Communication methods, devices and systems

Abstract

This application provides a communication method, apparatus, and system. The method includes: a first device sending indication information to a second device, the indication information indicating information about a first neural network model, the first neural network model including dedicated layer networks and common layer networks, the dedicated layer networks being dedicated to the first neural network model, and the common layer networks being a common network shared by the first neural network model and a second neural network model; the second device determining the information of the first neural network model based on the indication information. Because the neural network model is divided into common layer networks and dedicated layer networks, and the common layer networks are common networks for different neural network models, the overall overhead of indicating the information of the neural network model can be reduced. Furthermore, the amount of neural network model information pre-stored by the neural network model user (such as a terminal device) is also reduced accordingly, thereby reducing the storage overhead of the neural network model information.

Description

Technical Field

[0001] This application relates to the field of communication technology, and in particular to communication methods, devices and systems. Background Technology

[0002] New radio (NR) is a popular standard for next-generation cellular communication networks, covering three major scenarios: enhanced mobile broadband (eMBB), ultra-reliable and low-latency communication (uRLLC), and massive machine-type communication (mMTC). eMBB emphasizes high throughput, uRLLC emphasizes high reliability and low latency, and mMTC emphasizes massive connections. These new demands present significant challenges to modern communication systems and have also led to new research directions in modern communication theory. Existing communication theories often utilize system structure information and design the entire system using a top-down approach. This approach often encounters problems when facing complex application scenarios and massive amounts of data that are difficult to model, which is a limitation of existing communication theories. Therefore, how to establish new communication mechanisms to meet the needs of future wireless communication systems is an urgent problem to be solved. Summary of the Invention

[0003] This application provides communication methods, apparatus, and systems to reduce the storage overhead of information in neural network models and the signaling overhead during transmission.

[0004] In a first aspect, embodiments of this application provide a communication method, which can be executed by a second device or other devices such as a chip used in the second device. The second device can be a terminal device. The method includes: obtaining (e.g., receiving) indication information from a first device, the indication information being used to indicate information of a first neural network model, the first neural network model including a dedicated layer network and a common layer network, the dedicated layer network being dedicated to the first neural network model, and the common layer network being a common network between the first neural network model and a second neural network model; and determining information of the first neural network model based on the indication information.

[0005] In this application embodiment, a certain piece of information, such as information A, used to indicate another piece of information or feature, such as information B, may include:

[0006] Information A includes information B;

[0007] Information A indicates information B from one or more candidate information Bs, such as indicating an index of information B, wherein each candidate information B in the one or more candidate information Bs corresponds to a unique index used to identify that candidate information B; or...

[0008] Information A is used to determine information B. For example, information B can be obtained by performing mathematical operations or filtering on information A.

[0009] Based on the above scheme, the neural network model includes a common layer network and a dedicated layer network. The common layer network is a shared network for different neural network models, thereby reducing overhead. For example, the overall overhead of instructing the neural network model can be reduced. Furthermore, the amount of information about the neural network model pre-stored by the user (such as a terminal device) is also reduced accordingly, thereby reducing the storage overhead of the neural network model's information.

[0010] As one implementation method, the aforementioned indication information can be carried through broadcast messages (such as master information blocks (MIBs)), system messages (such as system information blocks (SIBs)), common messages, radio resource control (RRC) signaling, medium access control control elements (MAC CEs) dedicated to the second device, and / or downlink control information (DCIs) dedicated to the second device. Common messages can be common DCIs or common MAC CEs, etc. "Dedicated to the second device" can also be referred to as "specific to the second device."

[0011] In this embodiment of the application, the dedicated layer network is also referred to as the dedicated layer, the common layer network as the common layer, and the information of the neural network model as the neural network model information. The neural network model can also be referred to as the neural network.

[0012] As one possible implementation, after determining the information of the first neural network model, a new first neural network model can be determined based on this information, or an existing first neural network model can be updated. Specifically, the information of the first neural network model can be used to obtain a new first neural network model, or to replace an existing one. Alternatively, the existing first neural network model can be trained online based on the information of the first neural network model to obtain a trained and updated first neural network model. Optionally, before updating the existing first neural network model, it can be further determined that the performance of the first neural network model corresponding to the information of the first neural network model is better than the performance of the existing first neural network model.

[0013] As one possible implementation, the information of the first neural network model includes at least one of the following: neural network cascade structure information, common layer network structure information, common layer network parameters, dedicated layer network structure information, and dedicated layer network parameters. The neural network cascade structure information includes the cascade structure information of the dedicated layer network and the common layer network.

[0014] Optionally, the neural network cascade structure information is used to indicate the connection structure between the specialized layers and common layers contained in the neural network. For example, it is used to indicate the interconnections between neurons in specialized layers and neurons in common layers. Common layer network structure information is used to indicate the number of layers in the common layer and the inter-layer structure within the common layer. Common layer network parameters are used to indicate the weights, biases, and activation functions of each neuron in each layer of the common layer. Similarly, specialized layer network structure information is used to indicate the number of layers in the specialized layer and the inter-layer structure within the specialized layer. Specialized layer network parameters are used to indicate the weights, biases, and activation functions of each neuron in each layer of the specialized layer.

[0015] Optionally, when the information of the first neural network model includes at least two of the following: neural network cascade structure information, common layer network structure information, common layer network parameters, dedicated layer network structure information, and dedicated layer network parameters, the aforementioned indication information can indicate these at least two types of information through one or more messages. For example, when the information of the first neural network model includes at least one of common layer network structure information and common layer network parameters, and also includes at least one of neural network cascade structure information, dedicated layer network structure information, and dedicated layer network parameters, the aforementioned indication information includes first indication information and second indication information. The first indication information is used to indicate at least one of common layer network structure information and common layer network parameters, and the second indication information is used to indicate at least one of neural network cascade structure information, dedicated layer network structure information, and dedicated layer network parameters. Optionally, the first indication information is carried through MIB, SIB, or a common message. The second indication information is carried by RRC signaling, MAC CE, or dedicated DCI specific to the second device.

[0016] Based on the above scheme, users of neural network models (such as terminal devices) can accurately determine the information of the neural network model.

[0017] As one possible implementation, the information of the first neural network model includes neural network cascade structure information, including: the information of the first neural network model indicating the neural network cascade structure, or the information of the first neural network model being used to indicate the configured neural network cascade structure from a candidate set of neural network cascade structures.

[0018] Based on the above scheme, the cascaded structure of neural networks can be flexibly indicated.

[0019] As one possible implementation, the information of the first neural network model includes common layer network structure information, including: the information of the first neural network model indicating the common layer network structure, or the information of the first neural network model being used to indicate the configured common layer network structure from a set of candidate common layer network structures.

[0020] Based on the above scheme, the common layer network structure can be flexibly indicated.

[0021] As one possible implementation, the information of the first neural network model includes common layer network parameter information, including: the information of the first neural network model indicating the common layer network parameters, or the information of the first neural network model being used to indicate the configured common layer network parameters from a candidate set of common layer network parameters.

[0022] Based on the above scheme, common layer network parameter information can be flexibly indicated.

[0023] As one possible implementation, the information of the first neural network model includes dedicated layer network structure information, including: the information of the first neural network model indicating the dedicated layer network structure, or the information of the first neural network model being used to indicate the configured dedicated layer network structure from a set of dedicated layer network structure candidates.

[0024] Based on the above scheme, the network structure information of the dedicated layer can be flexibly indicated.

[0025] As one possible implementation, the information of the first neural network model includes dedicated layer network parameter information, including: the information of the first neural network model indicating the dedicated layer network parameters, or the information of the first neural network model being used to indicate the configured dedicated layer network parameters from a set of dedicated layer network parameter candidates.

[0026] Based on the above scheme, the parameters of the dedicated layer network can be flexibly indicated.

[0027] As one possible implementation, the aforementioned indication information carries an index of the neural network cascade structure information, an index of the dedicated layer network structure information, and an index of the dedicated layer network parameters. Alternatively, the aforementioned indication information carries an index of the neural network cascade structure information, an index of the dedicated layer network structure information, and the dedicated layer network parameters. Alternatively, the aforementioned indication information carries an index of the neural network cascade structure information, the dedicated layer network structure information, and the dedicated layer network parameters.

[0028] As one possible implementation, configuration information is received from the first device, which indicates one or more of the following: a candidate set of neural network cascade structure information, a candidate set of common layer network structure information, a candidate set of common layer network parameters, a candidate set of dedicated layer network structure information, and a candidate set of dedicated layer network parameters.

[0029] Based on the above scheme, configuration information is received in advance from the first device (such as a network device) to configure one or more of the following: the candidate set of neural network cascade structure information, the candidate set of common layer network structure information, the candidate set of common layer network parameters, the candidate set of dedicated layer network structure information, and the candidate set of dedicated layer network parameters. Thus, the first device only needs to indicate the index corresponding to the information in the candidate set through the indication information, which can reduce the overhead of the indication information.

[0030] As one possible implementation, auxiliary information is sent to the first device, which indicates the wireless system parameters between the first device and the second device.

[0031] Based on the above scheme, the first device (such as a network device) can make decisions on the parameters and structural information corresponding to the neural network model based on the auxiliary information, which helps to improve the accuracy of the neural network model.

[0032] As one possible implementation, capability information is sent to the first device, which indicates one or more of the following:

[0033] 1) Does it support using neural networks to replace or implement the functions of the communication module?

[0034] 2) Does it support neural network structures where public layer networks and private layer networks are cascaded?

[0035] 3) Does it support receiving dedicated layer network structure information and / or dedicated layer network parameters via signaling?

[0036] 4) Stored protocol-predefined or pre-configured dedicated layer network structure information;

[0037] 5) Stored protocol predefined or preconfigured dedicated layer network parameters;

[0038] 6) Memory space that can be used to store information about the cascaded structure of neural networks, information about the structure of common layers, parameters of common layers, information about the structure of private layers and / or parameters of private layers;

[0039] 7) Computing power information that can be used to run neural networks.

[0040] Based on the above scheme, the first device (such as a network device) can make decisions on the parameters and structural information corresponding to the neural network model and / or the method of sending decision instruction information based on the capability information, which helps to improve the accuracy of the neural network model and reduce the overhead of sending instruction information.

[0041] Secondly, embodiments of this application provide a communication method, which can be executed by a first device or other devices such as a chip used in the first device. The first device can be a network device or a terminal device. The method includes: determining indication information, which is used to indicate information of a first neural network model, the first neural network model including a dedicated layer network and a common layer network, the dedicated layer network being dedicated to the first neural network model, and the common layer network being a common network between the first neural network model and a second neural network model; and sending the indication information to the second device.

[0042] For information on the indication information and the first neural network model, please refer to the first part, which will not be repeated here.

[0043] As one possible implementation, configuration information is sent to the second device, which indicates one or more of the following: a candidate set of neural network cascade structure information, a candidate set of common layer network structure information, a candidate set of common layer network parameters, a candidate set of dedicated layer network structure information, and a candidate set of dedicated layer network parameters.

[0044] As one possible implementation, auxiliary information is received from the second device, which indicates the wireless system parameters between the first device and the second device. Optionally, based on the auxiliary information, it is determined that the neural network model information of the second device needs to be updated. Optionally, based on the auxiliary information, it is determined that the wireless system parameters between the first device and the second device have changed.

[0045] As one possible implementation, capability information is received from the second device, which indicates one or more of the following:

[0046] 1) Does it support using neural networks to replace or implement the functions of the communication module?

[0047] 2) Does it support neural network structures where public layer networks and private layer networks are cascaded?

[0048] 3) Does it support receiving dedicated layer network structure information and / or dedicated layer network parameters via signaling?

[0049] 4) Stored protocol-predefined or pre-configured dedicated layer network structure information;

[0050] 5) Stored protocol predefined or preconfigured dedicated layer network parameters;

[0051] 6) Memory space that can be used to store information about the cascaded structure of neural networks, information about the structure of common layers, parameters of common layers, information about the structure of private layers and / or parameters of private layers;

[0052] 7) Computing power information that can be used to run neural networks.

[0053] Thirdly, embodiments of this application provide a communication method, which can be executed by a second device or other devices such as a chip used in the second device. The second device can be a terminal device. The method includes: obtaining (receiving) first indication information from a first device, the first indication information being used to indicate public information of a first neural network model, the first neural network model including a dedicated layer network and a public layer network, the public information of the first neural network model including information of the public layer network; and obtaining (e.g., receiving) second indication information from the first device, the second indication information being used to indicate information of the dedicated layer network.

[0054] Based on the above scheme, the neural network model includes a common layer network and a dedicated layer network. The common layer network is a shared network for neural network models on different terminal devices, while the dedicated layer network is used independently by the neural network model on each terminal device. This reduces the overall overhead of information indicating the neural network model. Furthermore, the amount of information about the neural network model pre-stored by users of the neural network model (such as terminal devices) is also reduced accordingly, thereby reducing the storage overhead of the neural network model's information.

[0055] As one possible implementation, the information of the common layer network includes common layer network structure information and / or common layer network parameters.

[0056] As one possible implementation, the information of the dedicated layer network includes at least one of the following: neural network cascade structure information, dedicated layer network structure information, and dedicated layer network parameters. The neural network cascade structure information includes the cascade structure information of the dedicated layer network and the common layer network.

[0057] As one possible implementation method, the first indication information is carried through the MIB, system messages (such as SIB), or public messages.

[0058] As one possible implementation method, the second indication information is carried via RRC signaling, MAC CE, or DCI.

[0059] Fourthly, embodiments of this application provide a communication method, which can be executed by a first device or other devices such as a chip used in the first device. The first device can be a network device or a terminal device. The method includes: sending first indication information to a second device, the first indication information being used to indicate public information of a first neural network model, the first neural network model including a dedicated layer network and a public layer network, the public information of the first neural network model including information of the public layer network; and sending second indication information to the second device, the second indication information being used to indicate information of the dedicated layer network.

[0060] Information regarding the public layer network, the private layer network, the first instruction information, and the second instruction information can be found in the third section and will not be elaborated upon here.

[0061] Fifthly, an apparatus is provided, which may be a second device, a device within a second device, or a device compatible with a second device. In one design, the apparatus may include modules corresponding to the methods / operations / steps / actions described in the first aspect. These modules may be hardware circuits, software, or a combination of hardware circuits and software. In another design, the apparatus may include a processing module and a communication module. Exemplarily, the communication module is used to obtain instruction information from a first device. This instruction information indicates information about a first neural network model, which includes dedicated layer networks and common layer networks. The dedicated layer networks are dedicated to the first neural network model, and the common layer networks are common networks shared by the first neural network model and a second neural network model.

[0062] The determination module is used to determine the information of the first neural network model based on the indication information.

[0063] In a sixth aspect, an apparatus is provided, which may be a first device, a device within the first device, or a device compatible with the first device. In one design, the apparatus may include modules corresponding to each of the methods / operations / steps / actions described in the second aspect. These modules may be hardware circuits, software, or a combination of hardware circuits and software. In one design, the apparatus may include a processing module and a communication module. Exemplarily, the processing module is configured to determine indication information for indicating information about a first neural network model, which includes dedicated layer networks and common layer networks. The dedicated layer networks are dedicated to the first neural network model, and the common layer networks are common networks shared by the first neural network model and a second neural network model.

[0064] The communication module is used to send the instruction information to the second device.

[0065] In a seventh aspect, an apparatus is provided, which may be a second device, a device within a second device, or a device compatible with a second device. In one design, the apparatus may include modules corresponding to each of the methods / operations / steps / actions described in the third aspect; these modules may be hardware circuits, software, or a combination of hardware circuits and software. In one design, the apparatus may include a processing module and a communication module. Exemplarily,

[0066] A communication module is configured to acquire first indication information from a first device, the first indication information indicating common information of a first neural network model, the first neural network model including a dedicated layer network and a common layer network, the common information of the first neural network model including information of the common layer network; and to receive second indication information from the first device, the second indication information indicating information of the dedicated layer network. A processing module is configured to process the first and second indication information received by the communication module.

[0067] Eighthly, an apparatus is provided, which may be a first device, a device within the first device, or a device compatible with the first device. In one design, the apparatus may include modules corresponding to each of the methods / operations / steps / actions described in the fourth aspect. These modules may be hardware circuits, software, or a combination of hardware circuits and software. In one design, the apparatus may include a processing module and a communication module. Exemplarily, the communication module is configured to send first indication information to a second device, the first indication information indicating common information of a first neural network model, the first neural network model including dedicated layer networks and common layer networks, the common information of which includes information about the common layer networks; and to send second indication information to the second device, the second indication information indicating information about the dedicated layer networks. The processing module is configured to generate the first indication information and the second indication information.

[0068] Ninthly, embodiments of this application provide an apparatus, the apparatus including a processor for implementing the method described in the first aspect above. The apparatus may further include a memory for storing instructions and data. The memory is coupled to the processor, and when the processor executes the instructions stored in the memory, it can implement the method described in the first aspect above. The apparatus may further include a communication interface for communicating with other devices; exemplaryly, the communication interface may be a transceiver, circuit, bus, module, pin, or other type of communication interface. In one possible device, the apparatus includes:

[0069] Memory, used to store program instructions;

[0070] The processor is configured to use a communication interface to obtain indication information from a first device, the indication information being used to indicate information about a first neural network model, the first neural network model including a dedicated layer network and a common layer network, the dedicated layer network being dedicated to the first neural network model, and the common layer network being a common network between the first neural network model and a second neural network model; and to determine information about the first neural network model based on the indication information.

[0071] Tenthly, embodiments of this application provide an apparatus, the apparatus including a processor for implementing the method described in the second aspect above. The apparatus may further include a memory for storing instructions and data. The memory is coupled to the processor, and when the processor executes the instructions stored in the memory, it can implement the method described in the second aspect above. The apparatus may further include a communication interface for communicating with other devices; exemplaryly, the communication interface may be a transceiver, circuit, bus, module, pin, or other type of communication interface. In one possible device, the apparatus includes:

[0072] Memory, used to store program instructions;

[0073] The processor is configured to determine indication information for indicating information about a first neural network model, the first neural network model including a dedicated layer network and a common layer network, the dedicated layer network being dedicated to the first neural network model, and the common layer network being a common network between the first neural network model and a second neural network model; and to send the indication information to a second device via a communication interface.

[0074] Eleventhly, embodiments of this application provide an apparatus, the apparatus including a processor for implementing the method described in the third aspect above. The apparatus may further include a memory for storing instructions and data. The memory is coupled to the processor, and when the processor executes the instructions stored in the memory, it can implement the method described in the third aspect above. The apparatus may further include a communication interface for communicating with other devices; exemplaryly, the communication interface may be a transceiver, circuit, bus, module, pin, or other type of communication interface. In one possible device, the apparatus includes:

[0075] Memory, used to store program instructions;

[0076] The processor is configured to utilize a communication interface to: obtain first indication information from a first device, the first indication information being used to indicate public information of a first neural network model, the first neural network model including a dedicated layer network and a public layer network, the public information of the first neural network model including information of the public layer network; and obtain second indication information from the first device, the second indication information being used to indicate information of the dedicated layer network.

[0077] In a twelfth aspect, embodiments of this application provide an apparatus comprising a processor for implementing the method described in the fourth aspect above. The apparatus may further include a memory for storing instructions and data. The memory is coupled to the processor, and when the processor executes the instructions stored in the memory, it can implement the method described in the fourth aspect above. The apparatus may further include a communication interface for communicating with other devices; exemplaryly, the communication interface may be a transceiver, circuit, bus, module, pin, or other type of communication interface. In one possible device, the apparatus includes:

[0078] Memory, used to store program instructions;

[0079] The processor is configured to, via a communication interface, send first instruction information to a second device, the first instruction information indicating common information of a first neural network model, the first neural network model including a dedicated layer network and a common layer network, the common information of the first neural network model including information of the common layer network; and send second instruction information to the second device, the second instruction information indicating information of the dedicated layer network.

[0080] In a thirteenth aspect, embodiments of this application also provide a computer-readable storage medium, including instructions that, when executed on a computer, cause the computer to perform the method described in any one of the first to fourth aspects.

[0081] In a fourteenth aspect, this application also provides a computer program product, including instructions that, when executed on a computer, cause the computer to perform the method described in any one of the first to fourth aspects.

[0082] In a fifteenth aspect, embodiments of this application provide a chip system including a processor and potentially a memory, for implementing the methods described in any one of the first to fourth aspects. The chip system may be composed of chips or may include chips and other discrete devices.

[0083] In a sixteenth aspect, embodiments of this application provide a system comprising the means of the fifth aspect or the ninth aspect, and the means of the sixth aspect or the tenth aspect.

[0084] In a seventeenth aspect, embodiments of this application provide a system comprising the means of the seventh aspect or the eleventh aspect, and the means of the eighth aspect or the twelfth aspect. Attached Figure Description

[0085] Figure 1 This is a schematic diagram of a neuron structure provided in an embodiment of this application;

[0086] Figure 2 A schematic diagram of the layer relationships of a neural network provided in an embodiment of this application;

[0087] Figure 3 A schematic diagram of a CNN provided for an embodiment of this application;

[0088] Figure 4 A schematic diagram of an RNN provided for an embodiment of this application;

[0089] Figure 5 This is a schematic diagram of the network architecture provided in the embodiments of this application;

[0090] Figure 6 A schematic diagram of the neural network architecture for joint transmit / receive optimization of constellation modulation / demodulation provided in the embodiments of this application;

[0091] Figure 7 Example diagram of the cascaded form of the neural network provided in the embodiments of this application;

[0092] Figure 8(a) is a schematic diagram of the communication method provided in an embodiment of this application;

[0093] Figure 8(b) is a schematic diagram of the communication method provided in an embodiment of this application;

[0094] Figure 9 This is a schematic diagram of a communication method provided in an embodiment of this application;

[0095] Figure 10 This is a schematic diagram of a communication method provided in an embodiment of this application;

[0096] Figure 11 This is a schematic diagram of a communication method provided in an embodiment of this application;

[0097] Figure 12 This is a schematic diagram of a communication method provided in an embodiment of this application;

[0098] Figure 13 This is a schematic diagram of a communication method provided in an embodiment of this application;

[0099] Figure 14 This is a schematic diagram of a communication method provided in an embodiment of this application;

[0100] Figure 15(a) is a schematic diagram of the constellation neural network modulation module structure provided in an embodiment of this application;

[0101] Figure 15(b) is a schematic diagram of the neural network modeling constellation generation network provided in the embodiment of this application;

[0102] Figure 15(c) is a schematic diagram of the constellation neural network demodulation module structure provided in the embodiment of this application;

[0103] Figure 16 A schematic diagram of a communication device provided in an embodiment of this application;

[0104] Figure 17 This is a schematic diagram of a communication device provided in an embodiment of this application. Detailed Implementation

[0105] Machine learning (ML) has attracted widespread attention from academia and industry in recent years. Due to its significant advantages in handling structured information and massive amounts of data, many researchers in the field of communications have also turned their attention to machine learning. Machine learning-based communication technologies have enormous potential in signal classification, channel estimation, and performance optimization. Most existing communication systems are designed modularly, meaning they consist of multiple modules. For this modular communication architecture, researchers in the field have developed many techniques to optimize the performance of each module; however, optimal performance of each module does not necessarily mean optimal performance of the entire communication system. Some new research shows that end-to-end optimization (i.e., optimizing the entire communication system) is superior to optimizing a single model. Machine learning provides an advanced and powerful tool for maximizing end-to-end performance. In wireless communication systems, under complex and large-scale communication scenarios, channel conditions change rapidly. Many traditional communication models, such as massively multi-input multiple-output (MIMO) models, heavily rely on channel state information. Their performance deteriorates under nonlinear time-varying channels, making accurate acquisition of channel state information (CSI) crucial for system performance. By utilizing machine learning techniques, it is possible for communication systems to learn changing channel models and provide timely feedback on channel status.

[0106] Based on the above considerations, using machine learning technology in wireless communication can adapt to new demands in future wireless communication scenarios.

[0107] Machine learning is an important technological approach to achieving artificial intelligence. Machine learning can include supervised learning, unsupervised learning, and reinforcement learning.

[0108] Supervised learning, based on collected sample values ​​and labels, uses machine learning algorithms to learn the mapping relationship between sample values ​​and labels, and expresses this learned mapping relationship using a machine learning model. The process of training the machine learning model is the process of learning this mapping relationship. For example, in signal detection, the noisy received signal is the sample, and the corresponding real constellation point is the label. Machine learning aims to learn the mapping relationship between samples and labels through training, that is, to enable the machine learning model to learn a signal detector. During training, the model parameters are optimized by calculating the error between the model's predicted values ​​and the real labels. Once the mapping relationship is learned, it can be used to predict the sample label of each new sample. The mapping relationship learned in supervised learning can include linear mappings and nonlinear mappings. Based on the type of label, the learning task can be divided into classification tasks and regression tasks.

[0109] Unsupervised learning relies solely on collected sample values, using algorithms to discover inherent patterns within the samples. One type of unsupervised learning algorithm uses the samples themselves as supervisory signals, meaning the model learns the mapping relationship from samples to samples; this is also known as self-supervised learning. During training, model parameters are optimized by calculating the error between the model's predictions and the samples themselves. Self-supervised learning can be used for signal compression and decompression recovery applications; common algorithms include autoencoders and generative adversarial networks.

[0110] Reinforcement learning, unlike supervised learning, is a type of algorithm that learns problem-solving strategies through interaction with the environment. Unlike supervised and unsupervised learning, reinforcement learning problems do not have explicit "correct" action labels. The algorithm needs to interact with the environment to obtain reward signals from the environment, and then adjust its decision actions to obtain a larger reward signal value. For example, in downlink power control, the reinforcement learning model adjusts the downlink transmission power of each user based on the total system throughput feedback from the wireless network, aiming to achieve a higher system throughput. The goal of reinforcement learning is also to learn the mapping relationship between the environment state and the optimal decision action. However, because the label of the "correct action" cannot be obtained in advance, the network cannot be optimized by calculating the error between the action and the "correct action." Reinforcement learning training is achieved through iterative interaction with the environment.

[0111] Deep neural networks (DNNs) are a specific implementation of machine learning. According to the general approximation theorem, DNNs can theoretically approximate any continuous function, thus enabling them to learn arbitrary mappings. Traditional communication systems rely on extensive expert knowledge to design communication modules, while DNN-based deep learning communication systems can automatically discover hidden pattern structures from large datasets, establish mapping relationships between data, and achieve performance superior to traditional modeling methods.

[0112] The idea behind DNNs (Dual Neural Networks) originates from the neuronal structure of the brain. Each neuron performs a weighted summation of its input values, and then passes the weighted sum through an activation function to produce the output. For example... Figure 1 The diagram shown is a schematic of a neuron structure. Assume the neuron's input is x = [x0, x1, ... x]. n The weights corresponding to the inputs are w = [w0, w1, ... w1]. n The bias of the weighted summation is b. The activation function can take many forms. As an example, the activation function is: y = f(x) = max{0,x}. Then the output of a neuron is: Among them, w i x i Indicates w i With x i The product of the input and output layers. DNNs typically have a multi-layered structure, with each layer containing multiple neurons. The input layer processes the received values ​​through neurons and then passes them to the hidden layers. Similarly, the hidden layers then pass the calculation results to the final output layer, producing the final output of the DNN. For example... Figure 2 The diagram shown illustrates the layer relationships within a neural network.

[0113] DNNs typically have one or more hidden layers, which directly influence their ability to extract information and fit functions. Increasing the number of hidden layers or the number of neurons in each layer can improve the function fitting ability of a DNN. The parameters of each neuron include weights, biases, and activation functions. The set of parameters for all neurons in a DNN is called the DNN parameters (or neural network parameters). The weights and biases of neurons can be optimized through training, enabling the DNN to extract data features and express mapping relationships. DNNs generally use supervised or unsupervised learning strategies to optimize neural network parameters.

[0114] Depending on how the network is constructed, DNNs can include feedforward neural networks (FNNs), convolutional neural networks (CNNs), and recurrent neural networks (RNNs). Figure 2 The image shows a type of FNN, characterized by complete pairwise connections between neurons in adjacent layers. This makes FNNs typically require a large amount of storage space and result in high computational complexity.

[0115] like Figure 3 As shown, Figure 3This is a schematic diagram of a CNN. A CNN is a neural network specifically designed to process data with a grid-like structure. For example, time series data (discrete sampling along the time axis) and image data (two-dimensional discrete sampling) can both be considered as grid-like data. CNNs do not use all the input information at once for computation; instead, they use a fixed-size window to extract a portion of the information for convolution operations, which significantly reduces the computational cost of the neural network parameters. Furthermore, depending on the type of information extracted by the window (such as people and objects in an image representing different types of information), each window can use different convolution kernels, allowing CNNs to better extract features from the input data. Specifically, convolutional layers are used for feature extraction, resulting in feature maps. Pooling layers compress the input feature maps, reducing their size and simplifying the network's computational complexity. Fully connected layers map the learned "distributed feature representations" to the sample label space.

[0116] like Figure 4 The diagram shows a schematic of an RNN. An RNN is a type of DNN that utilizes feedback time-series information. Its input includes the current input value and its own output value from the previous time step. RNNs are suitable for acquiring temporally correlated sequence features, and are particularly applicable to applications such as speech recognition and channel coding / decoding. (Reference) Figure 4 A neuron can produce multiple outputs from inputs at multiple time points. For example, at time 0, the inputs are x0 and s0, and the outputs are y0 and s1; at time 1, the inputs are x1 and s1, and the outputs are y1 and s2, and so on, until at time t, the input is x... t and s t The output is y t and s t+1 .

[0117] The aforementioned FNN, CNN, and RNN are common neural network structures, all of which are constructed based on neurons.

[0118] Thanks to the advantages of machine learning in modeling and extracting information features, machine learning-based communication schemes can be designed, achieving good performance. These include CSI compressed feedback, adaptive constellation point design, and robust precoding. These schemes optimize transmission performance or reduce processing complexity by replacing the transmitting or receiving modules in the original communication system with neural network models. To support different application scenarios, different neural network model information can be predefined or configured, allowing the neural network model to adapt to the requirements of different scenarios. Neural network models typically have a large number of neural network parameters; for example, a layer in a fully connected neural network with 100 inputs and 100 outputs corresponds to 10^100 neural network parameters. Therefore, it is necessary to minimize the signaling overhead of storing and forwarding neural network parameters.

[0119] In one possible implementation, for each different wireless system parameter (including one or more of the following: wireless channel type, bandwidth, number of receiving antennas, number of transmitting antennas, modulation order, number of paired users, channel coding method, and coding code rate), a corresponding set of neural network model information (including neural network structure information and neural network parameters) is defined. Taking the design of an adaptive modulation constellation based on artificial intelligence (AI) as an example, when the number of transmitting antennas is 1 and the number of receiving antennas is 2, one set of neural network model information is needed to generate the corresponding modulation constellation; when the number of transmitting antennas is 1 and the number of receiving antennas is 4, another set of neural network model information is needed to generate the corresponding modulation constellation. Similarly, the corresponding neural network model information may differ for different wireless channel types, bandwidths, modulation orders, number of paired users, channel coding methods, and / or coding code rates.

[0120] For each different wireless system parameter, configuring a corresponding neural network model can achieve the best performance, but it also has the disadvantage of increased storage and signaling resource overhead. For example, if a neural network has 3, 4, and 2 different configurations for the receiving antenna, modulation order, and coding rate, respectively, then the number of neural networks that need to be pre-stored or transmitted in the air interface is 3*4*2, a total of 24, which consumes a lot of storage space or air interface signaling overhead.

[0121] In summary, when using neural networks in communication systems, different wireless system parameters lead to variations in the input and output dimensions, structure, and weights of the neural network. This results in network devices or terminal devices consuming significant storage space to store neural network model information (including neural network structure and parameters), or incurring substantial air interface signaling overhead to transmit neural network model information to terminal devices. Therefore, reducing the storage overhead of neural network model information or the signaling overhead during transmission is a pressing issue that needs to be addressed.

[0122] like Figure 5 The diagram illustrates the network architecture applicable to embodiments of this application. The communication system in this application can be a system including network devices and terminal devices, or a system including two or more terminal devices. In a communication system including network devices and terminal devices, the network device can send configuration information to the terminal device, and the terminal device performs corresponding configuration based on the configuration information. The network device can send downlink data to the terminal device, and the terminal device can send uplink data to the network device. In a communication system including two or more terminal devices (such as a vehicle-to-everything (V2X) network), terminal device 1 can send configuration information to terminal device 2, and terminal device 2 performs corresponding configuration based on the configuration information. Terminal device 1 can send data to terminal device 2, and terminal device 2 can also send data to terminal device 1.

[0123] The terminal device involved in the embodiments of this application can also be called a terminal, which can be a device with wireless transceiver capabilities. This terminal device can be deployed on land, including indoors or outdoors, handheld or vehicle-mounted; it can also be deployed on water (such as ships); and it can also be deployed in the air (such as airplanes, balloons, and satellites). The terminal device can be user equipment (UE), where UE includes handheld devices, vehicle-mounted devices, wearable devices, or computing devices with wireless communication capabilities. For example, the UE can be a mobile phone, tablet computer, or computer with wireless transceiver capabilities. The terminal device can also be a virtual reality (VR) terminal device, an augmented reality (AR) terminal device, a wireless terminal in industrial control, a wireless terminal in autonomous driving, a wireless terminal in telemedicine, a wireless terminal in a smart grid, a wireless terminal in a smart city, a wireless terminal in a smart home, etc. In the embodiments of this application, the device used to implement the terminal's functions can be the terminal itself; it can also be a device capable of supporting the terminal in implementing these functions, such as a chip system, which can be installed in the terminal or used in conjunction with the terminal. In this embodiment of the application, the chip system may be composed of chips or may include chips and other discrete devices.

[0124] The network devices involved in this application include access network devices, such as base stations (BS). A base station can be a device deployed in a wireless access network that can wirelessly communicate with terminals. Base stations may take various forms, such as macro base stations, micro base stations, relay stations, and access points. For example, the base station involved in this application can be a 5G base station or an evolved node B (eNB) in long-term evolution (LTE). A 5G base station can also be called a transmission reception point (TRP) or a 5G base station (next-generation node B, gNB). In this application, the apparatus for implementing the functions of the network device can be the network device itself; it can also be an apparatus capable of supporting the network device in implementing these functions, such as a chip system, which can be installed in the network device or used in conjunction with the network device.

[0125] The technical solutions provided in this application can be applied to various communication systems, such as LTE systems, 5G systems, wireless-fidelity (WiFi) systems, future sixth-generation mobile communication systems, or systems integrating multiple communication systems, etc. This application does not limit the applications. 5G can also be referred to as new radio (NR).

[0126] The technical solutions provided in this application can be applied to various communication scenarios, such as one or more of the following: eMBB communication, URLLC, machine-type communication (MTC), mMTC, device-to-device (D2D) communication, vehicle-to-everything (V2X) communication, vehicle-to-vehicle (V2V) communication, and the Internet of Things (IoT). In this application, the term "communication" can also be described as "transmission," "information transmission," "data transmission," or "signal transmission," etc. Transmission can include sending and / or receiving. In this application, the technical solution is described using communication between network devices and terminal devices as an example. Those skilled in the art can also use this technical solution for communication between other scheduling entities and subordinate entities, such as communication between macro base stations and micro base stations, and / or communication between a first terminal device and a second terminal device.

[0127] To address the storage overhead of neural network model information or the signaling overhead during transmission, this application provides a solution as follows: For different wireless system parameters, a multi-layered cascaded neural network structure including a common layer and dedicated layers is employed. The neural network model information for the common layer corresponds to the same information for multiple different wireless system parameters, while the neural network model information for the dedicated layers corresponds to different information. Furthermore, the neural network model information for the common layer can be predefined by the protocol or pre-configured to the terminal device by the network device. When the wireless system parameters change, the network device notifies the terminal device to update the neural network model information for the dedicated layers.

[0128] In this embodiment of the application, in a neural network-based communication system, some communication modules of the network device and / or terminal device can adopt a neural network model. For example... Figure 6The diagram shows a neural network architecture for joint transmit / receive optimization of constellation modulation / demodulation. Both the transmitting end's constellation modulation module and the receiving end's constellation demodulation module employ neural network models. The transmitting end's constellation neural network modulation module maps the bitstream to constellation symbols, while the constellation neural network demodulation module demaps the received constellation symbols to the log-likelihood ratio of the bit information. By training the neural network with collected channel data, optimal end-to-end communication performance can be achieved. In a communication system, the transmitting end can modulate the bitstream to be transmitted, obtaining modulation symbols, and then send these modulation symbols to the receiving end. When sending modulation symbols, the transmitting end can process them and send the processed signal to the receiving end. For example, this might involve one or more of the following operations: layer mapping, precoding, orthogonal frequency division multiplexing (OFDM) modulation, beamforming, and up-conversion. For ease of description, Figure 6 No further operations on the modulation symbols are listed in the document.

[0129] It should be noted that the communication method provided in this application is not limited to a specific scenario, such as CSI compressed feedback, adaptive constellation point design, robust precoding, etc., but can be applied to any neural network scenario.

[0130] In this embodiment, the neural network model at the receiving or transmitting end has a multi-layered cascaded network structure, including a common layer and dedicated layers. Each of the receiving and transmitting neural networks possesses a set of common layer model information (or common layer network information), which specifically includes common layer network structure information and common layer network parameters. For multiple different wireless system parameters, the common layer model information of both the receiving and transmitting ends remains unchanged. For each of these multiple different wireless system parameters, each of the receiving and transmitting neural networks possesses dedicated layer model information specific to that wireless system parameter. The dedicated layer model information specifically includes dedicated layer network structure information and dedicated layer network parameters. This embodiment can be described using a single common layer as an example. However, when applying the method of this embodiment in a communication system, the communication system may have one or more common layers, and the method described in this embodiment can be used separately for each scheme related to each common layer.

[0131] As one implementation method, multiple different wireless system parameters can be grouped into one or more groups, with each group containing one or more wireless system parameters. For a group of wireless system parameters, the group corresponds to the same common layer, and each wireless system parameter within the group corresponds to its own specific dedicated layer. The common layers of different groups of wireless parameters are designed independently; the common layers of different groups of wireless parameters can be the same or different, without restriction. Furthermore, different groups of wireless parameters can be for different receivers; for example, each receiver may correspond to one group of wireless parameters, or different groups of wireless parameters may be for different wireless parameters of the same receiver, without restriction.

[0132] Both the transmitting and receiving ends can possess multiple sets of dedicated layer model information (or dedicated layer network information). This dedicated layer model information specifically includes dedicated layer network structure information and dedicated layer network parameters. For different wireless system parameters, the transmitting and receiving ends can switch between using one or more of these dedicated layer model information sets. The common layer network structure can contain one or more neural networks. The common layer network structure can contain one or more of FNN, CNN, RNN, and other types of neural network structures. The dedicated layer network structure can contain one or more of FNN, CNN, RNN, and other types of neural network structures. Various cascading methods can be used between the common and dedicated layers.

[0133] As described above, the neural network model provided in this application embodiment includes a common layer and a dedicated layer. The parameters of this neural network model include common layer model information, dedicated layer model information, and neural network cascade structure information. Figure 7 The diagram shown is an example of a cascaded form of a neural network provided in an embodiment of this application. Figure 7 As shown in (a) above, the common layer can precede the dedicated layer, meaning the output of the common layer serves as the input to the dedicated layer. For example... Figure 7 As shown in (b) above, a common layer can also follow a dedicated layer, meaning the input to the common layer is the output of the dedicated layer. Figure 7 As shown in (c), the common layer can also be used between different dedicated layers; that is, the output of one dedicated layer can be used as the input of the common layer, and the output of the common layer can be used as the input of another dedicated layer. For example... Figure 7 As shown in (d), the common layer can also be connected in parallel with the dedicated layers between the dedicated layers. That is, the output of the first dedicated layer serves as the input of the common layer and the second dedicated layer, and the outputs of the common layer and the second dedicated layer serve as the input of the third dedicated layer. Figure 7 As shown in (e), a dedicated layer can also be used between different common layers, meaning the output of one part of the common layers can be used as the input of a dedicated layer, and the output of a dedicated layer can be used as the input of another part of the common layers. For example... Figure 7As shown in (f), the dedicated layer can also be connected in parallel with the common layer between the common layers, that is, the output of the first part of the common layer serves as the input of the dedicated layer and the second part of the common layer, and the output of the dedicated layer and the second part of the common layer serves as the input of the third part of the common layer.

[0134] For example, in the embodiments of this application, the following terms have the following meanings:

[0135] 1) Information on the cascaded structure of neural networks.

[0136] Neural network cascade structure information refers to the connection structure information between the specialized layers and common layers contained in a neural network. Specifically, it includes the interconnections between neurons in specialized layers and neurons in common layers. The neural network cascade structure indicated by this information can be... Figure 7 Any one of the six different cascade structures shown.

[0137] 2) Candidate set of information for neural network cascade structures.

[0138] The candidate set of neural network cascade structure information contains multiple neural network cascade structure information, and one neural network cascade structure information indicates a neural network cascade structure.

[0139] In the embodiments of this application, when describing one or more features A, such as neural network cascade structure information, dedicated layer network structure information, dedicated layer network parameters, common layer network structure information, and / or common layer network parameters, the one or more features A can be referred to as at least one feature A, or a set of features A, etc. The one or more features A can be represented in various forms such as a set, list, subset, or element in a set, without limitation. When configuring the feature A to be used from the one or more features A, the one or more features A can also be referred to as one or more candidate features A, at least one candidate feature A, a candidate set of features A, or a set of candidate features A, etc, without limitation.

[0140] 3) Dedicated layer network structure information.

[0141] The information about the network structure of a dedicated layer includes: the number of dedicated layers and the inter-layer structure within the dedicated layer (i.e., the connection method between neurons within the dedicated layer).

[0142] 4) Candidate set of network structure information for dedicated layer.

[0143] The candidate set of dedicated layer network structure information contains multiple dedicated layer network structure information.

[0144] 5) Dedicated layer network parameters.

[0145] The parameters of a dedicated layer network include: the weights, biases, and activation functions of each neuron in each dedicated layer.

[0146] 6) Candidate set of network parameters for the dedicated layer.

[0147] The candidate set of dedicated layer network parameters contains multiple dedicated layer network parameters.

[0148] 7) Common layer network structure information.

[0149] The network structure information of the common layer includes: the number of layers in the common layer and the inter-layer structure within the common layer (i.e., the connection method between neurons within the common layer).

[0150] 8) Candidate set of network structure information for the common layer.

[0151] The candidate set of common layer network structure information contains multiple common layer network structure information.

[0152] 9) Common layer network parameters.

[0153] The parameters of the common layer network include: the weights, biases, and activation functions of each neuron in each layer of the common layer.

[0154] 10) Candidate set of network parameters for the common layer.

[0155] The candidate set of common layer network parameters contains multiple common layer network parameters.

[0156] It should be noted that in the embodiments of this application, the dedicated layer is also referred to as the dedicated layer network, the common layer as the common layer network, and the neural network model information as the neural network model information.

[0157] To reduce the storage overhead of neural network model information or the signaling overhead during transmission, as shown in Figure 8(a), this application embodiment provides a schematic diagram of a communication method. This method can be executed by a first device or a chip used in the first device on the neural network model configuration side, and by a second device or a chip used in the second device on the neural network model usage side. In one scenario, the first device is a network device (such as a base station), and the second device is a terminal device (such as a UE). In another application scenario, the first device is a terminal device (such as a UE), and the second device is another terminal device (such as a UE). In yet another application scenario, the first device is an artificial intelligence device used to implement artificial intelligence technology, and the second device is a network device. In yet another application scenario, the first device is an artificial intelligence device, and the second device is a terminal device (such as a UE). The following explanation uses the execution of this communication method between the first device and the second device as an example.

[0158] In this communication method, multiple neural network models of the second device can share one or more common layers, and each can independently use a dedicated layer. That is, common layers can be shared, but dedicated layers can only be used exclusively.

[0159] The method includes the following steps:

[0160] Step 801a: The first device determines the instruction information.

[0161] The instruction information is used to indicate information about a first neural network model, which includes a dedicated layer network and a common layer network. The dedicated layer network is dedicated to the first neural network model, and the common layer network is a common network between the first neural network model and the second neural network model.

[0162] The second neural network model refers to a neural network model other than the first neural network model, or it can be described as a second neural network model that differs from the first neural network model. This second neural network model does not specifically refer to any particular neural network model. For example, both the first and second neural network models are neural network models for a second device. As another example, the first neural network model is a neural network model for a second device, and the second neural network model is a neural network model for a third device, where the third device is different from the second device. For example, the third device and the second device are different terminals.

[0163] In step 802a, the first device sends the indication information to the second device. Accordingly, the second device receives the indication information.

[0164] For example, the second device receives the indication information; or the second device obtains the indication information through other received signals, such as by processing the other signals (e.g., mathematical calculations, physical layer processing, or extraction). In this embodiment, the methods for obtaining other information are similar and will not be described in detail here.

[0165] Step 803a: The second device determines the information of the first neural network model based on the instruction information.

[0166] After determining the information of the first neural network model, the second device can either determine a new first neural network model based on that information, or update an existing first neural network model. Specifically, it can use the information of the first neural network model to initially obtain a new model, or replace an existing or currently used first neural network model. Alternatively, it can use the information of the first neural network model to perform online training on an existing or currently used first neural network model, obtaining an updated model. Optionally, before updating the existing or currently used first neural network model, the second device further determines that the performance of the first neural network model corresponding to the information of the first neural network model is superior to the performance of the existing or currently used first neural network model. That is, the second device only updates the existing or currently used first neural network model after determining that the performance of the first neural network model corresponding to the information of the first neural network model is superior to the performance of the existing or currently used first neural network model, thus avoiding a performance degradation of the first neural network model after updating it.

[0167] Based on the above scheme, the neural network model includes a common layer network and a dedicated layer network. The common layer network is a shared network for different neural network models, thereby reducing the overall overhead of information indicating the neural network model. Furthermore, the amount of neural network model information pre-stored by users (such as terminal devices) is also reduced accordingly, thus decreasing the storage overhead of the neural network model information.

[0168] As one possible implementation method, the information of the first neural network model indicated by the above-mentioned indication information includes at least one of the following: neural network cascade structure information, common layer network structure information, common layer network parameters, dedicated layer network structure information, and dedicated layer network parameters.

[0169] When the aforementioned indication information of the first neural network model includes information about the neural network cascade structure, this indication information or the information of the first neural network model can indicate the neural network cascade structure, or be used to indicate the configured neural network cascade structure from a candidate set of neural network cascade structures. That is, the neural network cascade structure information can directly indicate the neural network cascade structure, or indicate a neural network cascade structure from a candidate set of neural network cascade structures, thereby achieving flexible indication of the neural network cascade structure.

[0170] When the aforementioned indication information of the first neural network model includes common layer network structure information, this indication information or the information of the first neural network model can indicate the common layer network structure, or be used to indicate the configured common layer network structure from the candidate set of common layer network structures. That is, the common layer network structure information can directly indicate the common layer network structure, or indicate a common layer network structure from the candidate set of common layer network structures, thereby achieving flexible indication of the common layer network structure.

[0171] When the aforementioned indication information of the first neural network model includes common layer network parameter information, this indication information or the information of the first neural network model can indicate the common layer network parameters, or be used to indicate the configured common layer network parameters from the common layer network parameter candidate set. That is, the common layer network parameter information can directly indicate the common layer network parameters, or indicate a common layer network parameter from the common layer network parameter candidate set, thereby achieving flexible indication of the common layer network parameters.

[0172] When the aforementioned indication information of the first neural network model includes information about the dedicated layer network structure, this indication information or the information of the first neural network model can indicate the dedicated layer network structure, or be used to indicate the configured dedicated layer network structure from a set of dedicated layer network structure candidates. That is, the dedicated layer network structure information can directly indicate the dedicated layer network structure, or indicate a dedicated layer network structure from a set of dedicated layer network structure candidates, thereby achieving flexible indication of the dedicated layer network structure.

[0173] When the aforementioned indication information of the first neural network model includes dedicated layer network parameter information, this indication information or the information of the first neural network model can indicate dedicated layer network parameters, or be used to indicate the configured dedicated layer network parameters from a set of dedicated layer network parameter candidates. That is, the dedicated layer network parameter information can directly indicate dedicated layer network parameters, or indicate a dedicated layer network parameter from a set of dedicated layer network parameter candidates, thereby achieving flexible indication of dedicated layer network parameters.

[0174] The following describes how to implement the above-mentioned instruction information.

[0175] In one implementation, the aforementioned indication information carries an index of the neural network cascade structure, an index of the dedicated layer network structure, and an index of the dedicated layer network parameters. Optionally, the common layer network structure and common layer network parameters are predefined by the protocol or pre-configured by the first device on the second device.

[0176] Based on the above implementation method, the index of the neural network cascade structure information is used to indicate the neural network cascade structure in the configured neural network cascade structure candidate set. That is, the neural network cascade structure determined by the first device is a neural network cascade structure in the protocol-predefined or pre-configured neural network cascade structure candidate set. Then, the first device indicates the determined neural network cascade structure through this indication information. The index of the dedicated layer network structure information is used to indicate the dedicated layer network structure information in the configured dedicated layer network structure information candidate set. That is, the dedicated layer network structure information determined by the first device is a dedicated layer network structure information in the protocol-predefined or pre-configured dedicated layer network structure information candidate set. Then, the first device indicates the determined dedicated layer network structure information through this indication information. The index of the dedicated layer network parameters is used to indicate the dedicated layer network parameters in the configured dedicated layer network parameter candidate set. That is, the dedicated layer network parameters determined by the first device are a dedicated layer network parameter in the protocol-predefined or pre-configured dedicated layer network parameter candidate set. Then, the first device indicates the determined dedicated layer network parameter through this indication information.

[0177] As a second implementation, the aforementioned indication information carries an index of the neural network cascade structure, an index of the dedicated layer network structure, and dedicated layer network parameters. Optionally, the common layer network structure and common layer network parameters are predefined by the protocol or pre-configured by the first device on the second device.

[0178] Based on the above implementation method, the indices for the neural network cascade structure information and the dedicated layer network structure information are the same as in the first implementation method described above, and will not be repeated here. The dedicated layer network parameters can be obtained by the first device through training, and then the first device indicates the dedicated layer network parameters through this indication information.

[0179] As a third implementation method, the aforementioned indication information carries an index of the neural network cascade structure, information about the dedicated layer network structure, and parameters of the dedicated layer network. Optionally, the common layer network structure and parameters are predefined by the protocol or pre-configured by the first device on the second device.

[0180] Based on the above implementation method, the index of the neural network cascade structure information is the same as in the first implementation method, and will not be repeated here. The dedicated layer network structure information can be obtained by the first device through training, and then the first device indicates the dedicated layer network structure information through this indication information. The dedicated layer network parameters can be obtained by the first device through training, and then the first device indicates the dedicated layer network parameters through this indication information.

[0181] As one implementation method, prior to step 801a above, the first device further sends configuration information to the second device. This configuration information is used to indicate one or more of the following: a candidate set of neural network cascade structure information, a candidate set of common layer network structure information, a candidate set of common layer network parameters, a candidate set of dedicated layer network structure information, and a candidate set of dedicated layer network parameters. Based on this scheme, the first device sends configuration information to the second device in advance to configure one or more of the following: a candidate set of neural network cascade structure information, a candidate set of common layer network structure information, a candidate set of common layer network parameters, a candidate set of dedicated layer network structure information, and a candidate set of dedicated layer network parameters. Subsequently, the first device can indicate the index corresponding to the information in the candidate set through the indication information, thereby reducing the overhead of the indication information.

[0182] As one possible implementation, prior to step 801a, the second device further sends auxiliary information to the first device. This auxiliary information indicates the wireless system parameters between the first and second devices. The first device can then determine, based on this auxiliary information, that the neural network model information of the second device needs to be updated. Based on this scheme, the second device sends auxiliary information to the first device, enabling the first device to make decisions regarding the parameters and structure information corresponding to the neural network model based on this auxiliary information, which helps improve the accuracy of the neural network model.

[0183] As one possible implementation, prior to step 801a above, the second device further sends capability information of the second device to the first device, which indicates one or more of the following information of the second device:

[0184] 1) Does it support using neural networks to replace or implement the functions of the communication module?

[0185] The communication modules here include, but are not limited to: OFDM modulation module, OFDM demodulation module, constellation mapping module, constellation demapping module, channel coding module, and / or channel decoding module.

[0186] 2) Does it support neural network structures where public layer networks and private layer networks are cascaded?

[0187] 3) Does it support receiving dedicated layer network structure information and / or dedicated layer network parameters via signaling?

[0188] 4) Stored protocol-predefined or pre-configured dedicated layer network structure information;

[0189] 5) Stored protocol predefined or preconfigured dedicated layer network parameters;

[0190] 6) Memory space that can be used to store information about the cascaded structure of neural networks, information about the structure of common layers, parameters of common layers, information about the structure of private layers and / or parameters of private layers;

[0191] 7) Computing power information that can be used to run neural networks.

[0192] The computing power information here refers to the computing capabilities of running a neural network, such as the processor's processing speed and / or the amount of data the processor can process.

[0193] Based on the above scheme, the second device sends capability information to the first device, enabling the first device to make decisions on the parameters and structural information corresponding to the neural network model based on the capability information, as well as the method of sending decision instruction information, which helps to improve the accuracy of the neural network model and reduce the overhead of sending instruction information.

[0194] As shown in Figure 8(b), this application provides another schematic diagram of a communication method. The execution subject of this method is similar to that described in Figure 8(a) above, and will not be repeated here. The following description uses the execution of this communication method between a first device and a second device as an example.

[0195] In this communication method, in one possible scenario, the neural network models of multiple second devices can share the same one or more common layers, while each second device can independently use one or more dedicated layers. That is, common layers can be shared by multiple second devices, but dedicated layers can only be used by each second device exclusively.

[0196] The method includes the following steps:

[0197] Step 801b: The first device sends first instruction information to the second device. The first instruction information is used to indicate the common information of the first neural network model. The first neural network model includes a dedicated layer network and a common layer network. The common information of the first neural network model includes the information of the common layer network.

[0198] Step 802b: The first device sends a second indication message to the second device, the second indication message being used to indicate information about the private layer network.

[0199] The information of the common layer network includes common layer network structure information and common layer network parameters. The first indication information can indicate the common layer network structure information and / or the common layer network parameters. When the first indication information indicates common layer network structure information, it can directly indicate the common layer network structure, or indicate the index of the common layer network structure information from the candidate set of common layer network structure information. When the first indication information indicates common layer network parameters, it can directly indicate the common layer network parameters, or indicate the index of the common layer network parameters from the candidate set of common layer network parameters.

[0200] The information of the dedicated layer network includes dedicated layer network structure information and dedicated layer network parameters. The second indication information can indicate the dedicated layer network structure information and / or indicate the dedicated layer network parameters. When the second indication information indicates dedicated layer network structure information, it can directly indicate the dedicated layer network structure, or indicate the index of the dedicated layer network structure information from the dedicated layer network structure information candidate set. When the second indication information indicates dedicated layer network parameters, it can directly indicate the dedicated layer network parameters, or indicate the index of the dedicated layer network parameters from the dedicated layer network parameter candidate set.

[0201] After the second device acquires (e.g., receives) the first instruction information and the second instruction information, it can determine the information of the first neural network model based on the first instruction information and the second instruction information. Then, based on the information of the first neural network model, it can obtain the first neural network model, or it can update the first neural network model that is in use or that is already in use. Specific details are similar to the corresponding description in Figure 8(a), and will not be repeated here.

[0202] Based on the above scheme, the neural network model is divided into a common layer network and a dedicated layer network. The common layer network is a shared network for the neural network models of different second devices (such as terminal devices), while the dedicated layer network is a network used independently by the neural network model of each second device. This reduces the overall overhead of the information indicating the neural network model. Furthermore, the amount of information about the neural network model pre-stored by the second device is also reduced accordingly, thereby reducing the storage overhead of the neural network model information.

[0203] As one possible implementation method, the first indication information is carried through the MIB, system message, or public message.

[0204] As one possible implementation, the second indication information is carried by the second device-specific RRC signaling, MAC CE, or dedicated DCI.

[0205] The communication method described above will be explained in detail below with specific examples.

[0206] In Examples 1 to 6, the protocol predefines or the network device pre-configures common layer model information for the terminal device, specifically including common layer network structure information and common layer network parameters. Therefore, each time the neural network model information is updated, it is not necessary to update the common layer network structure information and common layer network parameters; in other words, only the neural network cascade structure information, dedicated layer network structure information, and dedicated layer network parameters need to be updated. Alternatively, the update cycle for the common layer network structure information and common layer network parameters can be set to a relatively long period, updating the common layer network structure information and common layer network parameters when the update cycle is reached. Or, the common layer network structure information and common layer network parameters can be almost never updated for a considerable period of time.

[0207] Furthermore, the explanation will be based on the example of the first device being a network device and the second device being a terminal device.

[0208] Example 1

[0209] like Figure 9 The diagram shown illustrates another communication method provided in an embodiment of this application. In this embodiment, the following information is predefined by the protocol:

[0210] 1) Candidate set of information for cascaded neural network structures;

[0211] 2) Common layer network structure information;

[0212] 3) Common layer network parameters;

[0213] 4) Candidate set of network structure information for the dedicated layer; and,

[0214] 5) Candidate set of network parameters for the dedicated layer.

[0215] In one possible implementation, the network device determines the neural network model information (including neural network cascade structure information, dedicated layer network structure information, and dedicated layer network parameters), and then instructs the terminal device on the neural network model information through indication information. This method can occur when the network device determines that the wireless system parameters have changed, when the network device initially configures the neural network model information for the terminal device, or when the network device updates the neural network model information; there are no restrictions on this. Specifically, the neural network cascade structure information is selected from a protocol-predefined candidate set of neural network cascade structure information; the dedicated layer network structure information can be selected from the protocol-predefined candidate set of dedicated layer network structure information or generated by the network device during online training; and the dedicated layer network parameters can be selected from the protocol-predefined candidate set of dedicated layer network parameters or generated by the network device during online training. The generation and determination methods of each variable in this embodiment are shown in Table 1.

[0216] Table 1

[0217]

[0218] like Figure 9 As shown, this embodiment includes the following steps:

[0219] Step 901: The terminal device sends capability information to the network device. Correspondingly, the network device receives the capability information.

[0220] Terminal devices can report their capability information to network devices. This capability information includes one or more of the following: whether the terminal device supports using neural networks to replace or implement the functions of some communication modules; whether it supports a neural network structure that cascades a public layer network and a private layer network; whether it stores protocol-predefined private layer network structure information and private layer network parameters; whether it supports receiving private layer network structure information and private layer network parameters via signaling; and computing power information that can be used to run neural networks.

[0221] Optionally, the network device will only notify the terminal device of relevant information about the dedicated layer network if the terminal device supports a neural network structure in which the public layer network and the dedicated layer network are cascaded.

[0222] Step 901 is optional. When step 901 is not performed, the network device can always notify the terminal device of relevant information about the terminal device's private layer network without referring to the terminal device's capability information.

[0223] In step 902, the terminal device sends auxiliary information to the network device. Accordingly, the network device acquires (e.g., receives) the auxiliary information.

[0224] The auxiliary information reported by the terminal device is used to help the network device determine whether the wireless system parameters have changed.

[0225] In the first implementation method, the auxiliary information includes information that directly characterizes wireless system parameters, such as bandwidth, the number of antennas of the terminal device, modulation order, and / or coding rate. In the second implementation method, the auxiliary information includes information that indirectly characterizes wireless system parameters, such as CSI, the moving speed of the terminal device, the geographical location of the terminal device, bit error rate, and estimated signal-to-noise ratio. In the third implementation method, the auxiliary information includes neural network information of the terminal device, such as information indicating the dedicated layer network structure and parameters that the terminal device is using or has stored.

[0226] Step 903: The network device determines whether the wireless system parameters have changed.

[0227] If the network device determines that the wireless system parameters have changed, it will execute step 904; otherwise, the process ends.

[0228] In one possible implementation, the network device can determine whether the wireless system parameters have changed based on auxiliary information reported by the terminal device. The determination method could be as follows: the network device determines the current wireless system parameters based on the auxiliary information reported by the terminal device, and then compares the determined current wireless system parameters with the historical wireless system parameters stored in the network device. If the change between the determined current wireless system parameters and the historical wireless system parameters stored in the network device is greater than a threshold value, then it is determined that the wireless system parameters have changed. Optionally, this threshold value is predefined.

[0229] In one possible implementation, the network device can measure the uplink channel radio system parameters and / or determine whether the radio system parameters have changed based on the configuration parameters of the terminal device known to the network device. The determination method is similar to the previous implementation and will not be described in detail.

[0230] Step 904: The network device determines whether the neural network model information of the terminal device needs to be configured.

[0231] The term "configuration" here can refer to initial configuration or updated configuration. For ease of explanation, this embodiment uses configuration as an example.

[0232] If the network device determines that the neural network model information of the terminal device needs to be configured, it will execute the following step 905; otherwise, the process ends.

[0233] If the network device determines that the wireless system parameters have changed, it further determines whether the neural network model information of the terminal device needs to be updated. Not all changes in wireless system parameters will trigger an update of the terminal device's neural network model information. For example, for small changes in channel delay spread, the terminal device's neural network model information may not need to be updated.

[0234] As one implementation method, network devices can approximate or classify the current wireless system parameters to obtain the current standard values ​​corresponding to the current wireless system parameters. Then, they compare these current standard values ​​with the historical standard values ​​corresponding to the historical wireless system parameters. If they are the same, it is determined that the neural network model information of the terminal device does not need to be updated; if they are different, it is determined that the neural network model information of the terminal device needs to be updated. Here, the historical wireless system parameters refer to the wireless system parameters corresponding to the neural network model information last notified to the terminal device by the network device.

[0235] For example, taking wireless system parameters including wireless channel type as an example, the predefined standard values ​​corresponding to the wireless channel types include: 30 nanoseconds (ns), 100 ns, and 300 ns. Specifically, channels with a latency spread less than 50 ns are classified as 30 ns wireless channel types, channels with a latency spread greater than or equal to 50 ns and less than 200 ns are classified as 100 ns wireless channel types, and channels with a latency spread greater than or equal to 200 ns are classified as 300 ns wireless channel types. For instance, if the network device determines that the current channel latency spread is 150 ns, then the current standard value corresponding to this current channel latency spread is 100 ns. If the historical standard value corresponding to the historical wireless system parameters is also 100 ns, then the network device determines that it does not need to update the neural network model information of the terminal device.

[0236] Step 905: The network device determines the updated neural network model information of the terminal device.

[0237] If the network device determines that the neural network model information of the terminal device needs to be updated, the network device will then continue to determine the updated neural network model information of the terminal device.

[0238] The updated neural network model information for the terminal device includes:

[0239] 1) Information on the cascaded neural network structure of the terminal device;

[0240] 2) Dedicated layer network structure information for terminal equipment; and,

[0241] 3) Dedicated layer network parameters for terminal equipment.

[0242] Based on the neural network model information generation methods shown in Table 1, network devices can determine the neural network model information of terminal devices according to either Method 1 or Method 2.

[0243] Method 1: The network device selects the neural network cascade structure information, dedicated layer network structure information, and dedicated layer network parameters corresponding to the current wireless system parameters of the terminal device from the protocol-predefined candidate sets of neural network cascade structure information, dedicated layer network structure information, and dedicated layer network parameter parameters.

[0244] For example, network devices store a mapping table between wireless system parameters and neural network cascade structure information, dedicated layer network structure information, and dedicated layer network parameters. The network device can look up the table based on the current wireless system parameters of the terminal device to obtain the neural network cascade structure information, dedicated layer network structure information, and dedicated layer network parameters corresponding to the current wireless system parameters of the terminal device.

[0245] In this method, the network structure information and parameters of the dedicated layer can be obtained through offline training.

[0246] It should be noted that in the mapping table, the information on the neural network cascade structure, the information on the dedicated layer network structure, and the parameters of the dedicated layer network can be represented using the indices of the neural network cascade structure information, the dedicated layer network structure information, and the dedicated layer network parameters, respectively. That is, the mapping table stores the mapping relationships between wireless system parameters and the indices of the neural network cascade structure information, the dedicated layer network structure information, and the dedicated layer network parameters.

[0247] Method 2: The network device selects the neural network cascade structure information corresponding to the current wireless system parameters of the terminal device from the candidate set of neural network cascade structure information predefined by the protocol, and obtains the dedicated layer network structure information and dedicated layer network parameters through online training (such as through machine learning or neural network training) based on the auxiliary information reported by the terminal device.

[0248] For example, network devices store a mapping table between wireless system parameters and neural network cascade structure information. The network device can look up the table based on the current wireless system parameters of the terminal device to obtain the neural network cascade structure information corresponding to the current wireless system parameters of the terminal device.

[0249] Furthermore, the network device also determines first indication information for indicating updated neural network model information.

[0250] Based on the above method one, the determined first indication information includes the following:

[0251] 1) Index of neural network cascade structure information of terminal devices;

[0252] 2) Index of dedicated layer network structure information for terminal equipment; and,

[0253] 3) Index of dedicated layer network parameters for terminal equipment.

[0254] Based on Method 1 above, the neural network cascade structure indicated by the index of the neural network cascade structure information of the terminal device can be... Figure 7 The dedicated layer network structure is one of the neural network cascade structures in the protocol, that is, one of the candidate sets of neural network cascade structure information predefined by the protocol. The dedicated layer network structure information indicated by the index of the dedicated layer network structure information of the terminal device can be one of the candidate sets of dedicated layer network structure information predefined by the protocol. The dedicated layer network parameters indicated by the index of the dedicated layer network parameters of the terminal device can be one of the candidate sets of dedicated layer network parameters predefined by the protocol.

[0255] Based on Method 2 above, if the dedicated layer network structure information and dedicated layer network parameters obtained through online training have corresponding indices in the mapping table, then the determined indication information can include the following information:

[0256] 1) Index of neural network cascade structure information of terminal devices;

[0257] 2) Indexes to the dedicated layer network structure information of terminal devices; and,

[0258] 3) Index of dedicated layer network parameters for terminal devices.

[0259] Based on Method 2 above, if the terminal device supports receiving dedicated layer network structure information and dedicated layer network parameters via signaling, the indication information determined by the network device can include the following information:

[0260] 1) Index of neural network cascade structure information of terminal devices;

[0261] 2) Dedicated layer network structure information of terminal equipment; and,

[0262] 3) Dedicated layer network parameters of terminal equipment.

[0263] It should be noted that if the terminal device supports receiving dedicated layer network structure information and dedicated layer network parameters via signaling, the indication information determined by the network device can carry either the dedicated layer network structure information and dedicated layer network parameters obtained through offline training, or the dedicated layer network structure information and dedicated layer network parameters obtained through online training. As one implementation method, the network device can test the dedicated layer network structure information and dedicated layer network parameters obtained through offline training and the dedicated layer network structure information and dedicated layer network parameters obtained through online training based on the auxiliary information reported by the terminal device, thereby selecting the dedicated layer network structure information and dedicated layer network parameters with better test results.

[0264] Step 906: The network device sends indication information to the terminal device. Accordingly, the terminal device obtains (e.g., receives) the indication information.

[0265] This instruction carries the following information:

[0266] 1) Index of neural network cascade structure information of terminal devices;

[0267] 2) Indexes to the dedicated layer network structure information of terminal devices; and,

[0268] 3) Index of dedicated layer network parameters for terminal devices.

[0269] Alternatively, the instruction message may carry the following information:

[0270] 1) Index of neural network cascade structure information of terminal devices;

[0271] 2) Dedicated layer network structure information of terminal equipment; and,

[0272] 3) Dedicated layer network parameters of terminal equipment.

[0273] Step 907: The terminal device determines the neural network model information.

[0274] Based on the instructions sent by the network device, the terminal device determines its neural network model information, including the cascaded structure information, dedicated layer network structure information, and dedicated layer network parameters. Then, the terminal device determines its own neural network based on this model information.

[0275] Optionally, in step 908, the terminal device determines whether to update the neural network model information of the terminal device.

[0276] The terminal device compares the new neural network (i.e. the neural network determined in step 907) with the neural network currently being used by the terminal device to determine whether to update the neural network model information currently being used by the terminal device based on the neural network model information sent by the network device.

[0277] One method for a terminal device to determine whether to update its neural network model information is as follows: The terminal device tests a new neural network and the locally used neural network based on the latest acquired auxiliary information, and then compares their performance. If the former performs better, the terminal device updates its neural network model information; otherwise, it does not update the neural network model information. For example, the terminal device uses newly acquired CSI to test constellation modulation / demodulation with both the new and the currently used neural networks. If the new neural network has a lower demodulation bit error rate, the terminal device updates its neural network based on the new neural network model information; otherwise, the terminal device continues to use the locally used neural network.

[0278] It should be noted that, as an alternative implementation, after step 906, steps 907 and 908 are not executed. Instead, step 907' is executed: the terminal device determines the neural network model information based on the instruction information, and trains new dedicated layer network structure information and new dedicated layer network parameters based on the neural network model information. That is, the terminal device uses the dedicated layer network structure information and dedicated layer network parameters indicated by the network device as reference information, and trains itself to obtain new dedicated layer network structure information and new dedicated layer network parameters through machine learning. The method for locally training the dedicated layer could be, for example, that the terminal device uses machine learning to train a neural network, based on locally collected wireless channel information, and the dedicated layer network structure information and dedicated layer network parameters received from the network device, to train the dedicated layer so that the bit error rate performance of the locally trained dedicated layer cascaded with the common layer approaches or exceeds the performance of the dedicated layer cascaded with the common layer.

[0279] Example 2

[0280] like Figure 10 The diagram shown illustrates another parameter indication method for a neural network provided in this application embodiment. In this embodiment, the following information is predefined by the protocol:

[0281] 1) Candidate set of information for cascaded neural network structures.

[0282] 2) Common layer network structure information.

[0283] 3) Common layer network parameters; and,

[0284] 4) Candidate set of network structure information for dedicated layer.

[0285] Compared to Embodiment 1 above, in this embodiment, the protocol does not predefine a candidate set of dedicated layer network parameters.

[0286] Similar to the description in Example 1, Table 2 shows the generation and determination methods of each variable in this example.

[0287] Table 2

[0288]

[0289] like Figure 10 As shown, this embodiment includes the following steps:

[0290] Step 1001: The terminal device sends capability information to the network device. Accordingly, the network device acquires (e.g., receives) the capability information.

[0291] Before the network device notifies the terminal device of the neural network to be used, the terminal device can report its capability information to the network device. This capability information includes one or more of the following: whether the terminal device supports using a neural network to replace or implement the functions of some communication modules, whether it supports a neural network structure that cascades a public layer network and a private layer network, whether it stores protocol-predefined private layer network structure information, whether it supports receiving private layer network structure information and private layer network parameters via signaling, and computing power information that can be used to run the neural network.

[0292] Optionally, the network device will only notify the terminal device of relevant information about the dedicated layer network if the terminal device supports a neural network structure in which the public layer network and the dedicated layer network are cascaded.

[0293] Step 1001 is optional. When step 1001 is not performed, the network device can always notify the terminal device of relevant information about the terminal device's dedicated layer without referring to the terminal device's capability information.

[0294] Steps 1002 to 1004 are the same. Figure 9 Steps 902 to 904 in the corresponding embodiment.

[0295] Step 1005: The network device determines the updated neural network model information of the terminal device.

[0296] If the network device determines that the neural network model information of the terminal device needs to be updated, the network device will then continue to determine the updated neural network model information of the terminal device.

[0297] The updated neural network model information for the terminal device includes:

[0298] 1) Information on the cascaded neural network structure of the terminal device;

[0299] 2) Dedicated layer network structure information for terminal equipment; and,

[0300] 3) Dedicated layer network parameters for terminal equipment.

[0301] Based on the neural network model information generation methods shown in Table 2, network devices can determine the neural network model information of terminal devices according to either Method 1 or Method 2.

[0302] Method 1: The network device selects the neural network cascade structure information and the dedicated layer network structure information corresponding to the current wireless system parameters of the terminal device from the protocol-predefined candidate set of neural network cascade structure information and candidate set of dedicated layer network structure information, respectively. Based on the auxiliary information reported by the terminal device, the dedicated layer network parameters are obtained through online training (such as through machine learning or neural network training).

[0303] For example, network devices store a mapping table between wireless system parameters and neural network cascade structure information and dedicated layer network structure information. The network device can look up the table based on the current wireless system parameters of the terminal device to obtain the neural network cascade structure information and dedicated layer network structure information corresponding to the current wireless system parameters of the terminal device.

[0304] In this method, the network structure information of the dedicated layer can be obtained through offline training.

[0305] It should be noted that in the mapping table, the information on the neural network cascade structure and the information on the dedicated layer network structure can be represented using the indices of the neural network cascade structure information and the dedicated layer network structure information, respectively. That is, the mapping table can store indices of wireless system parameters and neural network cascade structure information, indices of dedicated layer network structure information, and indices of dedicated layer network parameters.

[0306] Method 2: The network device selects the neural network cascade structure information corresponding to the current wireless system parameters of the terminal device from the candidate set of neural network cascade structure information predefined by the protocol, and obtains the dedicated layer network structure information and dedicated layer network parameters through online training (such as through machine learning or neural network training) based on the auxiliary information reported by the terminal device.

[0307] For example, network devices store a mapping table between wireless system parameters and neural network cascade structure information. The network device can look up the table based on the current wireless system parameters of the terminal device to obtain the neural network cascade structure information corresponding to the current wireless system parameters of the terminal device.

[0308] Furthermore, the network device also determines indication information for indicating updated neural network model information.

[0309] Based on Method 1 above, the determined indication information includes the following:

[0310] 1) Index of neural network cascade structure information of terminal devices;

[0311] 2) Index of dedicated layer network structure information for terminal equipment; and,

[0312] 3) Dedicated layer network parameters for terminal equipment.

[0313] Based on Method 1 above, the neural network cascade structure indicated by the index of the neural network cascade structure information of the terminal device can be... Figure 7This refers to a type of neural network cascade structure, specifically one of the candidate sets of neural network cascade structure information predefined by the protocol. The dedicated layer network structure information indicated by the index of the terminal device's dedicated layer network structure information can also be one of the candidate sets of dedicated layer network structure information predefined by the protocol.

[0314] Based on Method 2 above, if the dedicated layer network structure information obtained through online training has a corresponding index in the mapping table, then the determined indication information can include the following information:

[0315] 1) Index of neural network cascade structure information of terminal devices;

[0316] 2) Indexes to the dedicated layer network structure information of terminal devices; and,

[0317] 3) Dedicated layer network parameters of terminal equipment.

[0318] Based on Method 2 above, if the dedicated layer network structure information obtained through online training does not have a corresponding index in the mapping table, the indication information determined by the network device can include the following information:

[0319] 1) Index of neural network cascade structure information of terminal devices;

[0320] 2) Dedicated layer network structure information of terminal equipment;

[0321] 3) Dedicated layer network parameters of terminal equipment.

[0322] It should be noted that the indication information determined by the network device can carry either the dedicated layer network structure information obtained through offline training and the dedicated layer network parameters obtained through online training, or both. As one implementation method, the network device can test the dedicated layer network structure information obtained offline and the dedicated layer network parameters obtained online, as well as the dedicated layer network structure information obtained online, based on the auxiliary information reported by the terminal device, thereby selecting the dedicated layer network structure information and dedicated layer network parameters with the better test results.

[0323] It should be noted that if the network device can know the private layer network structure information and private layer network parameters currently being used by the terminal device, then after determining the latest private layer network structure information and private layer network parameters of the terminal device, the network device can also determine the amount of change between the latest private layer network parameters of the terminal device and the private layer network parameters currently being used. Therefore, it is not necessary to send the private layer network parameters of the terminal device to the terminal device, but to send the amount of change of the private layer network parameters of the terminal device.

[0324] Step 1006: The network device sends indication information to the terminal device. Accordingly, the terminal device obtains (e.g., receives) the indication information.

[0325] This instruction carries the following information:

[0326] 1) Index of neural network cascade structure information of terminal devices;

[0327] 2) Indexes to the dedicated layer network structure information of terminal devices; and,

[0328] 3) Dedicated layer network parameters of terminal equipment.

[0329] Alternatively, the instruction message may carry the following information:

[0330] 1) Index of neural network cascade structure information of terminal devices;

[0331] 2) Dedicated layer network structure information of terminal equipment; and,

[0332] 3) Dedicated layer network parameters of terminal equipment.

[0333] Step 1007: The terminal device determines the neural network model information.

[0334] Based on the instructions sent by the network device, the terminal device determines its neural network model information, including the cascaded structure information, dedicated layer network structure information, and dedicated layer network parameters. Then, the terminal device determines its own neural network based on this model information.

[0335] Step 1008, same as above. Figure 9 Step 908 in the corresponding embodiment.

[0336] It should be noted that, as an alternative implementation method, after step 1006, steps 1007 and 1008 are not executed, but step 907' is executed instead, as described above.

[0337] Example 3

[0338] like Figure 11 The diagram shown illustrates another parameter indication method for a neural network provided in this application embodiment. In this embodiment, the following information is predefined by the protocol:

[0339] 1) Candidate set of information for cascaded neural network structures.

[0340] 2) Common layer network structure information; and,

[0341] 3) Common layer network parameters.

[0342] Compared with the first embodiment described above, in this embodiment, the protocol does not predefine a candidate set of dedicated layer network structure information and a candidate set of dedicated layer network parameters.

[0343] Similar to Embodiment 1 above, the generation and determination methods of each variable in this embodiment are shown in Table 3.

[0344] Table 3

[0345]

[0346]

[0347] like Figure 11 As shown, this embodiment includes the following steps:

[0348] Step 1101: The terminal device sends capability information to the network device. Accordingly, the network device acquires (e.g., receives) the capability information.

[0349] Before the network device notifies the terminal device of the neural network to be used, the terminal device can report its capability information to the network device. This capability information includes one or more of the following: whether the terminal device supports using a neural network to replace or implement the functions of some communication modules, whether it supports a neural network structure that cascades a public layer network and a private layer network, whether it supports receiving private layer network structure information and private layer network parameters through signaling, and computing power information that can be used to run the neural network.

[0350] Optionally, the network device will only notify the terminal device of relevant information about the dedicated layer network if the terminal device supports a neural network structure in which the public layer network and the dedicated layer network are cascaded.

[0351] Step 1101 is optional. When step 1101 is not performed, the network device can always notify the terminal device of relevant information about the terminal device's dedicated layer without referring to the terminal device's capability information.

[0352] Steps 1102 to 1104 are the same. Figure 9 Steps 902 to 904 in the corresponding embodiment.

[0353] Step 1105: The network device determines the updated neural network model information of the terminal device.

[0354] If the network device determines that the neural network model information of the terminal device needs to be updated, the network device will then continue to determine the updated neural network model information of the terminal device.

[0355] The updated neural network model information for the terminal device includes:

[0356] 1) Information on the cascaded neural network structure of the terminal device;

[0357] 2) Dedicated layer network structure information for terminal equipment; and,

[0358] 3) Dedicated layer network parameters for terminal equipment.

[0359] Based on the neural network model information generation method shown in Table 3, the network device can determine the neural network model information of the terminal device according to the following method: The network device selects the neural network cascade structure information corresponding to the current wireless system parameters of the terminal device from the candidate set of neural network cascade structure information predefined by the protocol, and obtains the dedicated layer network structure information and dedicated layer network parameters through online training (such as through machine learning or neural network training) based on the auxiliary information reported by the terminal device.

[0360] For example, a network device might store a mapping table between wireless system parameters and neural network cascade structure information. The network device can look up the table based on the current wireless system parameters of the terminal device to obtain the corresponding neural network cascade structure information. It's important to note that the neural network cascade structure information in the mapping table can be represented using an index. In other words, the mapping table can store the mapping relationship between wireless system parameters and the indexes of neural network cascade structure information.

[0361] Furthermore, the network device also determines indication information for indicating updated neural network model information, which includes:

[0362] 1) Index of neural network cascade structure information of terminal devices;

[0363] 2) Dedicated layer network structure information for terminal equipment; and,

[0364] 3) Dedicated layer network parameters for terminal equipment.

[0365] Based on the above method, the neural network cascade structure indicated by the index of the neural network cascade structure information of the terminal device can be... Figure 7 It is one of the neural network cascade structures in the protocol, that is, one of the candidate sets of neural network cascade structure information predefined by the protocol.

[0366] Step 1106: The network device sends indication information to the terminal device. Accordingly, the terminal device obtains (e.g., receives) the indication information.

[0367] This instruction carries the following information:

[0368] 1) Index of neural network cascade structure information of terminal devices;

[0369] 2) Dedicated layer network structure information of terminal equipment; and,

[0370] 3) Dedicated layer network parameters of terminal equipment.

[0371] Step 1107: The terminal device determines the neural network model information.

[0372] Based on the instructions sent by the network device, the terminal device determines its neural network model information, including the cascaded structure information, dedicated layer network structure information, and dedicated layer network parameters. Then, the terminal device determines its own neural network based on this model information.

[0373] Step 1108, same as above. Figure 9 Step 908 in the corresponding embodiment.

[0374] It should be noted that, as an alternative implementation method, after step 1106, steps 1107 and 1108 are not executed, but step 907' is executed instead, as described above.

[0375] Example 4

[0376] like Figure 12 The diagram shown illustrates another parameter indication method for a neural network provided in this application embodiment. In this embodiment, the network device pre-configures the following information to the terminal device (e.g., the network device pre-configures the information to the terminal device via signaling or system messages):

[0377] 1) Candidate set of information for cascaded neural network structures;

[0378] 2) Common layer network structure information;

[0379] 3) Common layer network parameters;

[0380] 4) Candidate set of network structure information for the dedicated layer; and,

[0381] 5) Candidate set of network parameters for the dedicated layer.

[0382] Similar to Embodiment 1 above, the generation and determination methods of each variable in this embodiment are shown in Table 4.

[0383] Table 4

[0384]

[0385]

[0386] like Figure 12 As shown, this embodiment includes the following steps:

[0387] Step 1201: The terminal device sends capability information to the network device. Accordingly, the network device acquires (e.g., receives) the capability information.

[0388] Terminal devices can report their capability information to network devices. This capability information includes one or more of the following: whether the terminal device supports using neural networks to replace or implement the functions of some communication modules; whether it supports a neural network structure that cascades a common layer network and a dedicated layer network; whether it supports receiving dedicated layer network structure information and dedicated layer network parameters via signaling; whether it has memory space available for storing neural network cascade structure information, common layer network structure information, common layer network parameters, dedicated layer network structure information, and dedicated layer network parameters; and whether it has computing power information available for running neural networks.

[0389] Optionally, network devices only need to pre-configure neural network model information for terminal devices when the terminal device supports a neural network structure that cascades a common layer network and a private layer network.

[0390] The indexing of the dedicated layer network parameters in this step is optional. When this step of indexing the dedicated layer network parameters is not performed, the network device can always pre-configure the neural network model information to the terminal device without referring to the terminal device's capability information.

[0391] In step 1202, the network device sends configuration information to the terminal device. Accordingly, the terminal device obtains (e.g., receives) the configuration information.

[0392] The network device sends configuration information to the terminal device, which carries neural network model information. This neural network model information includes a candidate set of neural network cascade structure information, common layer network structure information, common layer network parameters, a candidate set of dedicated layer network structure information, and a candidate set of dedicated layer network parameters. This step can occur when the terminal device first accesses the wireless network, or when the network device needs to update the parameters configured in this step; there are no restrictions on when it does.

[0393] Optionally, the configuration information also carries index information. This index information includes the index of each type of neural network cascade structure in the candidate set of neural network cascade structure information, the index of each type of dedicated layer network structure in the candidate set of dedicated layer network structure information, and the index of each type of dedicated layer network parameter in the candidate set of dedicated layer network parameters.

[0394] Optionally, the configuration information also carries a first indication, which indicates the default neural network cascade structure information, the default dedicated layer network structure information, and the default dedicated layer network parameters. Here, "default" means alternative. For example, when the terminal device uses the neural network model for the first time, it can use the default neural network cascade structure information, the default dedicated layer network structure information, and the default dedicated layer network parameters. Alternatively, if an update error occurs when the network device updates the terminal device's neural network model, the terminal device will not update the neural network cascade structure information, the dedicated layer network structure information, and the dedicated layer network parameters, or it will use the default neural network cascade structure information, the default dedicated layer network structure information, and the default dedicated layer network parameters.

[0395] Steps 1203 to 1205 are the same. Figure 9 Steps 902 to 904 in the corresponding embodiment.

[0396] Step 1206: The network device determines the updated neural network model information of the terminal device.

[0397] If the network device determines that the neural network model information of the terminal device needs to be updated, the network device will then continue to determine the updated neural network model information of the terminal device.

[0398] The updated neural network model information for the terminal device includes:

[0399] 1) Information on the cascaded neural network structure of the terminal device;

[0400] 2) Dedicated layer network structure information for terminal equipment; and,

[0401] 3) Dedicated layer network parameters for terminal equipment.

[0402] Based on the neural network model information generation methods shown in Table 4, network devices can determine the neural network model information of terminal devices according to either Method 1 or Method 2.

[0403] Method 1: The network device selects the neural network cascade structure information, dedicated layer network structure information, and dedicated layer network parameters corresponding to the current wireless system parameters of the terminal device from the network device's pre-configured candidate set of neural network cascade structure information, candidate set of dedicated layer network structure information, and candidate set of dedicated layer network parameters.

[0404] For example, network devices store a mapping table between wireless system parameters and neural network cascade structure information, dedicated layer network structure information, and dedicated layer network parameters. The network device can look up the table based on the current wireless system parameters of the terminal device to obtain the neural network cascade structure information, dedicated layer network structure information, and dedicated layer network parameters corresponding to the current wireless system parameters of the terminal device.

[0405] In this method, the network structure information and parameters of the dedicated layer are obtained through offline training.

[0406] It should be noted that in the mapping table, the information on the neural network cascade structure, the information on the dedicated layer network structure, and the parameters of the dedicated layer network can be represented using the indices of the neural network cascade structure information, the dedicated layer network structure information, and the dedicated layer network parameters, respectively. That is, the mapping table stores the mapping relationships between wireless system parameters and the indices of the neural network cascade structure information, the dedicated layer network structure information, and the dedicated layer network parameters.

[0407] Method 2: The network device selects the neural network cascade structure information corresponding to the current wireless system parameters of the terminal device from the candidate set of neural network cascade structure information pre-configured by the network device, and obtains the dedicated layer network structure information and dedicated layer network parameters through online training (such as through machine learning or neural network training) based on the auxiliary information reported by the terminal device.

[0408] For example, network devices store a mapping table between wireless system parameters and neural network cascade structure information. The network device can look up the table based on the current wireless system parameters of the terminal device to obtain the neural network cascade structure information corresponding to the current wireless system parameters of the terminal device.

[0409] Furthermore, the network device also determines indication information for indicating updated neural network model information.

[0410] Based on Method 1 above, the determined indication information includes the following:

[0411] 1) Index of neural network cascade structure information of terminal devices;

[0412] 2) Index of dedicated layer network structure information for terminal equipment; and,

[0413] 3) Index of dedicated layer network parameters for terminal equipment.

[0414] Based on Method 1 above, the neural network cascade structure indicated by the index of the neural network cascade structure information of the terminal device can be... Figure 7The dedicated layer network structure is one of the neural network cascade structures in the network device, and is also one of the candidate sets of neural network cascade structure information pre-configured by the network device. The dedicated layer network structure information indicated by the index of the dedicated layer network structure information of the terminal device can be one of the candidate sets of dedicated layer network structure information pre-configured by the network device. The dedicated layer network parameters indicated by the index of the dedicated layer network parameters of the terminal device can be one of the candidate sets of dedicated layer network parameters pre-configured by the network device.

[0415] Based on Method 2 above, if the dedicated layer network structure information and dedicated layer network parameters obtained through online training have corresponding indices in the mapping table, then the determined indication information can include the following information:

[0416] 1) Index of neural network cascade structure information of terminal devices;

[0417] 2) Indexes to the dedicated layer network structure information of terminal devices; and,

[0418] 3) Index of dedicated layer network parameters for terminal devices.

[0419] Based on Method 2 above, if the terminal device supports receiving dedicated layer network structure information and dedicated layer network parameters via signaling, the indication information determined by the network device can include the following information:

[0420] 1) Index of neural network cascade structure information of terminal devices;

[0421] 2) Dedicated layer network structure information of terminal equipment; and,

[0422] 3) Dedicated layer network parameters of terminal equipment.

[0423] It should be noted that if the terminal device supports receiving dedicated layer network structure information and dedicated layer network parameters via signaling, the indication information determined by the network device can carry either the dedicated layer network structure information and dedicated layer network parameters obtained through offline training, or the dedicated layer network structure information and dedicated layer network parameters obtained through online training. As one implementation method, the network device can test the dedicated layer network structure information and dedicated layer network parameters obtained through offline training, as well as the dedicated layer network structure information and dedicated layer network parameters obtained through online training, based on the auxiliary information reported by the terminal device, thereby selecting the dedicated layer network structure information and dedicated layer network parameters with better test results.

[0424] Step 1207: The network device sends indication information to the terminal device. Accordingly, the terminal device obtains (e.g., receives) the indication information.

[0425] This instruction carries the following information:

[0426] 1) Index of neural network cascade structure information of terminal devices;

[0427] 2) Indexes to the dedicated layer network structure information of terminal devices; and,

[0428] 3) Index of dedicated layer network parameters for terminal devices.

[0429] Alternatively, the instruction message may carry the following information:

[0430] 1) Index of neural network cascade structure information of terminal devices;

[0431] 2) Dedicated layer network structure information of terminal equipment; and,

[0432] 3) Dedicated layer network parameters of terminal equipment.

[0433] Step 1208: The terminal device determines the neural network model information.

[0434] Based on the instructions sent by the network device, the terminal device determines its neural network model information, including the cascaded structure information, dedicated layer network structure information, and dedicated layer network parameters. Then, the terminal device determines its own neural network based on this model information.

[0435] Step 1209, same Figure 9 Step 908 in the corresponding embodiment.

[0436] It should be noted that, as an alternative implementation method, after step 1207, steps 1208 and 1209 are not executed, but step 907' is executed instead, as described above.

[0437] Example 5

[0438] like Figure 13 The diagram shown illustrates another parameter indication method for a neural network provided in this embodiment. In this embodiment, the network device pre-configures the following information to the terminal device:

[0439] 1) Candidate set of information for cascaded neural network structures.

[0440] 2) Common layer network structure information.

[0441] 3) Common layer network parameters; and,

[0442] 4) Candidate set of network structure information for dedicated layer.

[0443] Compared to Embodiment 4 above, in this embodiment, the network device does not have a pre-configured set of candidate network parameters for the dedicated layer.

[0444] Similar to Embodiment 1 above, the generation and determination methods of each variable in this embodiment are shown in Table 5.

[0445] Table 5

[0446]

[0447] like Figure 13 As shown, this embodiment includes the following steps:

[0448] Step 1301: The terminal device sends capability information to the network device. Accordingly, the network device acquires (e.g., receives) the capability information.

[0449] Before the network device pre-configures the neural network model information to the terminal device, the terminal device can report its capability information to the network device. This capability information includes one or more of the following: whether the terminal device supports using a neural network to replace or implement the functions of some communication modules; whether it supports a neural network structure that cascades a public layer network and a private layer network; whether it supports receiving private layer network structure information and private layer network parameters via signaling; whether it has memory space available to store public layer network structure information, public layer network parameters, private layer network structure information, and private layer network parameters; and whether it has computing power information available to run the neural network.

[0450] Optionally, the network device will only pre-configure neural network model information to the terminal device if the terminal device supports a neural network structure in which the public layer network and the private layer network are cascaded.

[0451] Step 1301 is optional. When step 1301 is not performed, the network device can always pre-configure the neural network model information to the terminal device without referring to the terminal device's capability information.

[0452] In step 1302, the network device sends configuration information to the terminal device. Accordingly, the terminal device obtains (e.g., receives) the configuration information.

[0453] When a terminal device first accesses the wireless network, the network device sends configuration information to the terminal device, which carries neural network model information. This neural network model information includes a candidate set of neural network cascade structure information, common layer network structure information, common layer network parameters, and a candidate set of dedicated layer network structure information.

[0454] Optionally, the configuration information also carries index information. This index information includes an index for each type of neural network cascade structure and an index for each type of dedicated layer network structure.

[0455] Optionally, the configuration information also carries a second indication, which indicates the default neural network cascade structure information and the dedicated layer network structure information.

[0456] Steps 1303 to 1305 are the same. Figure 12 Steps 1203 to 1205 in the corresponding embodiments.

[0457] Step 1306: The network device determines the updated neural network model information of the terminal device.

[0458] If the network device determines that the neural network model information of the terminal device needs to be updated, the network device will then continue to determine the updated neural network model information of the terminal device.

[0459] The updated neural network model information for the terminal device includes:

[0460] 1) Information on the cascaded neural network structure of the terminal device;

[0461] 2) Dedicated layer network structure information for terminal equipment; and,

[0462] 3) Dedicated layer network parameters for terminal equipment.

[0463] Based on the neural network model information generation methods shown in Table 5, network devices can determine the neural network model information of terminal devices according to either Method 1 or Method 2.

[0464] Method 1: The network device selects the neural network cascade structure information and the dedicated layer network structure information corresponding to the current wireless system parameters of the terminal device from the network device's pre-configured candidate set of neural network cascade structure information and candidate set of dedicated layer network structure information, respectively. Based on the auxiliary information reported by the terminal device, the dedicated layer network parameters are obtained through online training (such as through machine learning or neural network training).

[0465] For example, network devices store a mapping table between wireless system parameters and neural network cascade structure information and dedicated layer network structure information. The network device can look up the table based on the current wireless system parameters of the terminal device to obtain the neural network cascade structure information and dedicated layer network structure information corresponding to the current wireless system parameters of the terminal device.

[0466] In this method, the specialized layer network structure information is obtained through offline training.

[0467] It should be noted that in the mapping table, the information on the neural network cascade structure and the information on the dedicated layer network structure can be represented using the indices of the neural network cascade structure information and the dedicated layer network structure information, respectively. That is, the mapping table can store indices of wireless system parameters and neural network cascade structure information, indices of dedicated layer network structure information, and indices of dedicated layer network parameters.

[0468] Method 2: The network device selects the neural network cascade structure information corresponding to the current wireless system parameters of the terminal device from the candidate set of neural network cascade structure information pre-configured by the network device, and obtains the dedicated layer network structure information and dedicated layer network parameters through online training (such as through machine learning or neural network training) based on the auxiliary information reported by the terminal device.

[0469] For example, network devices store a mapping table between wireless system parameters and neural network cascade structure information. The network device can look up the table based on the current wireless system parameters of the terminal device to obtain the neural network cascade structure information corresponding to the current wireless system parameters of the terminal device.

[0470] Furthermore, the network device also determines indication information for indicating updated neural network model information.

[0471] Based on Method 1 above, the determined indication information includes the following:

[0472] 1) Index of neural network cascade structure information of terminal devices;

[0473] 2) Index of dedicated layer network structure information for terminal equipment; and,

[0474] 3) Dedicated layer network parameters for terminal equipment.

[0475] Based on Method 1 above, the neural network cascade structure indicated by the index of the neural network cascade structure information of the terminal device can be... Figure 7 This refers to a type of neural network cascade structure, which is also one of the candidate sets of neural network cascade structure information pre-configured by the network device. The dedicated layer network structure information indicated by the index of the terminal device's dedicated layer network structure information can be one of the candidate sets of dedicated layer network structure information pre-configured by the network device.

[0476] Based on Method 2 above, if the dedicated layer network structure information obtained through online training has a corresponding index in the mapping table, then the determined indication information can include the following information:

[0477] 1) Index of neural network cascade structure information of terminal devices;

[0478] 2) Index of dedicated layer network structure information for terminal devices;

[0479] 3) Dedicated layer network parameters of terminal equipment.

[0480] Based on Method 2 above, if the dedicated layer network structure information obtained through online training does not have a corresponding index in the mapping table, the indication information determined by the network device can include the following information:

[0481] 1) Index of neural network cascade structure information of terminal devices;

[0482] 2) Dedicated layer network structure information of terminal equipment; and,

[0483] 3) Dedicated layer network parameters of terminal equipment.

[0484] It should be noted that the indication information determined by the network device can carry either the dedicated layer network structure information obtained through offline training and the dedicated layer network parameters obtained through online training, or both. As one implementation method, the network device can test the dedicated layer network structure information obtained offline and the dedicated layer network parameters obtained online, as well as the dedicated layer network structure information obtained online, based on the auxiliary information reported by the terminal device, thereby selecting the dedicated layer network structure information and dedicated layer network parameters with the better test results.

[0485] It should be noted that if the network device can know the private layer network structure information and private layer network parameters currently being used by the terminal device, then after determining the latest private layer network structure information and private layer network parameters of the terminal device, the network device can also determine the amount of change between the latest private layer network parameters of the terminal device and the private layer network parameters currently being used. Therefore, it is not necessary to send the private layer network parameters of the terminal device to the terminal device, but to send the amount of change of the private layer network parameters of the terminal device.

[0486] Step 1307: The network device sends indication information to the terminal device. Accordingly, the terminal device obtains (e.g., receives) the indication information.

[0487] This instruction carries the following information:

[0488] 1) Index of neural network cascade structure information of terminal devices;

[0489] 2) Indexes to the dedicated layer network structure information of terminal devices; and,

[0490] 3) Dedicated layer network parameters of terminal equipment.

[0491] Alternatively, the instruction message may carry the following information:

[0492] 1) Index of neural network cascade structure information of terminal devices;

[0493] 2) Dedicated layer network structure information of terminal equipment; and,

[0494] 3) Dedicated layer network parameters of terminal equipment.

[0495] Step 1308: The terminal device determines the neural network model information.

[0496] Based on the instructions sent by the network device, the terminal device determines its neural network model information, including the cascaded structure information, dedicated layer network structure information, and dedicated layer network parameters. Then, the terminal device determines its own neural network based on this model information.

[0497] Step 1309, same as above. Figure 12 Step 1209 in the corresponding embodiment.

[0498] It should be noted that, as an alternative implementation method, after step 1307, steps 1308 and 1309 are not executed, but step 907' is executed instead, as described above.

[0499] Example 6

[0500] like Figure 14 The diagram shown illustrates another parameter indication method for a neural network provided in this embodiment. In this embodiment, the network device pre-configures the following information to the terminal device:

[0501] 1) Candidate set of information for cascaded neural network structures;

[0502] 2) Common layer network structure information; and,

[0503] 3) Common layer network parameters.

[0504] Compared with the above embodiment four, in this embodiment, the network device does not pre-configure the candidate set of dedicated layer network structure information and the candidate set of dedicated layer network parameters to the terminal device.

[0505] Similar to Embodiment 1 above, the generation and determination methods of each variable in this embodiment are shown in Table 6.

[0506] Table 6

[0507]

[0508] like Figure 14 As shown, this embodiment includes the following steps:

[0509] Step 1401: The terminal device sends capability information to the network device. Accordingly, the network device acquires (e.g., receives) the capability information.

[0510] Before the network device pre-configures the neural network model information to the terminal device, the terminal device can report its capability information to the network device. This capability information includes one or more of the following: whether the terminal device supports using a neural network to replace or implement the functions of some communication modules; whether it supports a neural network structure that cascades a public layer network and a private layer network; whether it supports receiving private layer network structure information and private layer network parameters via signaling; whether it has memory space available to store public layer network structure information, public layer network parameters, private layer network structure information, and private layer network parameters; and whether it has computing power information available to run the neural network.

[0511] Optionally, the network device will only pre-configure neural network model information to the terminal device if the terminal device supports a neural network structure in which the public layer network and the private layer network are cascaded.

[0512] Step 1401 is optional. When step 1401 is not performed, the network device can always pre-configure the neural network model information to the terminal device without referring to the terminal device's capability information.

[0513] In step 1402, the network device sends configuration information to the terminal device. Accordingly, the terminal device obtains (e.g., receives) the configuration information.

[0514] When a terminal device first accesses the wireless network, the network device sends configuration information to the terminal device, which carries neural network model information. This neural network model information includes a candidate set of neural network cascade structure information, common layer network structure information, and common layer network parameters.

[0515] Optionally, the configuration information also carries index information. This index information includes an index for each type of neural network cascade structure.

[0516] Optionally, the configuration information also carries a third indication, which indicates the default neural network cascade structure information.

[0517] Steps 1403 to 1405 are the same. Figure 12 Steps 1203 to 1205 in the corresponding embodiments.

[0518] Step 1406: The network device determines the updated neural network model information of the terminal device.

[0519] If the network device determines that the neural network model information of the terminal device needs to be updated, the network device will then continue to determine the updated neural network model information of the terminal device.

[0520] The updated neural network model information for the terminal device includes:

[0521] 1) Information on the cascaded neural network structure of the terminal device;

[0522] 2) Dedicated layer network structure information for terminal equipment; and,

[0523] 3) Dedicated layer network parameters for terminal equipment.

[0524] Based on the neural network model information generation method shown in Table 6, the network device can determine the neural network model information of the terminal device according to the following method: The network device selects the neural network cascade structure information corresponding to the current wireless system parameters of the terminal device from the candidate set of neural network cascade structure information pre-configured by the network device, and obtains the dedicated layer network structure information and dedicated layer network parameters through online training (such as through machine learning or neural network training) based on the auxiliary information reported by the terminal device.

[0525] For example, a network device might store a mapping table between wireless system parameters and neural network cascade structure information. The network device can look up the table based on the current wireless system parameters of the terminal device to obtain the corresponding neural network cascade structure information. It's important to note that the neural network cascade structure information in the mapping table can be represented using an index. In other words, the mapping table can store the mapping relationship between wireless system parameters and the indexes of neural network cascade structure information.

[0526] Furthermore, the network device also determines indication information for indicating updated neural network model information, which includes:

[0527] 1) Index of neural network cascade structure information of terminal devices;

[0528] 2) Dedicated layer network structure information for terminal equipment; and,

[0529] 3) Dedicated layer network parameters for terminal equipment.

[0530] Based on the above method, the neural network cascade structure indicated by the index of the neural network cascade structure information of the terminal device can be... Figure 7 It is one of the neural network cascade structures in the network, that is, one of the candidate sets of neural network cascade structure information pre-configured by the network device.

[0531] In step 1407, the network device sends indication information to the terminal device. Accordingly, the terminal device obtains (e.g., receives) the indication information.

[0532] This instruction carries the following information:

[0533] 1) Index of neural network cascade structure information of terminal devices;

[0534] 2) Dedicated layer network structure information of terminal equipment; and,

[0535] 3) Dedicated layer network parameters of terminal equipment.

[0536] Step 1408: The terminal device determines the neural network model information.

[0537] Based on the instructions sent by the network device, the terminal device determines its neural network model information, including the cascaded structure information, dedicated layer network structure information, and dedicated layer network parameters. Then, the terminal device determines its own neural network based on this model information.

[0538] Step 1409, same as above. Figure 12 Step 1209 in the corresponding embodiment.

[0539] It should be noted that, as an alternative implementation method, after step 1407, steps 1408 and 1409 are not executed, but step 907' is executed instead, as described above.

[0540] It should be noted that the above embodiments are illustrated using a communication system consisting of a network device and a terminal device as an example. As another application scenario, the present application can also be applied to a communication system consisting of two or more terminal devices. In this case, some or all of the operations performed by the network device in the above embodiments can be performed by the terminal device acting as the sending end, and the operations performed by the terminal device in the above embodiments can be performed by the terminal device acting as the receiving end. Optionally, in a communication system consisting of two or more terminal devices, the pre-configuration operation of the neural network model information performed by the network device in embodiments four to six can still be performed by the network device, while other operations performed by the network device are performed by the terminal device acting as the sending end of the communication system.

[0541] It should be noted that the above embodiments are illustrated using a predefined protocol or pre-configured network device set of common layer network structure information and common layer network parameters as an example. Alternatively, two or more sets of common layer network structure information and common layer network parameters can be predefined by the protocol or pre-configured by the network device. The network device then instructs the terminal device to select the chosen set of common layer network structure information and common layer network parameters via indication information. The method for selecting common layer network structure information and common layer network parameters can be referenced from the method for selecting dedicated layer network structure information and dedicated layer network parameters, and will not be elaborated further.

[0542] The following is combined Figure 6 The constellation neural network modulation module and constellation neural network demodulation module in this application provide a detailed explanation of how to use the neural network model provided in the embodiments of this application.

[0543] 1) Constellation Neural Network Modulation Module

[0544] The constellation neural network modulation module maps the bitstream into constellation modulation symbols, as shown in Figure 15(a), which is a schematic diagram of the constellation neural network modulation module structure. This module first maps the bitstream into one-hot codes. For example, for 4th-order constellation modulation, there are M = 4 constellation points in the constellation diagram. Therefore, each 2 bits of information is mapped to a 4-bit one-hot code, and the mapping relationship is shown in Table 7.

[0545] Table 7

[0546] Bit information Unique hot code 00 [0,0,0,1] 01 [0,0,1,0] 10 [0,1,0,0] 11 [1,0,0,0]

[0547] That is, a one-hot code is represented by M binary numbers. Each one-hot code has only one bit set to 1. Then, a constellation generation network is modeled using a neural network, the structure of which is shown in Figure 15(b).

[0548] The constellation generation network employs a two-layer fully connected network, plus a module for normalization and real-to-complex number transformation. The total input to the constellation generation network is the operating point signal-to-noise ratio (SNR), and the final output is a constellation diagram containing M=4 constellation points, such as [y1, y2, y3, y4], where y i The value of i ranges from 1 to 4, representing constellation points. Specifically, referring to Figure 15(b), taking M=2 as an example, the fully connected layer 1 (layer 1 for short) maps one input value (signal-to-noise ratio) to eight output values, the fully connected layer 2 (layer 2 for short) maps eight input values ​​to eight output values, and the module that performs the real-to-complex operation maps eight inputs to four outputs. These final four outputs constitute a constellation diagram containing four points, represented as [y1, y2, y3, y4]. The constellation diagram [y1, y2, y3, y4] is multiplied by the one-hot code obtained from the aforementioned transformation to obtain the constellation modulation symbol corresponding to each input (i.e., the bit information in the left column of Table 1).

[0549] 2) Constellation Neural Network Demodulation Module

[0550] The constellation neural network demodulation module demodulates the modulation symbols into log-likelihood ratios, as shown in Figure 15(c), which is a schematic diagram of the constellation neural network demodulation module structure. The total input of this module is the operating point signal-to-noise ratio and the received modulation symbols, and the final output is the log-likelihood ratio. The entire demodulation network is divided into a common layer and a dedicated layer. The common layer consists of three fully connected layers, and the input / output dimensions and activation functions are shown in Figure 15(c). The dedicated layer uses a single fully connected layer, and its output dimension is the number of bits corresponding to each modulation symbol. For different wireless parameters, the common layer and the private layer can adopt different structures and parameters, as described in the aforementioned embodiments.

[0551] refer to Figure 16 This is a schematic diagram of a communication device provided in an embodiment of this application. The communication device 1600 includes a transceiver unit 1610 and a processing unit 1620.

[0552] In the first embodiment, the communication device is used to implement the steps corresponding to the second device, receiving end, or terminal device in the above embodiments:

[0553] The transceiver unit 1610 is configured to acquire (e.g., receive) indication information from the first device, the indication information being used to indicate information about a first neural network model, the first neural network model including a dedicated layer network and a common layer network, the dedicated layer network being dedicated to the first neural network model, and the common layer network being a common network between the first neural network model and the second neural network model; the processing unit 1620 is configured to determine information about the first neural network model based on the indication information.

[0554] As one possible implementation method, the information of the first neural network model includes at least one of the following: neural network cascade structure information, common layer network structure information, common layer network parameters, dedicated layer network structure information, and dedicated layer network parameters.

[0555] As one possible implementation, the information of the first neural network model includes neural network cascade structure information, including: the information of the first neural network model indicating the neural network cascade structure, or the information of the first neural network model being used to indicate the configured neural network cascade structure from a set of neural network cascade structure candidates.

[0556] As one possible implementation, the information of the first neural network model includes common layer network structure information, including: the information of the first neural network model indicating the common layer network structure, or the information of the first neural network model being used to indicate the configured common layer network structure from a set of candidate common layer network structures.

[0557] As one possible implementation, the information of the first neural network model includes common layer network parameter information, including: the information of the first neural network model indicating the common layer network parameters, or the information of the first neural network model being used to indicate the configured common layer network parameters from a candidate set of common layer network parameters.

[0558] As one possible implementation, the information of the first neural network model includes dedicated layer network structure information, including: the information of the first neural network model indicating the dedicated layer network structure, or the information of the first neural network model being used to indicate the configured dedicated layer network structure from a set of dedicated layer network structure candidates.

[0559] As one possible implementation, the information of the first neural network model includes dedicated layer network parameter information, including: the information of the first neural network model indicating the dedicated layer network parameters, or the information of the first neural network model being used to indicate the configured dedicated layer network parameters from a set of dedicated layer network parameter candidates.

[0560] As one possible implementation, the transceiver unit 1610 is further configured to obtain (e.g., receive) configuration information from the first device, the configuration information indicating one or more of the following: a candidate set of neural network cascade structure information, a candidate set of common layer network structure information, a candidate set of common layer network parameters, a candidate set of dedicated layer network structure information, and a candidate set of dedicated layer network parameters.

[0561] As one possible implementation, the transceiver unit 1610 is further configured to send auxiliary information to the first device, the auxiliary information being used to indicate wireless system parameters between the first device and the second device.

[0562] As one possible implementation, the transceiver unit 1610 is further configured to send capability information to the first device, the capability information indicating one or more of the following:

[0563] 1) Does it support using neural networks to replace or implement the functions of the communication module?

[0564] 2) Does it support neural network structures where public layer networks and private layer networks are cascaded?

[0565] 3) Does it support receiving dedicated layer network structure information and dedicated layer network parameters via signaling?

[0566] 4) Stored protocol-predefined or pre-configured dedicated layer network structure information;

[0567] 5) Stored protocol predefined or preconfigured dedicated layer network parameters;

[0568] 6) Memory space that can be used to store information about the cascaded structure of neural networks, the structure of common layers, the parameters of common layers, the structure of private layers, and the parameters of private layers;

[0569] 7) Computing power information that can be used to run neural networks.

[0570] In the second embodiment, the communication device is used to implement the steps corresponding to the first device, the transmitting end, the terminal device, or the network device in the above embodiments:

[0571] Processing unit 1620 is used to determine indication information, the indication information being used to indicate information about a first neural network model, the first neural network model including a dedicated layer network and a common layer network, the dedicated layer network being dedicated to the first neural network model, and the common layer network being a common network between the first neural network model and the second neural network model; transceiver unit 1610 is used to send the indication information to a second device.

[0572] As one possible implementation method, the information of the first neural network model is the same as described above, and will not be repeated here.

[0573] As one possible implementation, the transceiver unit 1610 is further configured to send configuration information to the second device, the configuration information being used to indicate one or more of the following: a candidate set of neural network cascade structure information, a candidate set of common layer network structure information, a candidate set of common layer network parameters, a candidate set of dedicated layer network structure information, and a candidate set of dedicated layer network parameters.

[0574] As one possible implementation, the transceiver unit 1610 is also configured to acquire (e.g., receive) auxiliary information from the second device, the auxiliary information being used to indicate wireless system parameters between the first device and the second device.

[0575] As one possible implementation, the transceiver unit 1610 is also used to obtain (e.g., receive) capability information from the second device, which is the same as described above and will not be repeated here.

[0576] Optionally, the communication device may further include a storage unit for storing data or instructions (also referred to as code or program). Each of the aforementioned units can interact with or couple with the storage unit to implement the corresponding method or function. For example, the processing unit 1620 can read data or instructions from the storage unit, enabling the communication device to implement the method described in the above embodiments. The coupling in the embodiments of this application is an indirect coupling or communication connection between devices, units, or modules, which can be electrical, mechanical, or other forms, used for information exchange between devices, units, or modules.

[0577] In this embodiment, the division of units in the communication device is merely a logical functional division. In actual implementation, they can be fully or partially integrated onto a single physical entity, or they can be physically separated. Furthermore, all units in the communication device can be implemented entirely through software calls from processing elements; all units can be implemented entirely in hardware; or some units can be implemented through software calls from processing elements, while others are implemented in hardware. For example, each unit can be a separately established processing element, or it can be integrated into a chip within the communication device. Alternatively, it can be stored as a program in memory, called and executed by a processing element of the communication device. Moreover, these units can be fully or partially integrated together, or they can be implemented independently. The processing element mentioned here can also be called a processor, which can be an integrated circuit with signal processing capabilities. In the implementation process, each step of the above method or each of the above units can be implemented through integrated logic circuits in the processor element or through software calls from processing elements.

[0578] In one example, a unit in any of the above communication devices can be one or more integrated circuits configured to implement the above methods, such as: one or more application-specific integrated circuits (ASICs), or one or more digital signal processors (DSPs), or one or more field-programmable gate arrays (FPGAs), or a combination of at least two of these integrated circuit forms. As another example, when a unit in the communication device can be implemented in the form of a processing element scheduler, the processing element can be a general-purpose processor, such as a central processing unit (CPU) or other processor capable of calling programs. Furthermore, these units can be integrated together and implemented as a system-on-a-chip (SOC).

[0579] refer to Figure 17 This is a schematic diagram of a communication device provided in an embodiment of this application, used to implement the operations of the first device (i.e., the transmitting end, which may specifically be a network device or a terminal device) or the second device (i.e., the receiving end, which may specifically be a terminal device) in the above embodiments. Figure 17 As shown, the communication device includes a processor 1710 and an interface 1730. Optionally, the communication device also includes a memory 1720. The interface 1730 is used to enable communication with other devices. In this embodiment, the interface may also be referred to as a communication interface, and its specific form may be a transceiver, circuit, bus, module, pin, or other type of communication interface.

[0580] The method executed by the first or second device in the above embodiments can be implemented by the processor 1710 calling a program stored in memory (which can be memory 1720 in the first or second device, or external memory). That is, the first or second device may include the processor 1710, which executes the method executed by the first or second device in the above method embodiments by calling a program in memory. The processor here can be an integrated circuit with signal processing capabilities, such as a CPU. The first or second device can be implemented by one or more integrated circuits configured to implement the above methods. For example: one or more ASICs, or one or more microprocessors (DSPs), or one or more FPGAs, or a combination of at least two of these integrated circuit forms. Alternatively, the above implementation methods can be combined.

[0581] Specifically, Figure 16 The functions / implementation process of the transceiver unit 1610 and the processing unit 1620 can be obtained through Figure 17 The processor 1710 in the communication device 1700 shown calls computer-executable instructions stored in memory 1720 to implement the function. Alternatively, Figure 16 The function / implementation process of the processing unit 1620 can be achieved through... Figure 17 The processor 1710 in the communication device 1700 shown calls computer execution instructions stored in memory 1720 to implement this. Figure 16 The function / implementation process of the transceiver unit 1610 in the middle can be obtained through Figure 17 The interface 1730 in the communication device 1700 shown is used to implement this functionality. For example, the function / implementation process of the transceiver unit 1610 can be implemented by the processor calling program instructions in memory to drive the interface 1730.

[0582] In the embodiments of this application, the processor may be a general-purpose processor, a digital signal processor, an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components, capable of implementing or executing the methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor may be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this application can be directly manifested as being executed by a hardware processor, or executed by a combination of hardware and software modules within the processor.

[0583] In the embodiments of this application, the memory can be non-volatile memory, such as a hard disk drive (HDD) or a solid-state drive (SSD), or it can be volatile memory, such as random-access memory (RAM). Memory is any other medium capable of carrying or storing desired program code in the form of instructions or data structures, and accessible by a computer, but is not limited thereto. The memory in the embodiments of this application can also be a circuit or any other device capable of implementing storage functions, used to store program instructions and / or data.

[0584] Those skilled in the art will understand that the various numerical designations such as "first," "second," etc., involved in the embodiments of this application are merely for descriptive convenience and are not intended to limit the scope of the embodiments of this application, nor do they indicate a sequential order. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, and B alone. The character " / " generally indicates that the preceding and following related objects have an "or" relationship. "At least one" refers to one or more. "At least two" refers to two or more. "At least one," "any one," or similar expressions refer to any combination of these items, including any combination of single or multiple items. For example, at least one of a, b, or c can represent: a, b, c, ab, ac, bc, or abc, where a, b, and c can be single or multiple. "Multiple" refers to two or more, and other quantifiers are similar.

[0585] It should be understood that, in the various embodiments of this application, the sequence number of each process does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this invention.

[0586] The technical solutions provided in this application can be implemented, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented using software, they can be implemented, in whole or in part, in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, a terminal device, a network device, an artificial intelligence device, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available media may be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media (e.g., solid-state disks, SSDs), etc.

[0587] In the embodiments of this application, the embodiments may reference each other. For example, the methods and / or terms between method embodiments may reference each other, the functions and / or terms between device embodiments may reference each other, and the functions and / or terms between device embodiments and method embodiments may reference each other.

[0588] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A communication method characterized by comprising: Comprise: obtaining indication information from a first device, the indication information indicating information of a first neural network model, the first neural network model comprising a private layer network and a public layer network, the private layer network being private to the first neural network model, the public layer network being a common network of the first neural network model and a second neural network model, the information of the first neural network model comprising at least one of: neural network cascade structure information, public layer network structure information, public layer network parameters, private layer network structure information, or private layer network parameters; determining the information of the first neural network model according to the indication information; wherein the public network corresponds to a set of radio system parameters, each radio system parameter in the set of radio system parameters corresponding to a private layer network of the first neural network model or a private layer network of the second neural network model.

2. The method of claim 1, wherein, The information of the first neural network model comprises neural network cascade structure information, comprising: The information of the first neural network model indicates the neural network cascade structure, or The information of the first neural network model is used to indicate the configured neural network cascade structure from a neural network cascade structure candidate set.

3. The method of claim 1, wherein, The information of the first neural network model comprises public layer network structure information, comprising: The information of the first neural network model indicates the public layer network structure, or The information of the first neural network model is used to indicate the configured public layer network structure from a public layer network structure candidate set.

4. The method of claim 2, wherein, The information of the first neural network model comprises public layer network structure information, comprising: The information of the first neural network model indicates the public layer network structure, or The information of the first neural network model is used to indicate the configured public layer network structure from a public layer network structure candidate set.

5. The method according to any one of claims 1 to 4, characterized in that, The information of the first neural network model comprises public layer network parameter information, comprising: The information of the first neural network model indicates the public layer network parameters, or The information of the first neural network model is used to indicate the configured public layer network parameters from a public layer network parameter candidate set.

6. The method according to any one of claims 1 to 4, wherein, The information of the first neural network model comprises private layer network structure information, comprising: The information of the first neural network model indicates the private layer network structure, or The information of the first neural network model is used to indicate the configured private layer network structure from a private layer network structure candidate set.

7. The method of claim 5, wherein, The information of the first neural network model comprises private layer network structure information, comprising: The information of the first neural network model indicates the private layer network structure, or The information of the first neural network model is used to indicate the configured private layer network structure from a private layer network structure candidate set.

8. The method according to any one of claims 1 to 4, wherein The information of the first neural network model comprises private layer network parameter information, comprising: The information of the first neural network model indicates the private layer network parameters, or The information of the first neural network model is used to indicate the configured private layer network parameters from a private layer network parameter candidate set.

9. The method of claim 5, wherein, The information of the first neural network model comprises private layer network parameter information, comprising: The information of the first neural network model indicates the special layer network parameter, or The information of the first neural network model is used to indicate the configured special layer network parameter from a special layer network parameter candidate set.

10. The method according to any one of claims 1 to 4, wherein Further comprising: Receiving configuration information from the first device, the configuration information being used to indicate one or more of the following: a neural network cascade structure information candidate set, a common layer network structure information candidate set, a common layer network parameter candidate set, a special layer network structure information candidate set, and a special layer network parameter candidate set.

11. The method of claim 5, wherein, Further comprising: Receiving configuration information from the first device, the configuration information being used to indicate one or more of the following: a neural network cascade structure information candidate set, a common layer network structure information candidate set, a common layer network parameter candidate set, a special layer network structure information candidate set, and a special layer network parameter candidate set.

12. The method of claim 6, wherein, Further comprising: Receiving configuration information from the first device, the configuration information being used to indicate one or more of the following: a neural network cascade structure information candidate set, a common layer network structure information candidate set, a common layer network parameter candidate set, a special layer network structure information candidate set, and a special layer network parameter candidate set.

13. The method of claim 7, wherein, Further comprising: Receiving configuration information from the first device, the configuration information being used to indicate one or more of the following: a neural network cascade structure information candidate set, a common layer network structure information candidate set, a common layer network parameter candidate set, a special layer network structure information candidate set, and a special layer network parameter candidate set.

14. The method of claim 8, wherein, Further comprising: Receiving configuration information from the first device, the configuration information being used to indicate one or more of the following: a neural network cascade structure information candidate set, a common layer network structure information candidate set, a common layer network parameter candidate set, a special layer network structure information candidate set, and a special layer network parameter candidate set.

15. The method of claim 9, wherein, Further comprising: Receiving configuration information from the first device, the configuration information being used to indicate one or more of the following: a neural network cascade structure information candidate set, a common layer network structure information candidate set, a common layer network parameter candidate set, a special layer network structure information candidate set, and a special layer network parameter candidate set.

16. The method of any one of claims 1 to 4, wherein, Further comprising: Receiving configuration information from the first device, the configuration information being used to indicate one or more of the following: a neural network cascade structure information candidate set, a common layer network structure information candidate set, a common layer network parameter candidate set, a special layer network structure information candidate set, and a special layer network parameter candidate set.

17. The method of claim 5, wherein, Further comprising: Sending assistance information to the first device, the assistance information being used to indicate a wireless system parameter between the first device and a second device.

18. The method of claim 6, wherein, Further comprising: Sending assistance information to the first device, the assistance information being used to indicate a wireless system parameter between the first device and a second device.

19. The method of claim 7, wherein, Further comprising: Sending assistance information to the first device, the assistance information being used to indicate a wireless system parameter between the first device and a second device.

20. The method of claim 8, wherein, Further comprising: Sending assistance information to the first device, the assistance information being used to indicate a wireless system parameter between the first device and a second device.

21. The method of claim 9, wherein, Further comprising: Sending assistance information to the first device, the assistance information being used to indicate a wireless system parameter between the first device and a second device. Further comprising: sending assistance information to the first device, the assistance information being used to indicate radio system parameters between the first device and a second device.

22. The method of claim 10, wherein, Further comprising: sending assistance information to the first device, the assistance information being used to indicate radio system parameters between the first device and a second device.

23. The method of claim 11, wherein, Further comprising: sending assistance information to the first device, the assistance information being used to indicate radio system parameters between the first device and a second device.

24. The method of claim 12, wherein, Further comprising: sending assistance information to the first device, the assistance information being used to indicate radio system parameters between the first device and a second device.

25. The method of claim 13, wherein, Further comprising: sending assistance information to the first device, the assistance information being used to indicate radio system parameters between the first device and a second device.

26. The method of claim 14, wherein, Further comprising: sending assistance information to the first device, the assistance information being used to indicate radio system parameters between the first device and a second device.

27. The method of claim 15, wherein, Further comprising: sending assistance information to the first device, the assistance information being used to indicate radio system parameters between the first device and a second device.

28. The method of any one of claims 1 to 4, wherein, Further comprising: sending assistance information to the first device, the assistance information being used to indicate radio system parameters between the first device and a second device. Further comprising: sending capability information to the first device, the capability information being used to indicate one or more of the following information: 1) whether to support using neural network to replace or implement the function of communication module; 2) whether to support neural network structure of cascaded common layer network and dedicated layer network; 3) whether to support receiving dedicated layer network structure information and / or dedicated layer network parameters through signaling; 4) stored protocol predefined or preconfigured dedicated layer network structure information; 5) stored protocol predefined or preconfigured dedicated layer network parameters; 29. The method of claim 5, wherein, 6) memory space available for storing neural network cascaded structure information, common layer network structure information, common layer network parameters, dedicated layer network structure information and / or dedicated layer network parameters; 7) computing power information available for running neural network. Further comprising: sending capability information to the first device, the capability information being used to indicate one or more of the following information: 1) whether to support using neural network to replace or implement the function of communication module; 2) whether to support neural network structure of cascaded common layer network and dedicated layer network; 3) whether to support receiving dedicated layer network structure information and / or dedicated layer network parameters through signaling; 4) stored protocol predefined or preconfigured dedicated layer network structure information; 5) stored protocol predefined or preconfigured dedicated layer network parameters; 30. The method of claim 6, wherein, 6) memory space available for storing neural network cascaded structure information, common layer network structure information, common layer network parameters, dedicated layer network structure information and / or dedicated layer network parameters; 7) computing power information available for running neural network. Further comprising: sending capability information to the first device, the capability information being used to indicate one or more of the following information: 1) whether to support using neural network to replace or implement the function of communication module; 2) whether to support neural network structure of cascaded common layer network and dedicated layer network; 3) whether to support receiving dedicated layer network structure information and / or dedicated layer network parameters through signaling; 4) stored protocol predefined or preconfigured dedicated layer network structure information; 5) stored protocol pre-defined or pre-configured dedicated layer network parameters; 6) memory space available for storing neural network cascade structure information, common layer network structure information, common layer network parameters, dedicated layer network structure information, and / or dedicated layer network parameters; 7) computing power information available for running neural network.

31. The method of claim 7, wherein, Further comprising: sending, to the first device, capability information indicating one or more of the following: 1) whether to support using neural network to replace or implement the function of the communication module; 2) whether to support neural network structure of common layer network and dedicated layer network cascade; 3) whether to support receiving dedicated layer network structure information and / or dedicated layer network parameters through signaling; 4) stored protocol pre-defined or pre-configured dedicated layer network structure information; 5) stored protocol pre-defined or pre-configured dedicated layer network parameters; 6) memory space available for storing neural network cascade structure information, common layer network structure information, common layer network parameters, dedicated layer network structure information, and / or dedicated layer network parameters; 7) computing power information available for running neural network.

32. The method of claim 8, wherein, Further comprising: sending, to the first device, capability information indicating one or more of the following: 1) whether to support using neural network to replace or implement the function of the communication module; 2) whether to support neural network structure of common layer network and dedicated layer network cascade; 3) whether to support receiving dedicated layer network structure information and / or dedicated layer network parameters through signaling; 4) stored protocol pre-defined or pre-configured dedicated layer network structure information; 5) stored protocol pre-defined or pre-configured dedicated layer network parameters; 6) memory space available for storing neural network cascade structure information, common layer network structure information, common layer network parameters, dedicated layer network structure information, and / or dedicated layer network parameters; 7) computing power information available for running neural network.

33. The method of claim 9, wherein, Further comprising: sending, to the first device, capability information indicating one or more of the following: 1) whether to support using neural network to replace or implement the function of the communication module; 2) whether to support neural network structure of common layer network and dedicated layer network cascade; 3) whether to support receiving dedicated layer network structure information and / or dedicated layer network parameters through signaling; 4) stored protocol pre-defined or pre-configured dedicated layer network structure information; 5) stored protocol pre-defined or pre-configured dedicated layer network parameters; 6) memory space available for storing neural network cascade structure information, common layer network structure information, common layer network parameters, dedicated layer network structure information, and / or dedicated layer network parameters; 7) computing power information available for running neural network.

34. The method of claim 10, wherein, Further comprising: sending, to the first device, capability information indicating one or more of the following: 1) whether to support using neural network to replace or implement the function of the communication module; 2) whether to support neural network structure of common layer network and dedicated layer network cascade; 3) whether to support receiving dedicated layer network structure information and / or dedicated layer network parameters through signaling; 4) stored protocol pre-defined or pre-configured dedicated layer network structure information; 5) stored protocol pre-defined or pre-configured dedicated layer network parameters; 6) memory space available for storing neural network cascade structure information, common layer network structure information, common layer network parameters, dedicated layer network structure information, and / or dedicated layer network parameters; 7) computing power information available for running neural network. 5) stored protocol pre-defined or pre-configured dedicated layer network parameters; 6) memory space available for storing neural network cascade structure information, common layer network structure information, common layer network parameters, dedicated layer network structure information, and / or dedicated layer network parameters; 7) computing power information available for running neural network.

35. The method of claim 11, wherein, Further comprising: sending, to the first device, capability information indicating one or more of the following: 1) whether to support using neural network to replace or implement the function of the communication module; 2) whether to support neural network structure of common layer network and dedicated layer network cascade; 3) whether to support receiving dedicated layer network structure information and / or dedicated layer network parameters through signaling; 4) stored protocol pre-defined or pre-configured dedicated layer network structure information; 5) stored protocol pre-defined or pre-configured dedicated layer network parameters; 6) memory space available for storing neural network cascade structure information, common layer network structure information, common layer network parameters, dedicated layer network structure information, and / or dedicated layer network parameters; 7) computing power information available for running neural network.

36. The method of claim 12, wherein, Further comprising: sending, to the first device, capability information indicating one or more of the following: 1) whether to support using neural network to replace or implement the function of the communication module; 2) whether to support neural network structure of common layer network and dedicated layer network cascade; 3) whether to support receiving dedicated layer network structure information and / or dedicated layer network parameters through signaling; 4) stored protocol pre-defined or pre-configured dedicated layer network structure information; 5) stored protocol pre-defined or pre-configured dedicated layer network parameters; 6) memory space available for storing neural network cascade structure information, common layer network structure information, common layer network parameters, dedicated layer network structure information, and / or dedicated layer network parameters; 7) computing power information available for running neural network.

37. The method of claim 13, wherein, Further comprising: sending, to the first device, capability information indicating one or more of the following: 1) whether to support using neural network to replace or implement the function of the communication module; 2) whether to support neural network structure of common layer network and dedicated layer network cascade; 3) whether to support receiving dedicated layer network structure information and / or dedicated layer network parameters through signaling; 4) stored protocol pre-defined or pre-configured dedicated layer network structure information; 5) stored protocol pre-defined or pre-configured dedicated layer network parameters; 6) memory space available for storing neural network cascade structure information, common layer network structure information, common layer network parameters, dedicated layer network structure information, and / or dedicated layer network parameters; 7) computing power information available for running neural network.

38. The method of claim 14, wherein, Further comprising: sending, to the first device, capability information indicating one or more of the following: 1) whether to support using neural network to replace or implement the function of the communication module; 2) whether to support neural network structure of common layer network and dedicated layer network cascade; 3) whether to support receiving dedicated layer network structure information and / or dedicated layer network parameters through signaling; 4) stored protocol pre-defined or pre-configured dedicated layer network structure information; 5) stored protocol pre-defined or pre-configured dedicated layer network parameters; 6) memory space available for storing neural network cascade structure information, common layer network structure information, common layer network parameters, dedicated layer network structure information, and / or dedicated layer network parameters; 7) computing power information available for running neural network. 5) stored protocol pre-defined or pre-configured dedicated layer network parameters; 6) memory space available for storing neural network cascade structure information, common layer network structure information, common layer network parameters, dedicated layer network structure information, and / or dedicated layer network parameters; 7) computing power information available for running neural network.

39. The method of claim 15, wherein, Further comprising: sending, to the first device, capability information indicating one or more of the following: 1) whether to support using neural network to replace or implement the function of the communication module; 2) whether to support neural network structure of common layer network and dedicated layer network cascade; 3) whether to support receiving dedicated layer network structure information and / or dedicated layer network parameters through signaling; 4) stored protocol pre-defined or pre-configured dedicated layer network structure information; 5) stored protocol pre-defined or pre-configured dedicated layer network parameters; 6) memory space available for storing neural network cascade structure information, common layer network structure information, common layer network parameters, dedicated layer network structure information, and / or dedicated layer network parameters; 7) computing power information available for running neural network.

40. The method of claim 16, wherein, Further comprising: sending, to the first device, capability information indicating one or more of the following: 1) whether to support using neural network to replace or implement the function of the communication module; 2) whether to support neural network structure of common layer network and dedicated layer network cascade; 3) whether to support receiving dedicated layer network structure information and / or dedicated layer network parameters through signaling; 4) stored protocol pre-defined or pre-configured dedicated layer network structure information; 5) stored protocol pre-defined or pre-configured dedicated layer network parameters; 6) memory space available for storing neural network cascade structure information, common layer network structure information, common layer network parameters, dedicated layer network structure information, and / or dedicated layer network parameters; 7) computing power information available for running neural network.

41. The method of claim 17, wherein, Further comprising: sending, to the first device, capability information indicating one or more of the following: 1) whether to support using neural network to replace or implement the function of the communication module; 2) whether to support neural network structure of common layer network and dedicated layer network cascade; 3) whether to support receiving dedicated layer network structure information and / or dedicated layer network parameters through signaling; 4) stored protocol pre-defined or pre-configured dedicated layer network structure information; 5) stored protocol pre-defined or pre-configured dedicated layer network parameters; 6) memory space available for storing neural network cascade structure information, common layer network structure information, common layer network parameters, dedicated layer network structure information, and / or dedicated layer network parameters; 7) computing power information available for running neural network.

42. The method of claim 18, wherein, Further comprising: sending, to the first device, capability information indicating one or more of the following: 1) whether to support using neural network to replace or implement the function of the communication module; 2) whether to support neural network structure of common layer network and dedicated layer network cascade; 3) whether to support receiving dedicated layer network structure information and / or dedicated layer network parameters through signaling; 4) stored protocol pre-defined or pre-configured dedicated layer network structure information; 5) stored protocol pre-defined or pre-configured dedicated layer network parameters; 6) memory space available for storing neural network cascade structure information, common layer network structure information, common layer network parameters, dedicated layer network structure information, and / or dedicated layer network parameters; 7) computing power information available for running neural network. 5) stored protocol pre-defined or pre-configured dedicated layer network parameters; 6) memory space available for storing neural network cascade structure information, common layer network structure information, common layer network parameters, dedicated layer network structure information, and / or dedicated layer network parameters; 7) computing power information available for running neural network.

43. The method of claim 19, wherein, Further comprising: sending, to the first device, capability information indicating one or more of the following: 1) whether to support using neural network to replace or implement the function of the communication module; 2) whether to support neural network structure of common layer network and dedicated layer network cascade; 3) whether to support receiving dedicated layer network structure information and / or dedicated layer network parameters through signaling; 4) stored protocol pre-defined or pre-configured dedicated layer network structure information; 5) stored protocol pre-defined or pre-configured dedicated layer network parameters; 6) memory space available for storing neural network cascade structure information, common layer network structure information, common layer network parameters, dedicated layer network structure information, and / or dedicated layer network parameters; 7) computing power information available for running neural network.

44. The method of claim 20, wherein, Further comprising: sending, to the first device, capability information indicating one or more of the following: 1) whether to support using neural network to replace or implement the function of the communication module; 2) whether to support neural network structure of common layer network and dedicated layer network cascade; 3) whether to support receiving dedicated layer network structure information and / or dedicated layer network parameters through signaling; 4) stored protocol pre-defined or pre-configured dedicated layer network structure information; 5) stored protocol pre-defined or pre-configured dedicated layer network parameters; 6) memory space available for storing neural network cascade structure information, common layer network structure information, common layer network parameters, dedicated layer network structure information, and / or dedicated layer network parameters; 7) computing power information available for running neural network.

45. The method of claim 21, wherein, Further comprising: sending, to the first device, capability information indicating one or more of the following: 1) whether to support using neural network to replace or implement the function of the communication module; 2) whether to support neural network structure of common layer network and dedicated layer network cascade; 3) whether to support receiving dedicated layer network structure information and / or dedicated layer network parameters through signaling; 4) stored protocol pre-defined or pre-configured dedicated layer network structure information; 5) stored protocol pre-defined or pre-configured dedicated layer network parameters; 6) memory space available for storing neural network cascade structure information, common layer network structure information, common layer network parameters, dedicated layer network structure information, and / or dedicated layer network parameters; 7) computing power information available for running neural network.

46. The method of claim 22, wherein, Further comprising: sending, to the first device, capability information indicating one or more of the following: 1) whether to support using neural network to replace or implement the function of the communication module; 2) whether to support neural network structure of common layer network and dedicated layer network cascade; 3) whether to support receiving dedicated layer network structure information and / or dedicated layer network parameters through signaling; 4) stored protocol pre-defined or pre-configured dedicated layer network structure information; 5) stored protocol pre-defined or pre-configured dedicated layer network parameters; 6) memory space available for storing neural network cascade structure information, common layer network structure information, common layer network parameters, dedicated layer network structure information, and / or dedicated layer network parameters; 7) computing power information available for running neural network. 5) stored protocol pre-defined or pre-configured dedicated layer network parameters; 6) memory space available for storing neural network cascade structure information, common layer network structure information, common layer network parameters, dedicated layer network structure information, and / or dedicated layer network parameters; 7) computing power information available for running neural network.

47. The method of claim 23, wherein, Further comprising: sending, to the first device, capability information indicating one or more of the following: 1) whether to support using neural network to replace or implement the function of the communication module; 2) whether to support neural network structure of common layer network and dedicated layer network cascade; 3) whether to support receiving dedicated layer network structure information and / or dedicated layer network parameters through signaling; 4) stored protocol pre-defined or pre-configured dedicated layer network structure information; 5) stored protocol pre-defined or pre-configured dedicated layer network parameters; 6) memory space available for storing neural network cascade structure information, common layer network structure information, common layer network parameters, dedicated layer network structure information, and / or dedicated layer network parameters; 7) computing power information available for running neural network.

48. The method of claim 24, wherein, Further comprising: sending, to the first device, capability information indicating one or more of the following: 1) whether to support using neural network to replace or implement the function of the communication module; 2) whether to support neural network structure of common layer network and dedicated layer network cascade; 3) whether to support receiving dedicated layer network structure information and / or dedicated layer network parameters through signaling; 4) stored protocol pre-defined or pre-configured dedicated layer network structure information; 5) stored protocol pre-defined or pre-configured dedicated layer network parameters; 6) memory space available for storing neural network cascade structure information, common layer network structure information, common layer network parameters, dedicated layer network structure information, and / or dedicated layer network parameters; 7) computing power information available for running neural network.

49. The method of claim 25, wherein, Further comprising: sending, to the first device, capability information indicating one or more of the following: 1) whether to support using neural network to replace or implement the function of the communication module; 2) whether to support neural network structure of common layer network and dedicated layer network cascade; 3) whether to support receiving dedicated layer network structure information and / or dedicated layer network parameters through signaling; 4) stored protocol pre-defined or pre-configured dedicated layer network structure information; 5) stored protocol pre-defined or pre-configured dedicated layer network parameters; 6) memory space available for storing neural network cascade structure information, common layer network structure information, common layer network parameters, dedicated layer network structure information, and / or dedicated layer network parameters; 7) computing power information available for running neural network.

50. The method of claim 26, wherein, Further comprising: sending, to the first device, capability information indicating one or more of the following: 1) whether to support using neural network to replace or implement the function of the communication module; 2) whether to support neural network structure of common layer network and dedicated layer network cascade; 3) whether to support receiving dedicated layer network structure information and / or dedicated layer network parameters through signaling; 4) stored protocol pre-defined or pre-configured dedicated layer network structure information; 5) stored protocol pre-defined or pre-configured dedicated layer network parameters; 6) memory space available for storing neural network cascade structure information, common layer network structure information, common layer network parameters, dedicated layer network structure information, and / or dedicated layer network parameters; 7) computing power information available for running neural network. 5) stored protocol pre-defined or pre-configured dedicated layer network parameters; 6) memory space available for storing neural network cascade structure information, common layer network structure information, common layer network parameters, dedicated layer network structure information, and / or dedicated layer network parameters; 7) computing power information available for running neural networks.

51. The method of claim 27, wherein, Also comprising: sending capability information to the first device, the capability information indicating one or more of the following: 1) whether to support using neural networks to replace or implement the functions of the communication module; 2) whether to support a neural network structure of a common layer network and a dedicated layer network cascade; 3) whether to support receiving dedicated layer network structure information and / or dedicated layer network parameters through signaling; 4) stored protocol pre-defined or pre-configured dedicated layer network structure information; 5) stored protocol pre-defined or pre-configured dedicated layer network parameters; 6) memory space available for storing neural network cascade structure information, common layer network structure information, common layer network parameters, dedicated layer network structure information, and / or dedicated layer network parameters; 7) computing power information available for running neural networks.

52. A method of communication, the method comprising: Comprising: obtaining first indication information from the first device, the first indication information indicating common information of a first neural network model, the first neural network model comprising a dedicated layer network and a common layer network, the common information of the first neural network model comprising information of the common layer network, the information of the common layer network comprising common layer network structure information and / or common layer network parameters; obtaining second indication information from the first device, the second indication information indicating information of the dedicated layer network, the information of the dedicated layer network comprising at least one of the following: neural network cascade structure information, dedicated layer network structure information, or dedicated layer network parameters; wherein the common network corresponds to a set of radio system parameters, each radio system parameter in the set of radio system parameters corresponding to the dedicated layer network of the first neural network model or the dedicated layer network of the second neural network model.

53. The method of claim 52, wherein, The first indication information is carried through a broadcast message, a system message, or a common message.

54. The method of claim 52 or 53, wherein, The second indication information is carried through radio resource control (RRC) signaling specific to the second device, a medium access control (MAC) control element (CE), or downlink control information (DCI).

55. A method of communication, comprising: Comprising: determining indication information, the indication information indicating information of a first neural network model, the first neural network model comprising a dedicated layer network and a common layer network, the dedicated layer network being specific to the first neural network model, the common layer network being a common network of the first neural network model and a second neural network model, the information of the first neural network model comprising at least one of the following: neural network cascade structure information, common layer network structure information, common layer network parameters, dedicated layer network structure information, or dedicated layer network parameters; sending the indication information to the second device; wherein the common network corresponds to a set of radio system parameters, each radio system parameter in the set of radio system parameters corresponding to the dedicated layer network of the first neural network model or the dedicated layer network of the second neural network model.

56. The method of claim 55, wherein, Also comprising: sending configuration information to the second device, the configuration information being used to indicate one or more of the following: a candidate set of neural network cascade structure information, a candidate set of common layer network structure information, a candidate set of common layer network parameters, a candidate set of dedicated layer network structure information, and a candidate set of dedicated layer network parameters.

57. The method of claim 55, wherein, Also comprising: receiving assistance information from the second device, the assistance information being used to indicate wireless system parameters between the first device and the second device.

58. The method of claim 56, wherein, Also comprising: receiving assistance information from the second device, the assistance information being used to indicate wireless system parameters between the first device and the second device.

59. The method of any one of claims 55 to 58, wherein, Also comprising: receiving capability information from the second device, the capability information being used to indicate one or more of the following: 1) whether to support using neural network to replace or implement the function of the communication module; 2) whether to support the neural network structure of the cascade of common layer network and dedicated layer network; 3) whether to support receiving dedicated layer network structure information and dedicated layer network parameters through signaling; 4) stored protocol predefined or preconfigured dedicated layer network structure information; 5) stored protocol predefined or preconfigured dedicated layer network parameters; 6) memory space available for storing neural network cascade structure information, common layer network structure information, common layer network parameters, dedicated layer network structure information, and dedicated layer network parameters; 7) computing power information available for running neural network.

60. A method of communication, comprising: Comprising: sending first indication information to the second device, the first indication information being used to indicate common information of a first neural network model, the first neural network model comprising a dedicated layer network and a common layer network, the common information of the first neural network model comprising information of the common layer network, the information of the common layer network comprising common layer network structure information and / or common layer network parameters; sending second indication information to the second device, the second indication information being used to indicate information of the dedicated layer network, the information of the dedicated layer network comprising at least one of the following: neural network cascade structure information, dedicated layer network structure information, or dedicated layer network parameters; wherein the common network corresponds to a set of wireless system parameters, each wireless system parameter in the set of wireless system parameters corresponding to a dedicated layer network of the first neural network model or a dedicated layer network of a second neural network model.

61. A communications device, characterized by A method as claimed in any one of claims 1 to 54.

62. A communications device, characterized by A device comprising a processor and a memory, the memory and the processor coupled, the processor configured to perform a method as claimed in any one of claims 1 to 54.

63. A communications device, characterized by A device comprising a processor and a communication interface, The processor receives, from a first device via the communication interface, indication information indicating information of a first neural network model, the first neural network model comprising a dedicated layer network and a common layer network, the dedicated layer network being dedicated to the first neural network model, the common layer network being a common network of the first neural network model and a second neural network model, the information of the first neural network model comprising at least one of: neural network cascade structure information, common layer network structure information, common layer network parameters, dedicated layer network structure information, or dedicated layer network parameters; The processor is configured to determine the information of the first neural network model according to the indication information. The common network corresponds to a set of radio system parameters, each radio system parameter in the set of radio system parameters corresponding to the dedicated layer network of the first neural network model or the dedicated layer network of the second neural network model.

64. A communications device, characterized by A method as claimed in any one of claims 55 to 60.

65. A communications device, characterized by A device comprising a processor and a memory, the memory and the processor coupled, the processor configured to perform a method as claimed in any one of claims 55 to 60.

66. A communications device, characterized by A device comprising a processor and a communication interface, The processor is configured to determine indication information indicating information of a first neural network model, the first neural network model comprising a dedicated layer network and a common layer network, the dedicated layer network being dedicated to the first neural network model, the common layer network being a common network of the first neural network model and a second neural network model, the information of the first neural network model comprising at least one of: neural network cascade structure information, common layer network structure information, common layer network parameters, dedicated layer network structure information, or dedicated layer network parameters; The processor is configured to transmit, to a second device via the communication interface, the indication information. The common network corresponds to a set of radio system parameters, each radio system parameter in the set of radio system parameters corresponding to the dedicated layer network of the first neural network model or the dedicated layer network of the second neural network model.

67. A computer program product comprising instructions, wherein: The computer readable storage medium has stored thereon instructions which, when executed by a computer, cause the computer to perform a method as claimed in any one of claims 1 to 60.

68. A computer-readable storage medium, characterized in that, The computer readable storage medium has stored thereon instructions which, when executed by a computer, cause the computer to perform a method as claimed in any one of claims 1 to 60.

69. A communication system, characterized by A device as claimed in any one of claims 61 to 63, and / or a device as claimed in any one of claims 64 to 66.

Images