagram of a communication system provided by this application. Figure 1a In it, a network device and two terminal devices are exemplarily shown. Figure 1a Taking the communication system as a cellular communication system and the network device as a base station as an example. The network device can communicate with each terminal device through wireless signals. The network device can send downlink signals to the terminal device, and the terminal device can send uplink signals to the network device. The terminal devices can also receive and transmit wireless signals through a sidelink (SL).

[0083] Please refer to Figure 1b , which is a schematic diagram of another communication system provided by this application. Figure 1b In it, a network device and one terminal device are exemplarily shown. Figure 1b Taking the communication system as a wireless local area network (WLAN) system and the network device as a wireless access point (AP) as an example. The network device can communicate with each terminal device through wireless signals. The network device can send downlink signals to the terminal device, and the terminal device can send uplink signals to the network device.

[0084] In a wireless communication system (such as Figure 1a or Figure 1b the communication system shown), as a communication node that is a signal sender, the original data to be sent can go through multiple processing processes, including channel coding, modulation, etc.; correspondingly, as a communication node that is a signal receiver, the received signal can go through other processing processes corresponding to these multiple processing processes, including channel decoding, demodulation, etc., to recover the original data (or obtain an estimate of the original data). These processing processes can improve the reliability of data transmission.

[0085] After more than half a century of development, artificial intelligent (AI) technology has now fully entered the stage of industrial development. AI technology has penetrated into all fields and industries, including the wireless communication field. Specifically, in a wireless communication system, an AI model can be introduced into the transmitter and receiver of a device (network device or terminal device). For example, the transmitter of the device can perform processing such as channel coding and modulation on the data to be sent based on the AI model, and the receiver of the device can perform processing such as channel decoding and demodulation on the received signal based on the AI model, thereby improving system flexibility, spectrum efficiency, and system stability.

[0086] To improve the end-to-end performance of both communication parties, the AI models in the transmitters and receivers of both communication parties can be jointly optimized, that is, the transmitter / receiver trained by one party adapts to the receiver / transmitter trained by the other party. The functions of the transmitter and receiver are complex. For the data to be transmitted, the transmitter needs to perform processing such as encoding, rate matching, scrambling, modulation, layer mapping, precoding, resource element (RE) mapping, digital beamforming (BF), waveform shaping, digital-to-analog conversion, and analog BF. For the received wireless signal, the receiver needs to perform processing such as analog BF, analog-to-digital conversion, waveform reception, digital BF, de-RE mapping, channel equalization, de-layer mapping, demodulation, descrambling, de-rate matching, and decoding. Therefore, the AI models in the transmitter and receiver will include multiple AI modules, and each AI module is used to implement a specific function. When jointly optimizing the AI models of both communication parties, multi-module alignment is required, and the implementation of multi-module alignment is complex and costly. Therefore, how to reduce the complexity of multi-module alignment, reduce the overhead of multi-module alignment, and improve the efficiency of multi-module alignment has become an urgent problem to be solved.

[0087] To solve the above problems, the present application provides the following embodiments to reduce the complexity of multi-module alignment, reduce the overhead, and improve the alignment efficiency. Generally speaking, the present application verifies multiple modules in the models of both communication parties and screens out the modules that are not aligned between the two communication parties among the multiple modules. Then, training is performed in units of module groups to align the modules. A module group includes multiple modules, and the multiple modules in the module group include unaligned modules. Thus, there is no need to perform independent alignment for each module separately, which can improve the alignment efficiency and reduce the implementation complexity and overhead.

[0088] In the embodiments of the present application, the model includes multiple cascaded modules for implementing complex signal processing functions in the receiver or transmitter. Multiple cascaded modules mean that the output of one module serves as the input of one or more other modules. Among them, each module in the model is used to implement one or more functions in the receiver or transmitter, such as encoding, modulation, or information compression. At least some of the modules in the model can be AI modules. An AI module is a neural network model used to implement one or more functions in the receiver or transmitter, and it can be optimized through training. In one possible implementation, all the modules in the model are AI modules. In another possible implementation, among the multiple modules of the model, some of the modules can be AI modules and some can be non-AI modules. It should be noted that unless otherwise specified, the "model" in the present application refers to a model including multiple modules, and the "neural network model" corresponds to the modules in the "model".

[0089] As Figure 2 shown, Figure 2The flowchart of a training method provided for this application. This method is executed by a first communication device and a second communication device. The first communication device can be Figure 1a / Figure 1b a network device in / a chip or chip system in the network device, etc., and the second communication device is Figure 1a / Figure 1b a terminal device in / a chip or chip system in the terminal device, etc. Or, the first communication device can be Figure 1a / Figure 1b a terminal device in / a chip or chip system in the terminal device, etc., and the second communication device is Figure 1a / Figure 1b a network device in / a chip or chip system in the network device, etc. This embodiment includes the following steps:

[0090] S201: The first communication device and the second communication device verify whether the modules in the first model and the second model are aligned.

[0091] Among them, the first model is the model of the first communication device, and the second model is the model of the second communication device. The first model can be the model in the transmitter, and the second model is the model in the receiver. The second model is used to perform inverse processing on the data (signal) output by the first model. Or the first model can be the model in the receiver, and the second model is the model in the transmitter. The first model is used to perform inverse processing on the data (signal) output by the second model. Or, both the first model and the second model are the models in the transmitter. Or, both the first model and the second model are the models in the receiver.

[0092] The first model includes multiple modules, and the second model includes multiple modules. The multiple modules in the first model have a corresponding relationship with the multiple modules in the second model. In this embodiment, two modules with a corresponding relationship in the first model and the second model are referred to as a module pair. Optionally, the multiple modules in the first model and the multiple modules in the second model are in one-to-one correspondence. Two modules with a corresponding relationship in the first model and the second model are, for example, two modules for performing inverse operations. For example, a module for modulation in one model has a corresponding relationship with a module for demodulation in another model, a module for information compression in one model has a corresponding relationship with a module for information decompression in another model, a module for encoding in one model has a corresponding relationship with a module for decoding in another model, and so on. Examples are not given one by one here. Or, when both the first model and the second model are models in a transmitter / receiver, two modules with a corresponding relationship in the first model and the second model are, for example, two modules for performing the same operation. For example, a module for modulation in one model has a corresponding relationship with a module for modulation in another model, a module for information compression in one model has a corresponding relationship with a module for information compression in another model, a module for decoding in one model has a corresponding relationship with a module for decoding in another model, and so on. Examples are not given one by one here.

[0093] The first communication device and the second communication device can verify whether each module pair in the first model and the second model is aligned to determine the misaligned module pairs among the multiple module pairs.

[0094] In a possible implementation, the first model is a model in the transmitter of the first communication device, and the second model is a model in the receiver of the second communication device. Then, the first communication device sends verification data, and the second communication device verifies whether each pair of module pairs in the first model and the second model is aligned according to the verification data and feeds back the verification result to the first communication device.

[0095] In another possible implementation, the first model is a model in the receiver of the first communication device, and the second model is a model in the transmitter of the second communication device. Then, the second communication device sends verification data, and the first communication device verifies whether each pair of module pairs in the first model and the second model is aligned according to the verification data and feeds back the verification result to the second communication device.

[0096] In yet another possible implementation, the first model is a model in the transmitter of the first communication device, and the second model is a model in the transmitter of the second communication device. Then, the first communication device can send verification data, and the second communication device verifies according to the verification data. Or, the second communication device can also send verification data, and the first communication device verifies according to the verification data.

[0097] In yet another possible implementation, the first model is a model in the receiver of the first communication device, and the second model is a model in the receiver of the second communication device. Then, the first communication device can send the verification data, and the second communication device verifies based on the verification data. Alternatively, the second communication device can also send the verification data, and the first communication device verifies based on the verification data.

[0098] In one possible implementation, the verification data can include a model or a module. For example, the first model or at least one module in the first model (when the second communication device performs verification), or the second model or at least one module in the second model (when the first communication device performs verification). In another possible implementation, the verification data can include a data set. For example, the input data and / or output data of the first model.

[0099] There are various verification methods for verifying whether the modules in the first model and the second model are aligned. For example, each module pair can be independently verified. Also, for example, verification can be performed in units of module groups. For instance, the number of modules in the verified module group can vary from more to less, or from less to more, etc. The following describes these verification methods separately:

[0100] Verification method 1: Independently verify each module pair.

[0101] Here, the modules in the verified module pair are all AI modules. Taking the example where the first communication device sends the verification data to the second communication device, and the second communication device verifies based on the verification data. The first communication device sends the verification data corresponding to each module to the second communication device. The verification data corresponding to a module can include the input data and output data of that module. The second communication device verifies whether the module pair is aligned based on the verification data corresponding to each module and the corresponding module in the second model. When the first model is a model in the transmitter and the second model is a model in the receiver, the second communication device can use the output data of module a (a certain module in the first model) as the input data of the corresponding module b in the second model to obtain the output data of module b, and then calculate the loss function value based on the input data of module a and the output data of module b. If the loss function value is greater than the threshold, it can be confirmed that module a and module b are not aligned; otherwise, it can be determined that module a and module b are aligned. When both the first model and the second model are models in the transmitter or the receiver, the second communication device can use the input data of module a as the input data of the corresponding module b in the second model to obtain the output data of module b, and then calculate the loss function value based on the output data of module a and the output data of module b. If the loss function value is greater than the threshold, it can be confirmed that module a and module b are not aligned; otherwise, it can be determined that module a and module b are aligned.

[0102] The process in which the second communication device sends verification data to the first communication device and the first communication device performs verification based on the verification data is similar to the process in which the first communication device sends verification data to the second communication device and the second communication device performs verification based on the verification data. Therefore, it will not be elaborated here.

[0103] Verification method two: perform verification based on a module group, and the number of modules in the module group decreases from more to less.

[0104] Taking the first communication device sending verification data to the second communication device and the second communication device performing verification based on the verification data as an example. The first communication device sends the verification data corresponding to module group A1 in the first model to the second communication device. Optionally, module group A1 may include all the modules in the first model. Of course, module group A1 may also include some modules in the first model. When module group A1 in the first model is not aligned with the corresponding module group A2 in the second model, the first communication device sends the verification data corresponding to module group B1 in the first model to the second communication device, and module group B1 is a subset of module group A1. When module group B1 is not aligned with the corresponding module group B2 in the second model, the first communication device sends the verification data corresponding to module group C1 in the first model to the second communication device, and module group C1 is a subset of module group B1. And so on, until the module group in the first model is aligned with the corresponding module group in the second model, or the number of modules in the module group is 1. There are one or more module pairs included in two module groups with a corresponding relationship in the first model and the second model. That is, the modules in the module group in the first model have a corresponding relationship with the modules in the corresponding module group in the second model.

[0105] The module group may include one or a cascade of multiple modules. When the module group in the previous verification by the first communication device is not aligned, the first communication device reduces the scale of the module group for the next verification to screen out the modules in the first model and the second model that are not aligned. When the first model is the model in the transmitter, the first communication device can contract the module group towards the output side of the first model, that is, each module group for verification includes the modules on the output side of the first model until the module group to be verified is aligned with the corresponding module group in the second model, or the module group only includes one module on the output side of the first model and the module group cannot be further contracted. When the first model is the model in the receiver, the first communication device can contract the module group towards the input side of the first model, that is, each module group for verification includes the modules on the input side of the first model until the module group to be verified is aligned with the corresponding module group in the second model, or the module group only includes one module on the input side of the first model and the module group cannot be further contracted.

[0106] Exemplarily, as Figure 3 shown, Figure 3 is a schematic diagram of a model verification method provided by the present application. Figure 3In this case, take the first model as the model in the transmitter and the second model as the model in the receiver. The first model includes Module 1, Module 2, and Module 3, and the second model includes Module 4, Module 5, and Module 6. Among them, Module 1 and Module 6 form a pair of modules, Module 2 and Module 5 form a pair of modules, and Module 3 and Module 4 form a pair of modules. During the first verification, the first communication device sends the verification data of Module Group A1 to the second communication device. Module Group A1 includes Module 1, Module 2, and Module 3. The verification data of Module Group A1 includes, for example, the input data and output data of Module Group A1. The second communication device verifies Module Group A2 in the second model according to the verification data of Module Group A1. Module Group A2 includes Module 4, Module 5, and Module 6. If Module Group A1 and Module Group A2 are not aligned, the second communication device sends a verification result indicating that Module Group A1 and Module Group A2 are not aligned to the first communication device. The first communication device sends the verification data of Module Group B1 to the second communication device. Module Group B1 includes Module 2 and Module 3. The verification data of Module Group B1 includes, for example, the input data and output data of Module Group B1. The second communication device verifies Module Group B2 in the second model according to the verification data of Module Group B1. Module Group B2 includes Module 4 and Module 5. If Module Group B1 and Module Group B2 are not aligned, the second communication device sends a verification result indicating that Module Group B1 and Module Group B2 are not aligned to the first communication device. The first communication device sends the verification data of Module Group C1 to the second communication device. Module Group C1 includes Module 3. The verification data of Module Group C1 includes, for example, the input data and output data of Module Group C1. The second communication device verifies Module Group C2 in the second model according to the verification data of Module Group C1. Module Group C2 includes Module 4. If Module Group C1 and Module Group C2 are not aligned, the second communication device sends a verification result indicating that Module Group C1 and Module Group C2 are not aligned to the first communication device. The method of verifying whether module groups are aligned based on the verification data corresponding to the module groups is similar to the method of verifying whether modules are aligned based on the verification data corresponding to the modules, so it will not be elaborated here.

[0107] Optionally, when the module groups are aligned, the verification can be stopped. For example, if Module Group A1 and Module Group A2 are not aligned, and when Module Group B1 and Module Group B2 are aligned, the first communication device can stop the verification, thereby reducing the verification overhead and improving the verification efficiency.

[0108] According to the verification result, the first communication device and the second communication device can determine the misaligned module pairs. That is, the first communication device can determine the misaligned modules in the first model (in this embodiment, the module in the first model that is not aligned with the corresponding module in the second model is called the target module), and the second communication device can determine the misaligned modules in the second model. Exemplarily, if module group A1 is not aligned with module group A2, but module group B1 is aligned with module group B2, the misaligned module in the first model is determined to be module 1. For another example, if module group A1 is not aligned with module group A2, module group B1 is not aligned with module group B2, but module group C1 is aligned with module group C2, the misaligned modules in the first model include module 2, or the misaligned modules in the first model include module 1 and module 2.

[0109] The process in which the second communication device sends verification data to the first communication device and the first communication device performs verification based on the verification data is similar to the process in which the second communication device performs verification based on the verification data of the first communication device, so it will not be elaborated here.

[0110] Through the verification from the whole to the modules, the size of the module group is reduced when the module groups are not aligned, and the verification can be stopped when the module groups are aligned, thereby improving the verification efficiency.

[0111] Verification method three: perform verification based on module groups, and the number of modules in the module groups increases from less to more.

[0112] Taking the example where the first communication device sends verification data to the second communication device and the second communication device performs verification based on the verification data. The first communication device sends the verification data corresponding to module group E1 in the first model to the second communication device. Module group E1 includes, for example, one module in the first model. When module group E1 is aligned with the corresponding module group E2 in the second model, the first communication device sends the verification data corresponding to module group F1 in the first module group to the second communication device, and module group E1 is a subset of module group F1. When module group F1 is aligned with module group F2 in the second model, the first communication device sends the verification data corresponding to module group G1 in the first module group to the second communication device, and module group F1 is a subset of module group G1. And so on, until the module group in the first module group is not aligned with the corresponding module group in the second model, or the module group includes all the modules in the first model. There are one or more module pairs in the two module groups with a corresponding relationship in the first model and the second model. That is, the modules in the module group in the first model have a corresponding relationship with the modules in the corresponding module group in the second model.

[0113] The module group may include one or multiple cascaded modules. When the module group alignment of the previous verification is completed, the first communication device enlarges the scale of the module group for the next verification to screen the misaligned modules in the first model. When the first model is the model in the transmitter, the first communication device may enlarge the module group from the output side of the first model to the input side of the first model. That is, the module group for each verification includes the modules on the output side of the first model until the verified module group is misaligned with the corresponding module group in the second model, or the module group includes all the modules in the first model and cannot be further enlarged. When the first model is the model in the receiver, the first communication device may enlarge the module group from the input side of the first model to the output side of the first model. That is, the module group for each verification includes the modules on the input side of the first model until the verified module group is misaligned with the corresponding module group in the second model, or the module group includes all the modules in the first model and cannot be further enlarged.

[0114] Exemplarily, as Figure 3 shown, at the first verification, the first communication device sends the verification data of module group E1 to the second communication device. Module group E1 includes module 3. The verification data of module group E1 may include, for example, the input data and output data of module group E1. The second communication device verifies the module group E2 in the second model according to the verification data of module group E1. Module group E2 includes module 4. If module group E1 is aligned with module group E2, the second communication device sends a verification result indicating the alignment of module group E1 and module group E2 to the first communication device. The first communication device sends the verification data of module group F1 to the second communication device. Module group F1 includes module 2 and module 3. The verification data of module group F1 may include, for example, the input data and output data of module group F1. The second communication device verifies the module group F2 in the second model according to the verification data of module group F1. Module group F2 includes module 4 and module 5. If module group F1 is aligned with module group F2, the second communication device sends a verification result indicating the alignment of module group F1 and module group F2 to the first communication device. The first communication device sends the verification data of module group G1 to the second communication device. Module group G1 includes module 1, module 2, and module 3. The verification data of module group G1 may include, for example, the input data and output data of module group G1. The second communication device verifies the module group G2 in the second model according to the verification data of module group G1. Module group G2 includes module 4, module 5, and module 6. If module group G1 is aligned with module group G2, the second communication device sends a verification result indicating the alignment of module group G1 and module group G2 to the first communication device. The method for verifying whether the module groups are aligned based on the verification data corresponding to the module groups is similar to the method for verifying whether the modules are aligned based on the verification data corresponding to the modules, so it will not be elaborated here.

[0115] Optionally, when performing verification based on the third verification method, if the module groups are not aligned, the first communication device may stop the verification, thereby reducing the verification overhead and improving the verification efficiency.

[0116] According to the verification result, the first communication device and the second communication device may determine the misaligned module pairs. That is, the first communication device may determine the misaligned modules in the first model, and the second communication device may determine the misaligned modules in the second model. Exemplarily, if module group E1 is not aligned with module group E2, it can be determined that module 3 is not aligned with module 4. Also, for example, if module group E1 is aligned with module group E2 and module group F1 is not aligned with module group F2, it can be determined that the misaligned modules in the first model include module 2, or the misaligned modules in the first model include module 1 and module 2.

[0117] Verification method four: Perform verification based on module groups, and the number of modules in the module group is fixed.

[0118] For example, multiple modules in the first model and the second model may be divided into multiple module groups, and the number of modules in each module group is K, where K is an integer greater than or equal to 2. The multiple modules in the module group are cascaded, so that the module group can be verified as a whole. Optionally, the modules in each verified module group do not overlap, so as to improve the model verification efficiency. It can be understood that when the number of modules in the first model cannot be divided evenly by K, the number of modules in a module group may be less than K.

[0119] Exemplarily, as Figure 4 shown, Figure 4 is a schematic diagram of another model verification method provided by this application. Figure 4In this case, the first model is the model in the transmitter, and the second model is the model in the receiver. Taking K = 2 as an example, the first model includes module 7, module 8, module 9, and module 10, and the second model includes module 11, module 12, module 13, and module 14. Among them, module 7 and module 14 form a pair of modules, module 8 and module 13 form a pair of modules, module 9 and module 12 form a pair of modules, and module 10 and module 11 form a pair of modules. The first model includes module group H1 (including module 7 and module 8) and I1 (including module 9 and module 10), and the second model includes module group H2 (including module 11 and module 12) and I2 (including module 13 and module 14). Module group H1 corresponds to module group H2, and module group I1 corresponds to module group I2. The first communication device can send the verification data corresponding to module group H1 to the second communication device, and the second communication device verifies whether module group H1 and module group H2 are aligned according to the verification data corresponding to module group H1 and module group H2. The first communication device can send the verification data corresponding to module group I1 to the second communication device, and the second communication device verifies whether module group I1 and module group I2 are aligned according to the verification data corresponding to module group I1 and module group I2. The verification data corresponding to module group H1 and the verification data corresponding to module group I1 can be sent through the same message or through different messages, and there is no restriction here.

[0120] Optionally, if the module group is not aligned, it can be regarded that all the modules in the module group are not aligned. Similarly, if the module group is aligned, it can be regarded that the modules in the module group are aligned.

[0121] Of course, at least two of the above verification method 1 and verification method 4 can be combined. For example, when combining verification method 1 and verification method 4, the unaligned module groups can be screened first through verification method 4, and then the unaligned modules among them can be determined through verification method 1. Another example is the combination of verification method 2 and verification method 3. For example, multiple modules in the first model can be divided into two parts, and the modules in each part are cascaded. One part is verified through verification method 2, and the other part is verified through verification method 3. Another example is the combination of verification method 1 and verification method 3. When there are still remaining modules that have not been verified when the module group is not aligned, the remaining unverified modules can be verified through verification method 1.

[0122] Through the above verification method, the target modules that are not aligned in the first model can be determined, and then module alignment can be performed based on the target modules.

[0123] S202: The second communication device sends training data. Correspondingly, the first communication device receives the training data. The training data is used to train the first module group, and the first module group includes the unaligned target modules.

[0124] Among them, the training data is used to train the first module group, and the training data is obtained based on the second module group in the second model. The training data may include a training sample set constructed based on the second module group. For example, the training sample set includes the input data and output data of the second module group. Alternatively, the training data may include gradients. Alternatively, the training data may include the second module group. Alternatively, the training data may include the second model.

[0125] The first module group includes all the modules in the first model or some cascaded modules. The second module group includes all the modules in the second model or some cascaded modules. The first module group includes modules corresponding to the modules in the second module group. The first module group includes a target module, and the target module is a module in the first model that is not aligned with the corresponding module in the second module group. When there are multiple target modules in the first model, the first module group may include at least one target module in the first model. The first module group and the second module group are used to implement the same function. Alternatively, the first module group and the second module group are used to implement inverse functions.

[0126] In a possible implementation, the modules in the first module group correspond one-to-one with the modules in the second module group.

[0127] In another possible implementation, the first module group may further include at least one adaptation module. The adaptation module is inserted into the first model after determining the target module in the first model. The adaptation module is used to align the first module group and the second module group. How to align the first module group and the second module group through the adaptation module will be described in S203. It should be noted that the first module group is trained and adjusted based on the second module group, so the second module group may not include the adaptation module or the module corresponding to the adaptation module. That is, the modules in the first module group other than the adaptation module correspond one-to-one with the modules in the second module group.

[0128] The first communication device may determine the first module group after determining the unaligned target module, and then request the training data corresponding to the second module group from the second communication device. Each module in the first model and the second model may have a corresponding module identifier. In a possible implementation, the module identifiers of the two modules in the module pair may be the same. The module identifiers of the modules in the first model may be determined according to the processing order of the module pair data, and the module identifiers of the modules in the second model may be determined according to the processing order of the module pair data. Figure 3Taking the first model and the second model as examples, the module identifiers of module 1 and module 6 can be the same, the module identifiers of module 2 and module 5 can be the same, and the module identifiers of module 3 and module 4 can be the same. In another possible implementation, the module identifiers of the two modules in the module pair can be different, and the first communication device can store the mapping relationship between each module in the first model and each module in the second model. In the request for training data sent by the first communication device to the second communication device, the module identifiers of the modules in the first module group (excluding the identifier of the adaptation module) can be carried, or the module identifiers of the modules in the second module group can be carried. Thus, the second communication device can send the training data for training the first module group to the first communication device.

[0129] Alternatively, it can also be that after the second communication device determines the misaligned target module, it determines the second module group, sends the training data corresponding to the second module group to the first communication device, and indicates to the first communication device the modules included in the second module group. Thus, the first communication device can determine the first module group and use the training data to train the first module group.

[0130] S203: The first communication device trains the first module group according to the training data.

[0131] When validating the first model and the second model and determining the misaligned target module in the first model, the first model or the module group containing the target module can be trained to align the first model with the second model, or the target module with the corresponding module in the second model.

[0132] In this embodiment, there are multiple training methods. Classified by the optimization objective, training can be performed with the single-module performance as the objective, or training can be performed with the multi-module performance as the objective. Classified by the parameter adjustment object, the parameters of the target module can be adjusted, and the parameters of the adaptation module can also be adjusted. The adaptation module is different from the target module. The adaptation module is inserted into the first model after the target module is confirmed and is used for adaptation and alignment. The above training methods will be described in detail below.

[0133] Training method 1: With the optimization of single-module performance as the objective, adjust the parameters of the adaptation module.

[0134] In this training method, the first module group includes a target module and an adaptation module. Correspondingly, the second module group includes a module corresponding to the target module. When training based on this training method, the parameters of the target module are frozen, and the parameters of the adaptation module are adjusted to make the performance of the combination of the target module and the adaptation module (the first module group) meet the performance requirements. It is possible to determine whether the first module group and the second module group are aligned according to the loss function value between the first module group and the second module group. When the first module group and the second module group are aligned, the training can be stopped. During inference, the target module and the adaptation module participate in the inference as a whole.

[0135] The target module is adjacent to the adaptation module, and the target module and the adaptation module are cascaded. The adaptation module can be set on the input side of the target module. During training, the training data is input into the adaptation module, and after being processed by the adaptation module, it is input into the target module. The loss function value is calculated using the output data of the target module. The adaptation module can also be located on the output side of the target module. During training, the training data is input into the target module, and after being processed by the target module, it is input into the adaptation module. The loss function value is calculated using the output data of the adaptation module.

[0136] Since the target module is used to implement one or more functions in the receiver or transmitter, and the adaptation module is used to achieve the alignment between the first module group and the second module group, the parameters in the adaptation module can be fewer than those in the target module, which can reduce the training data required in the training process, reduce the training complexity, and improve the training efficiency. In addition, the adaptation module can adjust the parameters according to different second models or scenarios / tasks / training data, etc., while the parameters of the target module remain unchanged, which can ensure the generality of the target module. When switching the second model / scenario / task / training data, only the adaptation module needs to be switched, instead of training the corresponding target module separately for each second model / scenario / task / training data.

[0137] Exemplarily, as Figure 5 shown, when the first model includes multiple target modules, a corresponding adaptation module is set on the input side or the output side of each module. Each first module group includes an adaptation module and a target module, and each first module group is trained separately to make the first module group to which each target module belongs aligned with the corresponding second module group.

[0138] Training method two aims to optimize the performance of multiple modules and adjusts the parameters of the target module.

[0139] The first module group can include all the modules in the first model. Or, the first module group can include some of the modules in the first model. The first module group includes multiple modules in the first model. Correspondingly, the second module group includes multiple modules corresponding to the multiple modules in the first module group.

[0140] When the first model includes multiple target modules, multiple target modules can be trained jointly. In this case, the first module group can include multiple target modules. Exemplarily, taking the first model in Figure 3 as an example, if module 1 and module 2 are target modules and module 3 is an aligned module, the first module group can include module 1 and module 2.

[0141] Optionally, the first module group can also include aligned modules (simply referred to as aligned modules in this article). For example, since multiple modules in the first model are connected in a cascaded manner, when one or more aligned modules are included between two target modules, the first module group can include these aligned modules. Or, aiming to optimize the performance of the first model, the first module group includes all modules in the first model, and the first module group can include aligned modules. During training, the aligned modules in the first module group participate in data processing, but the parameters of the aligned modules will not be adjusted, that is, the parameters of the aligned modules are frozen during the training process. Exemplarily, taking the first model in Figure 3 as an example, if module 1 and module 3 are target modules and module 2 is an aligned module, the first module group can include module 1, module 2, and module 3. Or, if module 1 and module 2 are target modules and module 3 is an aligned module, when aiming to optimize the performance of the first model, the first module group can include module 1, module 2, and module 3.

[0142] In this training method, the first module group includes multiple target modules. Optimizing the target modules in the first module group with the performance of the first module group as the goal can optimize multiple target modules simultaneously, reduce the module alignment overhead, and improve the alignment efficiency.

[0143] Training method three: Aiming to optimize the performance of multiple modules, adjust the parameters of the adaptation modules.

[0144] The first module group can include all modules in the first model. Or, the first module group can include some modules in the first model. When the first model includes multiple target modules, multiple target modules can be trained jointly. That is, the first module group can include multiple target modules. Correspondingly, the second module group includes multiple modules corresponding to other modules except the adaptation modules in the first module group.

[0145] The difference from training method two is that in addition to including multiple target modules and aligned modules (if any), the first module group also includes adaptation modules. Exemplarily, taking the first model in Figure 3 as an example, if module 1 and module 2 are target modules and module 3 is an aligned module, the first module group can include module 1, module 2, and at least one adaptation module.

[0146] In one implementation, each target module in the first module group corresponds to an adaptation module. For example, the first module group includes N target modules and N adaptation modules, and the N target modules and the N adaptation modules are in one-to-one correspondence. In Training Method 3, N is an integer greater than or equal to 2. Each adaptation module can be arranged adjacent to the corresponding target module.

[0147] In another implementation, the number of adaptation modules can be 1, which can reduce the training complexity. For example, the first module group includes M target modules and 1 adaptation module. M is an integer greater than or equal to 2. The adaptation module can be arranged on the input side of the first module group, or on the output side of the first module group, or between any two modules in the first module group, and there is no limitation here.

[0148] When training with training data, the parameters of the adaptation module can be adjusted, and the parameters of the target module and the aligned module (if any) can be frozen. When the loss function value of the first module group and the second module group is less than the threshold, the training can be completed. When inferring using the first model, the adaptation module participates in the inference. By inserting target modules into the first module group and adjusting the parameters of the adaptation module with the performance of the first module group as the goal during training, without adjusting the parameters of each target module, the training complexity can be reduced and the training efficiency can be improved. Moreover, by jointly training multiple target modules, the interaction times between the first communication device and the second communication device during training can be reduced, and the air interface overhead during training can be reduced.

[0149] In this embodiment, the training data is the training data obtained based on the second module group corresponding to the first module group in the second model. The training data includes, for example, the input data and output data of the second module group. If the first model and the second model are models for implementing inverse functions, when training the first module group, the output data of the second module group can be used as the input data of the first module group to obtain the output data of the first module group, and then the loss function value can be calculated according to the input data of the second module group and the output data of the first module group, and based on this loss function value, the parameters of the target module can be adjusted using the gradient descent method. If the first model and the second model are models for implementing the same function, when training the first module group, the input data of the second module group is used as the input data of the first module group to obtain the output data of the first module group, and then the loss function value is calculated according to the output data of the second module group and the output data of the first module group, and based on this loss function value, the parameters of the target module are adjusted using the gradient descent method.

[0150] Alternatively, the training data includes, for example, gradients obtained based on the second module group. Specifically, the gradients can be obtained based on the input data of the first module group, the output data of the first module group, and the output data of the second module group. The first communication device adjusts the parameters of the target module according to the gradients. Alternatively, the training data includes, for example, the second module group or the second model. Thus, the first communication device can construct a data set for training based on the second module group and train the first module group based on the data set.

[0151] In this embodiment, Training Method 1, Training Method 2, and Training Method 3 can be adaptively switched. For example, when it is difficult for the first module group to converge during training using Training Method 3, the training can be switched to Training Method 1 or Training Method 2.

[0152] When the loss function value between the first module group and the corresponding second module group is less than the preset threshold, it can be confirmed that the first module group is aligned with the second module group. The first communication device can associate the first module group and the second module group. The first communication device can configure a first identifier for the first module group, configure a second identifier for the second module group, and send the second identifier to the second communication module. The first communication device associates the first identifier and the second identifier to associate the first module group and the second module group. Alternatively, the first communication device can determine a third identifier that associates the first module group and the second module group to associate the first module group and the second module group through the third identifier. The first identifier is associated with the second identifier, or the third identifier is associated with the first model and the second model. This association relationship can be used for model selection / switching / activation / monitoring, etc. in subsequent processes.

[0153] Through Training Method 3 provided in this embodiment, when training for different second models or training data, the parameters of only the adaptation module can be adjusted with the end-to-end performance of the first model and the second model as the goal, so that the first model adapts to different second models / training data. Different second models or training data are applicable to different scenarios or tasks.

[0154] In a possible implementation manner, when training for different second models or training data, the structure of the adaptation module remains unchanged. The parameters of the adaptation module are adjusted for each second model or the second model / training data corresponding to the training data to obtain the parameters of the adaptation module corresponding to different second models or training data. The parameters corresponding to different second models or training data are associated with the second model or the scenario / task. When the first model is used for inference subsequently, according to the current second model or scenario / task, the associated parameters are loaded into the adaptation module for inference. Exemplarily, Figure 6As shown, parameter p1 corresponds to scenario / task 1, parameter p2 corresponds to scenario / task 2, and parameter p3 corresponds to scenario / task 3. When task 1 is executed, parameter p1 is loaded into the adaptation module for inference. When task 2 is executed, parameter p2 is loaded into the adaptation module for inference. When task 3 is executed, parameter p3 is loaded into the adaptation module for inference.

[0155] In another possible implementation, the adaptation module includes multiple sub-modules. When training for different second models or scenarios / tasks, the sub-modules in the adaptation module that participate in parameter adjustment can be changed so that different second models or scenarios / tasks correspond to different sub-modules in the adaptation module. When the second model or scenario / task is relatively complex, more sub-modules and more parameters can be selected to participate in the training to ensure the performance of the first model when applied to the second model / scenario / task. When inferring for different second models or scenarios / tasks, the corresponding sub-modules in the target module can be selected for activation. Exemplarily, as Figure 7 shown, the target module includes, for example, 3 sub-modules w1, w2, and w3. Task 1 corresponds to sub-module w3, task 2 corresponds to sub-modules w1 and w2, and task three corresponds to sub-modules w1, w2, and w3. When the first model is used to execute task 1, only sub-module w3 can be activated, while sub-modules w1 and w2 are not activated. When the first model is used to execute task 2, sub-modules w3 and w2 can be activated, while sub-module w1 is not activated. When the first model is used to execute task 3, sub-modules w1, w3, and w2 can be activated.

[0156] Figure 7 The adaptation module shown in can also be applied to training scenarios with different complexities. For example, when there are fewer target modules in the first module group, fewer sub-modules among them can be used for training. When the number of target modules in the first module group is larger, more sub-modules among them can be used for training. Exemplarily, taking Figure 7 as an example, if the first module group includes module j and the adaptation module, only sub-module w3 of the adaptation module can be trained. If the first module group includes modules i, j, and the adaptation module, sub-modules w2 and w3 of the adaptation module can be trained. If the first module group includes modules h, i, j, and the adaptation module, sub-modules w1, w2, and w3 of the adaptation module can be trained.

[0157] It can be understood that Figures 3 - 7 the model in is only for illustration and should not be construed as a limitation to this application. For example, Figures 3 - 7Taking the first model as the model in the transmitter and the second model as the model in the receiver as an example. Of course, the first model can also be the model in the receiver, and the second model can also be the model in the transmitter, or both the first model and the second model are the models in the transmitter, or both are the models in the receiver, which is not limited here. For another example, the number of modules in the first model and the second model is only used as an example, and the number of modules in the first model and the second model can be more or less, and this application does not limit this. For another example, the position of the adaptation module in the first model / first module group is only for illustration. The adaptation module can be set on the input side of the target module / first module group / first model, or can be set on the output side of the target module / first module group / first model, which is not limited here.

[0158] In this embodiment, by verifying the modules in the first model and the second model, the target modules that are not aligned in the first model and the second model are determined, and then the first module group is determined based on the performance optimization target and the unaligned target modules. The first module group can include multiple target modules. With the goal of optimizing the performance of the first module group, the optimization of multiple target modules / adaptation modules can be completed simultaneously, thereby improving the alignment efficiency of the modules in the first model and the second model and reducing the overhead of aligning the first model.

[0159] Please refer to Figure 8 , an embodiment of the present application provides a communication device 800. The communication device 800 can implement the functions of the second communication device or the first communication device in the above method embodiment, and thus can also achieve the beneficial effects possessed by the above method embodiment. In the embodiment of the present application, the communication device 800 can be the first communication device (or the second communication device), or an integrated circuit or component inside the second communication device (or the first communication device), such as a chip.

[0160] It should be noted that the transceiver unit 802 can include a sending unit and a receiving unit, which are respectively used to perform sending and receiving.

[0161] In a possible implementation manner, when the device 800 is used to execute the method performed by the first communication device in the foregoing embodiment, the device 800 includes a processing unit 801 and a transceiver unit 802; the transceiver unit 802 is used to receive training data, and the training data is used to train the first module group. The training data is obtained based on the second module group in the second model. The first module group includes all the modules in the first model or cascaded partial modules. The second module group includes the modules in the second model corresponding to the modules in the first module group. The first module group includes target modules, and the target modules are the modules in the first model that are not aligned with the corresponding modules in the second module group. The processing unit 801 is used to train the first module group according to the training data.

[0162] In a possible implementation, the first module group includes an adaptation module, and the adaptation module and the target module are different modules. The processing unit 801 is configured to adjust the parameters in the adaptation module according to the training data.

[0163] In a possible implementation, the adaptation module is adjacent to the target module; or the adaptation module is disposed on the output side or the input side of the first module group.

[0164] In a possible implementation, the first module group includes N target modules and N adaptation modules, the N target modules and the N adaptation modules are in one-to-one correspondence, and each adaptation module is adjacent to the corresponding target module; N is an integer greater than or equal to 1.

[0165] In a possible implementation, the first module group includes M target modules and 1 adaptation module; M is an integer greater than or equal to 2.

[0166] In a possible implementation, the processing unit 801 is configured to adjust the parameters in the target module according to the training data.

[0167] In a possible implementation, the training data includes at least one of the following: the input data of the second module group and the output data of the second module group; the gradient obtained based on the input data of the first module group, the output data of the first module group, and the output data of the second module group; the second module group; the second model.

[0168] In a possible implementation, the transceiver unit 802 is configured to send or receive verification data, and the verification data is used to determine the target module.

[0169] In a possible implementation, the verification data includes first verification data, and the first verification data is used to verify whether the third module group and the fourth module group are aligned. The third module group includes all the modules or some cascaded modules in the first model, and the fourth module group includes the modules corresponding to the modules in the third module group in the second model.

[0170] In a possible implementation, the verification data includes second verification data, and the second verification data is used to verify whether the fifth module group and the sixth module group are aligned. The fifth module group is a subset of the third module group, and the sixth module group is a subset of the fourth module group.

[0171] In a possible implementation, the verification data includes third verification data, and the third verification data is used to verify whether the seventh module group and the eighth module group are aligned. The third module group is a subset of the seventh module group, and the fourth module group is a subset of the eighth module group.

[0172] In a possible implementation, when the device 800 is used to execute the method performed by the second communication device in the foregoing embodiment, the device 800 includes a processing unit 801 and a transceiver unit 802. The processing unit 801 is used to determine training data. The transceiver unit 802 is used to send the training data. The training data is used to train the first module group, and the training data is obtained based on the second module group in the second model. The first module group includes all modules or cascaded partial modules in the first model. The second module group includes modules corresponding to the modules in the first module group in the second model. The first module group includes a target module, and the target module is a module in the first model that is not aligned with the corresponding module in the second module group.

[0173] In a possible implementation, the training data includes at least one of the following: input data of the second module group and output data of the second module group; gradients obtained based on the input data of the first module group, the output data of the first module group, and the output data of the second module group; the second module group; the second model.

[0174] In a possible implementation, the transceiver unit 802 is used to send or receive verification data, and the verification data is used to determine the target module.

[0175] In a possible implementation, the verification data includes first verification data, and the first verification data is used to verify whether the third module group is aligned with the fourth module group. The third module group includes all modules or cascaded partial modules in the first model. The fourth module group includes modules corresponding to the modules in the third module group in the second model.

[0176] In a possible implementation, the verification data includes second verification data, and the second verification data is used to verify whether the fifth module group is aligned with the sixth module group. The fifth module group is a subset of the third module group, and the sixth module group is a subset of the fourth module group.

[0177] In a possible implementation, the verification data includes third verification data, and the third verification data is used to verify whether the seventh module group is aligned with the eighth module group. The third module group is a subset of the seventh module group, and the fourth module group is a subset of the eighth module group.

[0178] In a possible implementation, the first module group includes an adaptation module, and the adaptation module and the target module are different modules.

[0179] In a possible implementation, the adaptation module is adjacent to the target module; or the adaptation module is disposed on the output side or the input side of the first module group.

[0180] In a possible implementation, the first module group includes N target modules and N adaptation modules. The N target modules correspond to the N adaptation modules one by one, and each adaptation module is adjacent to the corresponding target module; N is an integer greater than or equal to 1.

[0181] In a possible implementation, the first module group includes M target modules and 1 adaptation module; M is an integer greater than or equal to 2.

[0182] It should be noted that for the information execution process and other contents of the units of the communication device 800 above, please refer to the description in the method embodiments shown in the foregoing of this application for details, which will not be elaborated here.

[0183] Please refer to Figure 9 , which is another schematic structural diagram of the communication device 900 provided by this application. The communication device 900 includes a logic circuit 901 and an input / output interface 902. Among them, the communication device 900 can be a chip or an integrated circuit.

[0184] Among them, Figure 8 the shown transceiver unit 802 can be a communication interface, and this communication interface can be Figure 9 the input / output interface 902 in , and this input / output interface 902 can include an input interface and an output interface. Alternatively, this communication interface can also be a transceiver circuit, and this transceiver circuit can include an input interface circuit and an output interface circuit.

[0185] Optionally, the logic circuit 901 is used to determine first information, and this first information indicates the model alignment mode supported by the first communication device; the input / output interface 902 is used to send this first information.

[0186] Optionally, the input / output interface 902 is used to receive second information, and this second information indicates a target model alignment mode. The target model alignment mode is the model alignment mode supported by a second communication device, and the target alignment mode is one of the model alignment modes supported by the first communication device; the logic circuit 901 is used to perform model alignment based on the target model alignment mode.

[0187] Among them, the logic circuit 901 and the input / output interface 902 can also execute other steps performed by the first communication device or the second communication device in any embodiment and achieve the corresponding beneficial effects, which will not be elaborated here.

[0188] In a possible implementation, Figure 8 the shown processing unit 801 can be Figure 9 the logic circuit 901 in .

[0189] Optionally, the logic circuit 901 may be a processing device, and the functions of the processing device may be implemented partially or entirely by software. Among them, the functions of the processing device may be implemented partially or entirely by software.

[0190] Optionally, the processing device may include a memory and a processor. Among them, the memory is used to store a computer program, and the processor reads and executes the computer program stored in the memory to perform the corresponding processing and / or steps in any one of the method embodiments.

[0191] Optionally, the processing device may only include a processor. The memory for storing the computer program is located outside the processing device, and the processor is connected to the memory through a circuit / wire to read and execute the computer program stored in the memory. Among them, the memory and the processor may be integrated together, or may also be physically independent of each other.

[0192] Optionally, the processing device may be one or more chips, or one or more integrated circuits. For example, the processing device may be one or more field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), system-on-chip (SoCs), central processing units (CPUs), network processors (NPs), digital signal processing circuits (DSPs), microcontroller units (MCUs), programmable logic devices (PLDs), or other integrated chips, or any combination of the above chips or processors, etc.

[0193] Please refer to Figure 10 , for the communication device 1000 involved in the above embodiments provided by the embodiments of the present application. The communication device 1000 may specifically be the communication device as the terminal device in the above embodiments. Figure 10 The example shown is implemented by the terminal device (or a component in the terminal device).

[0194] Among them, a possible schematic diagram of the logical structure of the communication device 1000. The communication device 1000 may include but is not limited to at least one processor 1001 and a communication port 1002.

[0195] Among them, Figure 8The shown transceiver unit 802 can be a communication interface, and this communication interface can be Figure 10 the communication port 1002 in it. This communication port 1002 can include an input interface and an output interface. Alternatively, this communication port 1002 can also be a transceiver circuit, and this transceiver circuit can include an input interface circuit and an output interface circuit.

[0196] Further optionally, the device can further include at least one of a memory 1003 and a bus 1004. In the embodiments of the present application, the at least one processor 1001 is used to control and process the actions of the communication device 1000.

[0197] In addition, the processor 1001 can be a central processing unit, a general-purpose processor, a digital signal processor, an application-specific integrated circuit, a field-programmable gate array or other programmable logic devices, transistor logic devices, hardware components or any combination thereof. It can implement or execute various exemplary logical blocks, modules and circuits described in combination with the disclosure content of the present application. This processor can also be a combination that realizes computing functions, such as a combination including one or more microprocessors, a combination of a digital signal processor and a microprocessor, and so on. Those skilled in the art can clearly understand that for the convenience and conciseness of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the foregoing method embodiments, and will not be elaborated here.

[0198] It should be noted that Figure 10 the shown communication device 1000 can specifically be used to implement the steps implemented by the terminal device in the foregoing method embodiments, and achieve the technical effects corresponding to the terminal device. Figure 10 For the specific implementation manners of the shown communication device, reference can be made to the descriptions in the foregoing method embodiments, and will not be elaborated one by one here.

[0199] Please refer to Figure 11 , which is a schematic structural diagram of the communication device 1100 involved in the above embodiments provided for the embodiments of the present application. This communication device 1100 can specifically be the communication device as a network device in the above embodiments. Figure 11 The shown example is implemented by a network device (or components in a network device). Among them, the structure of this communication device can refer to Figure 11 the structure shown.

[0200] The communication device 1100 includes at least one processor 1111 and at least one network interface 1114. Further optionally, the communication device further includes at least one memory 1112, at least one transceiver 1113, and one or more antennas 1115. The processor 1111, the memory 1112, the transceiver 1113, and the network interface 1114 are connected, for example, connected by a bus. In the embodiments of the present application, this connection may include various interfaces, transmission lines, or buses, etc., and this embodiment does not limit this. The antenna 1115 is connected to the transceiver 1113. The network interface 1114 is used to enable the communication device to communicate with other communication devices through a communication link. For example, the network interface 1114 may include a network interface between the communication device and the core network device, such as an S1 interface. The network interface may include a network interface between the communication device and other communication devices (such as other network devices or core network devices), such as an X2 or Xn interface.

[0201] Wherein, Figure 8 The shown transceiver unit 802 may be a communication interface, and this communication interface may be Figure 11 the network interface 1114 in, and the network interface 1114 may include an input interface and an output interface. Alternatively, the network interface 1114 may also be a transceiver circuit, and this transceiver circuit may include an input interface circuit and an output interface circuit.

[0202] The processor 1111 is mainly used to process communication protocols and communication data, and control the entire communication device, execute software programs, and process data of software programs. For example, it is used to support the communication device to perform the actions described in the embodiments. The communication device may include a baseband processor and a central processor. The baseband processor is mainly used to process communication protocols and communication data, and the central processor is mainly used to control the entire terminal device, execute software programs, and process data of software programs. Figure 11 The processor 1111 in may integrate the functions of the baseband processor and the central processor. Those skilled in the art can understand that the baseband processor and the central processor may also be independent processors and are interconnected through technologies such as a bus. Those skilled in the art can understand that the terminal device may include multiple baseband processors to adapt to different network modes, the terminal device may include multiple central processors to enhance its processing ability, and various components of the terminal device may be connected through various buses. The baseband processor may also be expressed as a baseband processing circuit or a baseband processing chip. The central processor may also be expressed as a central processing circuit or a central processing chip. The function of processing communication protocols and communication data may be built into the processor or stored in the memory in the form of a software program, and the processor executes the software program to implement the baseband processing function.

[0203] The memory is mainly used to store software programs and data. The memory 1112 can exist independently and be connected to the processor 1111. Optionally, the memory 1112 can be integrated with the processor 1111, for example, integrated within a single chip. Among them, the memory 1112 can store the program code for implementing the technical solution of the embodiments of the present application, and is controlled by the processor 1111 for execution. Various types of computer program codes being executed can also be regarded as the driver programs of the processor 1111.

[0204] Figure 11 Only one memory and one processor are shown. In an actual terminal device, there can be multiple processors and multiple memories. The memory can also be referred to as a storage medium or a storage device, etc. The memory can be a storage element on the same chip as the processor, that is, an on-chip storage element, or an independent storage element, and the embodiments of the present application do not make any limitations in this regard.

[0205] The transceiver 1113 can be used to support the reception or transmission of radio frequency signals between the communication device and the terminal. The transceiver 1113 can be connected to the antenna 1115. The transceiver 1113 includes a transmitter Tx and a receiver Rx. Specifically, one or more antennas 1115 can receive radio frequency signals. The receiver Rx of the transceiver 1113 is used to receive the radio frequency signals from the antenna, convert the radio frequency signals into digital baseband signals or digital intermediate frequency signals, and provide the digital baseband signals or digital intermediate frequency signals to the processor 1111, so that the processor 1111 can perform further processing on the digital baseband signals or digital intermediate frequency signals, such as demodulation processing and decoding processing. In addition, the transmitter Tx in the transceiver 1113 is also used to receive the modulated digital baseband signals or digital intermediate frequency signals from the processor 1111, convert the modulated digital baseband signals or digital intermediate frequency signals into radio frequency signals, and transmit the radio frequency signals through one or more antennas 1115. Specifically, the receiver Rx can selectively perform one-stage or multi-stage down-conversion processing and analog-to-digital conversion processing on the radio frequency signals to obtain digital baseband signals or digital intermediate frequency signals, and the order of the down-conversion processing and the analog-to-digital conversion processing can be adjusted. The transmitter Tx can selectively perform one-stage or multi-stage up-conversion processing and digital-to-analog conversion processing on the modulated digital baseband signals or digital intermediate frequency signals to obtain radio frequency signals, and the order of the up-conversion processing and the digital-to-analog conversion processing can be adjusted. The digital baseband signals and the digital intermediate frequency signals can be collectively referred to as digital signals.

[0206] The transceiver 1113 may also be referred to as a transceiver unit, a transceiver, a transceiver device, etc. Optionally, the devices used to implement the receiving function in the transceiver unit can be regarded as the receiving unit, and the devices used to implement the transmitting function in the transceiver unit can be regarded as the transmitting unit. That is, the transceiver unit includes a receiving unit and a transmitting unit. The receiving unit may also be referred to as a receiver, an input port, a receiving circuit, etc., and the transmitting unit may be referred to as a transmitter, a transmitter, or a transmitting circuit, etc.

[0207] It should be noted that Figure 11 The illustrated communication device 1100 can specifically be used to implement the steps implemented by the network device in the foregoing method embodiments and achieve the corresponding technical effects of the network device. Figure 11 For the specific implementation manners of the illustrated communication device 1100, reference can be made to the descriptions in the foregoing method embodiments, and details are not described herein one by one.

[0208] Please refer to Figure 12 , which is a schematic structural diagram of the communication device involved in the foregoing embodiments provided by the embodiments of the present application.

[0209] It can be understood that the communication device 120 includes, for example, modules, units, components, circuits, or interfaces, etc., which are appropriately configured together to execute the technical solutions provided by the present application. The communication device 120 may be the terminal device or network device described above, or a component (such as a chip) in these devices, for implementing the methods described in the following method embodiments. The communication device 120 includes one or more processors 121. The processor 121 may be a general-purpose processor or a dedicated processor, etc. For example, it may be a baseband processor or a central processing unit. The baseband processor can be used to process communication protocols and communication data, and the central processing unit can be used to control the communication device (such as a RAN node, a terminal, or a chip, etc.), execute software programs, and process the data of the software programs.

[0210] Optionally, in one design, the processor 121 may include a program 123 (sometimes also referred to as code or instructions), and the program 123 can be run on the processor 121, so that the communication device 120 executes the methods described in the following embodiments. In another possible design, the communication device 120 includes a circuit ( Figure 12 not shown).

[0211] Optionally, the communication device 120 may include one or more memories 122, on which there is a program 124 (sometimes also referred to as code or instructions), and the program 124 can be run on the processor 121, so that the communication device 120 executes the methods described in the above method embodiments.

[0212] Optionally, the processor 121 and / or the memory 122 may include AI modules 127, 128, which are used to implement AI-related functions. The AI modules may be implemented in software, hardware, or a combination of both. For example, the AI module may include a radio intelligence control (RIC) module. For example, the AI module may be a near real-time RIC or a non-real-time RIC.

[0213] Optionally, data may also be stored in the processor 121 and / or the memory 122. The processor and the memory may be provided separately or integrated together.

[0214] Optionally, the communication device 120 may further include a transceiver 125 and / or an antenna 126. The processor 121 is sometimes also referred to as a processing unit and controls the communication device (such as a RAN node or a terminal). The transceiver 125 is sometimes also referred to as a transceiver unit, a transceiver, a transceiver circuit, or a transceiver, etc., and is used to implement the transceiver function of the communication device through the antenna 126.

[0215] Among them, Figure 8 the shown transceiver unit 802 may be a communication interface, and this communication interface may be Figure 12 the transceiver 125 in it. The transceiver 125 may include an input interface and an output interface. Alternatively, the transceiver 125 may also be a transceiver circuit, and this transceiver circuit may include an input interface circuit and an output interface circuit.

[0216] The embodiment of the present application further provides a computer-readable storage medium, which is used to store one or more computer-executable instructions. When the computer-executable instructions are executed by a processor, the processor executes the method described in the possible implementation manners of the first communication device or the second communication device in the foregoing embodiments.

[0217] The embodiment of the present application further provides a computer program product (or referred to as a computer program). When the computer program product is executed by the processor, the processor executes the method in the possible implementation manners of the above first communication device or second communication device.

[0218] The embodiments of the present application further provide a chip system, which includes at least one processor for supporting a communication device to implement the functions involved in the possible implementation manners of the above communication device. Optionally, the chip system further includes an interface circuit, and the interface circuit provides program instructions and / or data for the at least one processor. In a possible design, the chip system may further include a memory for storing the necessary program instructions and data of the communication device. The chip system may be composed of chips or may include chips and other discrete devices, where the communication device may specifically be the first communication device or the second communication device in the foregoing method embodiments.

[0219] The embodiments of the present application further provide a communication system, and the network system architecture includes the first communication device and the second communication device in any of the above embodiments.

[0220] In several embodiments provided in the present application, it should be understood that the disclosed systems, devices, and methods may be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the units is only a logical function division, and there may be other division methods in actual implementation. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Another point is that the displayed or discussed coupling or direct coupling or communication connection to each other may be through some interfaces, and the indirect coupling or communication connection of the devices or units may be in electrical, mechanical, or other forms.

[0221] The units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place or may be distributed to multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0222] In addition, each functional unit in various embodiments of the present application may be integrated into one processing unit, or each unit may exist physically alone, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of a software functional unit. If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it may be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present application, in essence, or the part that makes a contribution, or all or part of the technical solution, may be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present application. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs that can store program codes.

Claims

1. A training method, characterized in that, The method includes: Receiving training data for training a first module group, where the training data is obtained based on a second module group in a second model. The first module group includes all modules or a cascaded partial module in a first model, the second module group includes all modules or a cascaded partial module in the second model, the first module group includes modules corresponding to the modules in the second module group, and the first module group includes a target module, where the target module is a module in the first model that is not aligned with the corresponding module in the second module group; Training the first module group according to the training data.

2. The method according to claim 1, characterized in that, The first module group includes an adaptation module, and the adaptation module and the target module are different modules. Training the first module group according to the training data includes: Adjusting the parameters of the adaptation module according to the training data.

3. The method according to claim 2, characterized in that, The adaptation module is adjacent to the target module; or the adaptation module is disposed on the output side or the input side of the first module group.

4. The method according to claim 2 or 3, characterized in that, The first module group includes N target modules and N adaptation modules, the N target modules and the N adaptation modules correspond one by one, and each adaptation module is adjacent to the corresponding target module; N is an integer greater than or equal to 1.

5. The method according to claim 2 or 3, characterized in that, The first module group includes M target modules and 1 adaptation module; M is an integer greater than or equal to 2.

6. The method according to claim 1, characterized in that, Training the first module group according to the training data includes: Adjusting the parameters of the target module according to the training data.

7. The method according to any one of claims 1 to 6, characterized in that, The training data includes at least one of the following: The input data of the second module group and the output data of the second module group; The gradient obtained based on the input data of the first module group, the output data of the first module group, and the output data of the second module group; The second module group; The second model.

8. The method according to any one of claims 1 to 7, characterized in that, The method further includes: Sending or receiving verification data for determining the target module.

9. The method according to claim 8, characterized in that, The verification data includes first verification data for verifying whether a third module group and a fourth module group are aligned. The third module group includes all modules or a cascaded partial module in the first model, and the fourth module group includes the modules corresponding to the modules in the third module group in the second model.

10. The method according to claim 9, characterized in that, The verification data includes second verification data for verifying whether a fifth module group and a sixth module group are aligned. The fifth module group is a subset of the third module group, and the sixth module group is a subset of the fourth module group.

11. The method according to claim 9, characterized in that, The verification data includes third verification data for verifying whether a seventh module group and an eighth module group are aligned. The third module group is a subset of the seventh module group, and the fourth module group is a subset of the eighth module group.

12. A training method, characterized in that, The method includes: Send training data, where the training data is used to train a first module group, the training data is obtained based on a second module group in a second model, the first module group includes all modules or cascaded partial modules in a first model, the second module group includes all modules or cascaded partial modules in the second model, the first module group includes modules corresponding to the modules in the second module group, and the first module group includes a target module, where the target module is a module in the first model that is not aligned with the corresponding module in the second module group.

13. The method according to claim 12, characterized in that, The training data includes at least one of the following: Input data of the second module group and output data of the second module group; Gradients obtained based on the input data of the first module group, the output data of the first module group, and the output data of the second module group; The second module group; The second model.

14. The method according to claim 12 or 13, characterized in that, The method further includes: Sending or receiving verification data, where the verification data is used to determine the target module.

15. The method according to claim 14, characterized in that, The verification data includes first verification data, where the first verification data is used to verify whether a third module group is aligned with a fourth module group, the third module group includes all modules or cascaded partial modules in the first model, and the fourth module group includes modules corresponding to the modules in the third module group in the second model.

16. The method according to claim 15, characterized in that, The verification data includes second verification data, where the second verification data is used to verify whether a fifth module group is aligned with a sixth module group, the fifth module group is a subset of the third module group, and the sixth module group is a subset of the fourth module group.

17. The method according to claim 15, characterized in that, The verification data includes third verification data, where the third verification data is used to verify whether a seventh module group is aligned with an eighth module group, the third module group is a subset of the seventh module group, and the fourth module group is a subset of the eighth module group.

18. A communication device, characterized in that, Includes modules for performing the method according to any one of claims 1 to 17.

19. A communication device, characterized in that, Includes at least one processor, where the at least one processor is coupled to a memory; the at least one processor is used to perform the method according to any one of claims 1 to 17.

20. The communication device according to claim 19, characterized in that, The communication device is a chip or a chip system.

21. A readable storage medium, characterized in that, A computer program or instruction is stored in the storage medium, and when the computer program or instruction is executed by the communication device, the method according to any one of claims 1 to 17 is implemented.

22. A computer program product, characterized in that, Includes instructions, and when the instructions run on a computer, the computer is caused to execute the method according to any one of claims 1 to 17.

Images