A physical layer security transmission method and system based on a graph neural network

Abstract

The application discloses a physical layer security transmission method and system based on a graph neural network, and belongs to the technical field of communication signal processing. Firstly, a scenario in which multiple eavesdroppers and legal users communicate with a base station is constructed, the instantaneous receiving rate of a receiver is analyzed, and a problem model in which a transmission beamforming is a variable and in which a maximum legal user receiving rate of the system and a minimum eavesdropper receiving rate of the system are taken as targets is formed. Then, a framework based on a graph neural network is developed, in which a beamforming vector is solved through neural network training. A random gradient descent method is adopted to optimize neural network parameters, so that the purpose of effectively suppressing eavesdropping and improving physical layer security is achieved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of communication signal processing, and particularly relates to a physical layer security transmission method and system based on a graph neural network. BACKGROUND

[0002] With the rapid development of wireless communication technology, communication security problems are increasingly valued. The continuous growth of computing power makes the previously considered unbreakable key no longer reliable, thereby highlighting the advantages of physical layer security over key-based security. By establishing a secure communication link between legitimate communicators, physical layer security technology helps to ensure that the transmitted information is stable and reliable, thereby ensuring the reliability and authenticity of the wireless communication system.

[0003] The concept of physical layer security provides a robust framework for protecting wireless communication by utilizing the physical characteristics of the transmission medium. However, it also faces challenges such as dynamic channel conditions and computational complexity. The conventional physical layer security solving method adopts successive convex approximation method or general power iteration method, both of which have high requirements for computing power, so when the physical environment is complex and changeable, these two methods are difficult to extend and use, and cannot fully meet the growing dynamic demand of physical layer security. In this special case, the academia proposes a more computationally efficient method to improve physical layer security, namely deep neural networks (DNN). DNN can counteract the nonlinearity and non-convexity of the physical layer security problem, and by utilizing parallel processing and avoiding the iteration process in the physical layer security solving process, it maximizes the reduction of computing resource requirements. Sangseok Yun et al. used DNN to solve the physical layer security optimization problem of MISO eavesdropping channel, and proved that the proposed scheme is always superior to the traditional scheme in ideal and actual training scenarios. DNN extracts potential features from a large number of Euclidean space structures, learns and updates network parameters, which relies on the rapid development of computing resources, which in turn supports the further prosperity of computing resources.

[0004] Although DNNs have strong performance in handling physical layer security, they are limited to processing grid-like data such as images or sequences, where input features have a fixed dimensional structure. However, due to the influence of factors such as channel condition changes, user mobility, beam interference, and the like, the wireless communication environment of physical layer security has inherent complexity and dynamics. Traditional Euclidean geometric representation is insufficient to accurately capture the relationship in the wireless network. In contrast, graph neural networks (GNNs) are specifically designed to directly process non-Euclidean space graph structure data, which provides a more flexible and adaptive framework for capturing the irregularity and dynamics of the wireless communication scenario. As an emerging neural network model, GNNs have been used to solve various problems in wireless communication and have achieved commendable results, such as radio resource management, user scheduling, and rate-maximized beamforming vector design, which have commonalities with physical layer security problems. GNNs are not only more suitable for modeling complex communication scenarios and problems, but also have excellent scalability, which can adapt to changes in graph structure without affecting model application due to the addition or deletion of nodes, which is a capability that DNNs lack, making GNNs particularly advantageous in solving physical layer security problems. SUMMARY

[0005] The technical problem to be solved by the present application is to provide a physical layer security transmission method and system based on a graph neural network, which researches the potential of graph neural networks in physical layer communication security, uses graph neural networks to achieve high scalability and efficient strategy design, and solves the technical problems of long operation time and poor generalization performance in solving physical layer security in wireless communication scenarios.

[0006] The technical scheme adopted by the present application is as follows:

[0007] A physical layer security transmission method based on a graph neural network, comprising the following steps:

[0008] A scenario in which a plurality of eavesdroppers and legitimate user receivers receive base station downlink communication signals is constructed;

[0009] Based on a given direction of arrival of the base station transmitter signal, the channel gain between the base station transmitter and the legitimate users and eavesdroppers is obtained, and the expression of the received signal at the legitimate users and eavesdroppers is determined;

[0010] According to the obtained expression of the received signal at the legitimate users and eavesdroppers, a problem model is constructed, taking each legitimate user's corresponding transmit beamforming vector as a variable and maximizing the security rate of the anti-eavesdropping system as an objective function;

[0011] Based on the constructed scenario and the obtained problem model, the nodes and connecting edges of the mapping graph of the scenario and the problem are defined;

[0012] The graph neural network encoder is designed based on the solved problem model, and the message generation and message aggregation methods in the graph neural network are defined.

[0013] The loss function, learning method and parameters of the neural network are designed, the neural network is trained, the problem model with generalization ability is obtained, and the physical layer security is improved.

[0014] Preferably, in the scenario where several eavesdroppers and legal users receive the downlink communication signals of the base station, the signals transmitted by the multi-antenna base station are received at uniformly and randomly distributed legal users in the area, and there is a one-to-one correspondence between eavesdroppers and corresponding users, and the eavesdroppers are randomly distributed within meters away from the corresponding users, and the user and eavesdropper pair set is represented as .

[0015] Preferably, the received signal of the legal user is :

[0016]

[0017] wherein, , represents the communication channel vector from the base station located at to the legal user , represents the additive white noise at the receiving end of the legal user , which follows a complex Gaussian distribution, is the signal transmitted by the base station to the user , is the number of user-eavesdropper pairs in the area, is the symbol of conjugate transpose;

[0018] The received signal of the eavesdropper is :

[0019]

[0020] wherein, represents the communication channel vector from the base station to the eavesdropper , represents the additive white noise at the receiving end of the eavesdropper , which follows a complex Gaussian distribution.

[0021] Preferably, the problem model is described as follows:

[0022]

[0023]

[0024] wherein, is the system safety rate, is, is the legitimate user receiving the target signal at a rate, is the eavesdropper receiving the target signal at a rate, is the base station maximum transmit power, is the base station beamforming vector to the user .

[0025] Preferably, the system safety rate is:

[0026]

[0027] Preferably, the legitimate user receiving the target signal at a rate and the eavesdropper receiving the target signal at a rate is:

[0028]

[0029]

[0030] wherein, is the base station channel gain to the user , is the noise power, is the base station maximum transmit power, is the base station beamforming vector to the user , is the number of user eavesdroppers in the region, is the base station channel gain to the eavesdropper .

[0031] Preferably, in the graph neural network, the initial feature of a node is represented as:

[0032]

[0033] wherein, is the operation of taking the real part of the base station beamforming vector to the user , is the operation of taking the imaginary part of the base station beamforming vector to the user , is the operation of taking the real part of the base station channel gain to the user , for taking the imaginary part of the channel gain from the base station to the user , for taking the real part of the channel gain from the base station to the eavesdropper , for taking the imaginary part of the channel gain from the base station to the eavesdropper .

[0034] Preferably, in the graph neural network encoder, the feature of the node in the first message passing is set as the feature of the node in the first message passing, and the node generates a message to be passed in the first message passing ; the message is converted into aggregated information of a single fixed dimension using a function as an aggregation function, a partial update method is adopted, part of the node features is reserved as an effective feature storage area of the original channel information, and the update variable of each layer is stored as an embedded feature of the remaining part of the node feature The first part of the above-mentioned represents the part of the node feature that stores the embedded feature, and is also the part of the feature that stores the update in each round of message passing. Preferably, the loss function of the neural network is :

[0035]

[0036]

[0037] wherein, is the system security rate, is the rate at which the legitimate user receives the target signal, is the rate at which the eavesdropper receives the target signal, is the number of users and eavesdroppers in the area. In a second aspect, an embodiment of the present application provides a physical layer secure transmission system based on a graph neural network, comprising:

[0038] A scene module constructs a scene in which a plurality of eavesdroppers and legitimate users receive downlink communication signals from a base station;

[0039] A signal module obtains channel gains from a base station transmitter to legitimate users and eavesdroppers based on a given direction of arrival of the base station transmitter signal, and determines expressions of received signals at the legitimate users and the eavesdroppers;

[0040]

[0041] ​​​​A function module constructs a problem model with each legitimate user corresponding to a transmission beamforming vector as a variable and a maximum security rate of the anti-eavesdropping system as an objective function based on the obtained expressions of the legitimate users and the eavesdropper;

[0042] A network module defines nodes and connection edges of a mapping graph of the scene and the problem based on the constructed scene and the obtained problem model, and designs a graph neural network encoder based on the solved problem model and defines a message generation and message aggregation method in the graph neural network;

[0043] A promotion module designs a loss function, a learning method and parameters of the neural network, trains the neural network, obtains a problem model with generalization ability, and realizes physical layer security promotion.

[0044] In a third aspect, a computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the steps of the above-mentioned physical layer security transmission method based on a graph neural network when executing the computer program.

[0045] In a fourth aspect, an embodiment of the present application provides a computer readable storage medium including a computer program, and the computer program implements the steps of the above-mentioned physical layer security transmission method based on a graph neural network when executed by a processor.

[0046] In a fifth aspect, a chip includes a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the steps of the above-mentioned physical layer security transmission method based on a graph neural network when executing the computer program.

[0047] In a sixth aspect, an embodiment of the present application provides an electronic device including a computer program, and the computer program implements the steps of the above-mentioned physical layer security transmission method based on a graph neural network when executed by the electronic device.

[0048] Compared with the prior art, the present application has at least the following beneficial effects:

[0049] A physical layer security transmission method based on graph neural network is provided, which considers the existence of multiple eavesdroppers in a legal communication link and proposes to use a graph neural network on the base station to generate a beamforming strategy to resist eavesdropping and improve the physical layer communication security. From the optimization point of view, the problem is a non-convex problem, which is difficult to solve effectively. In addition, the communication relationship and interference relationship in the wireless communication scene are complex and change frequently, and it is difficult to use a general method to solve the problem for different numbers of legal users. Based on this, a framework based on graph neural network is developed, which models the legal users, eavesdroppers, communication relationship and eavesdropping relationship in the wireless communication scene as the nodes and edges of the graph, and uses the beamforming vector as the variable to improve the physical layer security in the wireless communication scene. The random gradient descent method is used to optimize the neural network parameters to effectively improve the physical layer security of wireless communication. Numerical results show that this method can significantly improve the strategy determination efficiency while achieving near-optimal generalization performance.

[0050] Further, the design of the scene where there are several eavesdroppers and legal users receiving the base station downlink communication signal conforms to the general scene characteristics in wireless communication in reality, is easy to generalize to the real base station communication scene, and the design of multiple users and multiple eavesdroppers fully reflects the complexity and randomness of the communication scene, improving the adaptability of the scheme to various situations.

[0051] Further, based on the given direction of the base station transmitter signal arrival, the channel gain between the base station transmitter and the legal users and eavesdroppers is obtained, and the expression of the received signal at the legal users and eavesdroppers is determined, which lays the foundation for the necessary conditions for calculating the security rate and provides a theoretical basis for the subsequent problem model representation.

[0052] Further, the problem model is set as the difference between the legal user receiving rate and the eavesdropper eavesdropping rate, which clearly defines the problem solving goal in a simple form, and clearly defines the optimization variables and constraints, which provides favorable conditions for subsequent problem solving.

[0053] Further, the initial feature of the node is represented as two parts, one part is fixed during the processing of the multi-layer graph neural network and is set as the real and imaginary part arrangement vector of the legal user receiving channel and the eavesdropper eavesdropping channel; the other part changes dynamically during the processing of the multi-layer graph neural network and is used to save the neural network layer variables and the final output results of the neural network. The feature setting of the node serves the design of the message passing method in the graph neural network encoder and provides conditions for the subsequent message passing in the graph neural network encoder.

[0054] Further, in the graph neural network encoder, is the feature of the node in the th message passing, and the node In the first message passing, the message to be passed is generated ; using the function as an aggregation function, the message is converted into single fixed dimension aggregation information, using the partial update method, part of the node features are reserved as the effective feature storage area of the original channel information, and the update variable of each layer is stored as the embedded feature of the remaining part of the node feature, The first part of the node feature stores the embedded feature, and is also the part of the feature update in each round of message passing process. By controlling the consistency of the dimension of the input and output of each layer in the graph neural network encoder, the smooth update process of the node features between the neural network layers is ensured, and by maintaining the unchanged part of the node feature, i.e. the physical channel feature, in the graph neural network encoder, the learning ability of the neural network for the channel feature is strengthened, which is beneficial to improve the training effect of the neural network.

[0055] It can be understood that the beneficial effects of the above-mentioned second aspect can be referred to the related description in the above-mentioned first aspect, which will not be repeated here.

[0056] In summary, the present application applies the graph neural network technology to the communication scene with eavesdropping, uses the beamforming vector as a variable, and solves complex problems through the training of the neural network. In this regard, the neural network can approximate the internal relationship between the system input and output of various complex mathematical forms to achieve the best physical layer security communication performance.

[0057] The technical solutions of the present application will be further described in detail below with the help of the drawings and embodiments. BRIEF DESCRIPTION OF DRAWINGS

[0058] Figure 1 A wireless transmission system model diagram constructed by the present application;

[0059] Figure 2 A flowchart of the present application;

[0060] Figure 3 A schematic diagram of the relationship between the physical layer security rate of the system of the present application under different schemes and the number of user eavesdroppers;

[0061] Figure 4 A schematic diagram of the relationship between the processing time of the system of the present application under different schemes and the number of user eavesdroppers;

[0062] Figure 5 A schematic diagram of a computer device provided by an embodiment of the present application;

[0063] Figure 6 A block diagram of a chip provided by an embodiment of the present application. DETAILED DESCRIPTION

[0064] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are some of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative effort belong to the scope of the present application.

[0065] In the description of the present application, it should be understood that the terms "comprising" and "including" indicate the presence of described features, integers, steps, operations, elements, and / or components, but do not exclude one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0066] It should also be understood that the terms used in the specification of the present application are only for the purpose of describing specific embodiments and are not intended to limit the present application. As used in the specification and the appended claims of the present application, the singular forms "a", "an" and "the" are intended to include the plural forms unless the context clearly indicates otherwise.

[0067] It should be further understood that the term "and / or" used in the specification and the appended claims of the present application means any combination of one or more of the associated listed items and all possible combinations, and includes these combinations, for example, A and / or B can represent three cases of A alone, A and B together, and B alone. In addition, the character " / " in the present application generally represents an "or" relationship between the front and rear associated objects.

[0068] It should be understood that although the terms first, second, third, etc. may be used in the embodiments of the present application to describe the preset ranges, etc., these preset ranges should not be limited to these terms. These terms are only used to distinguish the preset ranges from each other. For example, the first preset range can also be referred to as the second preset range, and similarly, the second preset range can also be referred to as the first preset range without departing from the scope of the embodiments of the present application.

[0069] Depending on the context, the word "if" as used herein can be interpreted to mean "when" or "while" or "in response to the determination" or "in response to the detection." Similarly, the phrase "if determined" or "if detecting (a stated condition or event)" can be interpreted to mean "when determined" or "in response to the determination" or "when detecting (a stated condition or event)" or "in response to the detection (a stated condition or event)," depending on the context.

[0070] The various structural diagrams according to the disclosed embodiments of the present application are shown in the drawings. These diagrams are not drawn to scale, in which certain details are exaggerated for clarity and others omitted. The shapes and relative sizes of the various regions, layers, and their relative positions are shown only by way of example, and may deviate in actuality due to manufacturing tolerances or technical limitations, and regions / layers with different shapes, sizes, and relative positions can be additionally designed according to actual needs by those skilled in the art.

[0071] The present application provides a physical layer security transmission method based on graph neural network, which reveals the potential of graph neural network in secure communication. Considering the existence of multiple eavesdroppers in the legal communication link, and suggesting to use graph neural network on the base station to generate beamforming strategy to resist eavesdropping. First, the scene of multiple eavesdroppers and legal users communicating with the base station is constructed, and the instantaneous receiving rate of the receiver is analyzed, and the problem model is formed with the maximum legal user receiving rate of the system and the minimum eavesdropper receiving rate of the system as the target, and the transmit beamforming as the variable. Then a framework based on graph neural network is developed, in which the beamforming vector is trained by neural network to solve; the random gradient descent method is used to optimize the neural network parameters, so as to effectively suppress eavesdropping and improve the physical layer security.

[0072] Referring to Figure 2 , the present application is a physical layer security transmission method based on graph neural network, comprising the following steps:

[0073] S1, constructing a scene in which several eavesdroppers and legal users receive the base station downlink communication signal;

[0074] Referring to Figure 1 , the present application constructs a wireless communication system model diagram, considering a downlink MISO communication system containing multiple users and eavesdroppers in a fixed area. In the fixed area, the signal is transmitted by a multi-antenna base station to uniformly randomly distributed single-antenna legal users in the area, and corresponding eavesdroppers are also distributed in the area, which simultaneously eavesdrop on the corresponding user signals, and the eavesdroppers are randomly distributed within meters from the corresponding user, then such a user and eavesdropper pair set is represented as

[0075] S2, based on the given direction of the known base station transmitter signal, the channel gain between the base station transmitter and the legal users and eavesdroppers is obtained, and then the expression of the received signal at the legal users and eavesdroppers is obtained;

[0076] According to the positions of the base station, legal users and eavesdroppers in the considered system, the base station is The signal transmitted by the antenna is transmitted to the user through an air channel The beamforming vector of the base station to the user in this process can be expressed as When the maximum transmission power of the base station is , the beamforming vector satisfies The signal transmitted by the base station is expressed as:

[0077]

[0078] wherein is the signal transmitted by the base station to the user , and satisfies When the signal transmitted by the base station is received by the user receiving end through an air channel, the signal received by the user is further expressed on the basis of the above formula:

[0079]

[0080] wherein , represents the communication channel vector from the base station to the user , and in the formula represents the additive white noise of the user receiving end in compliance with a complex Gaussian distribution. The signal received by the eavesdropper receiving end is expressed in an approximate form:

[0081]

[0082]

[0083] wherein represents the communication channel vector from the base station to the eavesdropper , and represents the additive white noise of the eavesdropper receiving end in compliance with a complex Gaussian distribution.

[0084] S3. According to the signal expression of the base station transmitter signal, a problem model is constructed, in which the transmission beamforming of each legal user is taken as a variable, and the maximum security rate of the anti-eavesdropping system is taken as an objective function.

[0085] For the system considered, we expect to defend eavesdropping to protect the required legal transmission link. Therefore, when the fixed base station position is taken, a more suitable beamforming vector group of the system is designed to improve the security transmission performance of the system, so as to improve the receiving rate of the legal user and suppress the eavesdropping rate of the eavesdropper. On the basis of the received signal expression, the rate of the target signal received by each user and the eavesdropper is expressed by using the Shannon formula:​​​

[0086]

[0087]

[0088] Therefore, the system's safe rate is expressed as:

[0089]

[0090] Therefore, taking the transmitted beamforming vector as a variable, considering the limitation of maximum transmitted power, and setting the objective function as maximizing the safe rate of this system, the problem is described as follows:

[0091]

[0092]

[0093] S4. Based on the scenario in step S1, define the nodes and connecting edges in the graph neural network;

[0094] Based on the scenario in step S1, the nodes and connecting edges in the graph neural network are defined. The graph representation processed by the graph neural network is a geometric representation of points and edges. The organization of points and edges can reflect various relationships in reality. The closer this organizational relationship of the graph data is to the actual physical relationships, the more advantageous it is for processing specific data. How to design the graph representation of data requires comprehensive consideration in conjunction with graph data processing methods.

[0095] To simulate the scenario in step S1, we define a fully connected graph structure with each legitimate user and their corresponding eavesdropper as a node. We assume the edges have no features, and the channel vectors of the legitimate user and the eavesdropper are decomposed and reassembled according to their real and imaginary parts. Then, the nodes... initial features Represented as:

[0096]

[0097] S5. Design a graph neural network encoder and define the methods for message generation and message aggregation in the graph neural network.

[0098] In graph neural networks (Graph Neural Networks), the interaction of features between different nodes needs to be implemented by defining methods for message generation and aggregation. Generally, Graph Neural Networks process graph data through message passing. Messages refer to information such as the features of nodes and edges. Messages are passed along the relationships between nodes and the direction of edges in the graph structure, allowing related nodes to influence each other, and nodes to update their features by incorporating the influence of their related nodes.

[0099] The message passing process in graphs and convolution in images share many similarities: both extract features from neighboring nodes and use these extracted features to update the features of the central node. The difference lies in that image convolution is performed in a plane, and can only extract features from a fixed number of adjacent nodes at fixed positions. The number of neighboring nodes extracted depends on the size of the convolution kernel. However, message passing in graphs operates in non-Euclidean space, and the central node can extract features from all its neighboring nodes without considering the graph's structure. GNNs capture more structural information from the graph and combine the relationships between nodes with the representation vectors of node features.

[0100] Message generation in the message passing process is the process by which a node generates the information to be transmitted using its own characteristics. Let's assume... For nodes In the first The characteristics of this message passing, then the node In the The message to be delivered generated in this message passing process Represented as:

[0101]

[0102] in, It refers to the first In this message generation function, a multilayer perceptron is used as the message generation function in a deep graph neural network.

[0103] Neighbor message aggregation is the second part of the message passing process. It occurs after messages have traveled along the edges of the graph data. A node receives messages from multiple neighbors, and aggregation refers to converting these messages into aggregated information of a single fixed dimension, preparing for subsequent feature updates. The function acts as an aggregation function to select the message with the greatest impact, and the selected message participates in the aggregation process of neighbor messages:

[0104]

[0105] A partial update method is adopted, retaining a portion of node features as an effective feature storage region for the original channel information, and storing the update variables of each layer as embedded features of the remaining part of the node features. When using the node features of the previous layer in the feature update of each layer, it is equivalent to repeatedly emphasizing the effective features of the original channel information, effectively avoiding deviation from the learning direction in the multi-layer graph convolution process, and helping to accelerate the learning convergence to the expected value, as shown below:

[0106]

[0107] in, The former Item represents the part of the node feature that stores the embedding feature and also the part that stores the feature update in each round of message passing.

[0108] S6, loss function, learning method and parameter of the whole neural network are designed, the neural network is trained, and a model with good generalization ability is obtained.

[0109] The loss function of the network is defined as the negative value of the system security rate:

[0110]

[0111] By adjusting the parameters and multiple rounds of training, the loss function converges finally, and the beamforming strategy of system security communication is obtained.

[0112] In another embodiment of the application, a physical layer security transmission system based on a graph neural network is provided, which can be used to implement the above-mentioned physical layer security transmission method based on a graph neural network.

[0113] The scene module constructs a scene in which there are several eavesdroppers and legitimate users receiving downlink communication signals of the base station.

[0114] The signal module obtains the channel gain between the base station transmitter and the legitimate users and eavesdroppers based on the given direction of the base station transmitter signal arrival, and determines the expression of the received signal at the legitimate users and eavesdroppers.

[0115] The function module constructs a problem model with the transmission beamforming vector corresponding to each legitimate user as the variable and the maximum security rate of the anti-eavesdropping system as the objective function based on the obtained expression of the received signal at the legitimate users and eavesdroppers.

[0116] The network module defines the nodes and connection edges of the mapping graph of the constructed scene and the obtained problem model based on the solved problem model, and designs a graph neural network encoder based on the solved problem model to define the message generation and message aggregation method in the graph neural network.

[0117] The improvement module designs the loss function, learning method and parameter of the neural network, trains the neural network, obtains a problem model with generalization ability, and realizes the improvement of physical layer security.

[0118] In another embodiment of the present application, a terminal device is provided, which comprises a processor and a memory, the memory being configured to store a computer program, the computer program comprising program instructions, and the processor being configured to execute the program instructions stored in the computer storage medium. The processor can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc., which are the computing core and control core of the terminal, and are suitable for implementing one or more instructions, and are particularly suitable for loading and executing one or more instructions to implement a corresponding method flow or a corresponding function. The processor in the embodiments of the present application can be used for the operation of the physical layer security transmission method based on the graph neural network, including:

[0119] A scenario in which there are eavesdroppers and legitimate users receiving downlink communication signals of a base station is constructed; based on a given direction of arrival of a base station transmitter signal, channel gains between the base station transmitter and the legitimate users and the eavesdroppers are obtained, and expressions of received signals at the legitimate users and the eavesdroppers are determined; based on the obtained expressions of received signals at the legitimate users and the eavesdroppers, a problem model is constructed, in which each legitimate user's corresponding transmit beamforming vector is taken as a variable, and maximizing the security rate of the anti-eavesdropping system is taken as an objective function; based on the constructed scenario and the obtained problem model, nodes and connecting edges of a graph mapping the scenario and the problem are defined; based on the solved problem model, a graph neural network encoder is designed, and a method of message generation and message aggregation in the graph neural network is defined; a loss function, a learning method and parameters of the neural network are designed, the neural network is trained, a problem model with generalization ability is obtained, and physical layer security is improved.

[0120] Please refer to Figure 5 , the terminal device is a computer device, the computer device 60 of the embodiment comprises a processor 61, a memory 62, and a computer program 63 stored in the memory 62 and executable on the processor 61, and the computer program 63 implements the physical layer security transmission method based on the graph neural network in the embodiment when executed by the processor 61, to avoid repetition, which will not be described here. Alternatively, the computer program 63 implements the functions of each model / unit in the physical layer security transmission system based on the graph neural network in the embodiment when executed by the processor 61, to avoid repetition, which will not be described here.

[0121] The computer device 60 can be a desktop computer, a notebook computer, a palm computer, a cloud server, and the like. The computer device 60 can include, but is not limited to, a processor 61, a memory 62. Those skilled in the art can understand that Figure 5 The computer device 60 is only an example and does not constitute a limitation on the computer device 60, and can include more or fewer components than shown, or combine certain components, or different components, for example, the computer device can also include an input / output device, a network access device, a bus, and the like.

[0122] The processor 61 can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs) or other programmable logic devices, discrete gates or transistor logic components, discrete hardware components, and the like. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor.

[0123] The memory 62 can be an internal storage unit of the computer device 60, such as a hard disk or a memory of the computer device 60. The memory 62 can also be an external storage device of the computer device 60, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, and the like.

[0124] Further, the memory 62 can include both an internal storage unit and an external storage device of the computer device 60. The memory 62 is used to store computer programs and other programs and data required by the computer device. The memory 62 can also be used to temporarily store data that has been output or will be output.

[0125] Please refer to Figure 6 The terminal device is a chip, and the chip 600 of the embodiment includes a processor 622, the number of which can be one or more, and a memory 632 for storing a computer program executable by the processor 622. The computer program stored in the memory 632 can include one or more than one module each corresponding to a set of instructions. In addition, the processor 622 can be configured to execute the computer program to perform the above-mentioned physical layer security transmission method based on the graph neural network.

[0126] Additionally, the chip 600 can also include a power supply component 626 that can be configured to perform power management of the chip 600, and a communication component 650 that can be configured to implement communication of the chip 600, for example, wired or wireless communication. In addition, the chip 600 can also include an input / output interface 658. The chip 600 can operate based on an operating system stored in the memory 632.

[0127] In another embodiment of the present application, the present application also provides a storage medium, specifically a computer readable storage medium, which is a memory device in a terminal device, used to store programs and data. It can be understood that the computer readable storage medium herein can include a built-in storage medium in the terminal device, and of course can also include an expansion storage medium supported by the terminal device. The computer readable storage medium provides a storage space, which stores an operating system of the terminal. In addition, one or more instructions suitable for being loaded and executed by the processor are also stored in the storage space, and these instructions can be one or more computer programs. It should be noted that the computer readable storage medium herein can be a high-speed RAM memory, or a non-volatile memory such as at least one disk memory.

[0128] The one or more instructions stored in the computer readable storage medium can be loaded and executed by the processor to implement the corresponding steps of the physical layer security transmission method based on the graph neural network in the above embodiments; the one or more instructions in the computer readable storage medium are loaded and executed by the processor as follows:

[0129] A scenario in which there are eavesdroppers and legitimate users receiving downlink communication signals of a base station is constructed; based on a given direction of arrival of the base station transmitter signal, channel gains between the base station transmitter and the legitimate users and the eavesdroppers are obtained, expressions of received signals at the legitimate users and the eavesdroppers are determined; based on the obtained expressions of received signals at the legitimate users and the eavesdroppers, a problem model is constructed, taking the transmission beamforming vectors corresponding to each legitimate user as variables, and taking the maximum security rate of the anti-eavesdropping system as the objective function; based on the constructed scenario and the obtained problem model, nodes and connecting edges of a graph mapping the scenario and the problem are defined; based on the solved problem model, a graph neural network encoder is designed, and methods of message generation and message aggregation in the graph neural network are defined; a loss function, a learning method and parameters of the neural network are designed, the neural network is trained, a problem model with generalization ability is obtained, and physical layer security is improved.

[0130] In order to make the purposes, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some but not all of the embodiments of the present application. The components of the embodiments of the present application described and shown in the drawings herein can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the claimed present application, but only represents selected embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the scope of protection of the present application.

[0131] Referring to Figure 3 , a schematic diagram of the physical layer security rate changing with the number of user eavesdroppers under different schemes is given. It can be seen from Figure 3 that when the number of users increases or decreases, the graph neural network can directly obtain the high-security rate of the transmit beamforming strategy. This is because the message passing mechanism of the graph makes the number of nodes and the connection mode of the graph unrestricted. However, the performance of the scheme proposed by us is inferior to that of the OPT scheme (i.e., the transmit beamforming uses an optimization method), but the OPT scheme needs to be adjusted for different numbers of users, which poses a major challenge to the practical engineering application involving complex and variable wireless communication scenarios. In addition, the MLP (i.e., using a multi-layer perception to generate a beamforming strategy) as a way of machine learning needs to redesign the structure of the neural network and retrain for different numbers of users and eavesdroppers, which illustrates the potential of the graph neural network in expanding the ability to protect physical layer communication security.

[0132] Referring to Figure 4 , a schematic diagram of the processing time changing with the number of user eavesdroppers under different schemes is given. When the number of user eavesdroppers increases, the processing time required by the system also increases. The time consumption of the graph neural network stably remains at a very low time consumption, and in most cases, the time required to execute the OPT once is much longer than executing the graph neural network operation 50 times on the same data set, which is due to the large number of matrix calculations and iterations involved in the optimization process, which poses a major challenge to the practical engineering application involving delay-sensitive tasks, which illustrates the high timeliness of the graph neural network in protecting physical layer communication security.

[0133] In summary, the physical layer security transmission method and system based on the graph neural network provides a solution for physical layer communication security by using the graph neural network to generate a beamforming vector in the presence of multiple eavesdroppers in a legitimate communication link. The present application formulates the problem as a security rate maximization problem and develops a graph neural network-based solution to beamforming. The random gradient descent method is used to optimize the parameters of the neural network, effectively suppressing eavesdropping and improving the reception rate of legitimate users.

[0134] Those skilled in the art can clearly understand that, for the convenience and brevity of description, only the division of the above functional units and modules is exemplified, and in actual application, the above functions can be completed by different functional units and modules according to needs, that is, the internal structure of the device is divided into different functional units or modules to complete all or part of the functions described above. Each functional unit or module in the embodiment can be integrated in one processing unit, or each unit can be physically present separately, or two or more units can be integrated in one unit. The integrated unit can be realized in the form of hardware or software. In addition, the specific names of each functional unit or module are only for easy distinction, and do not limit the protection scope of the present application. The specific working process of the units and modules in the system can refer to the corresponding process in the foregoing method embodiments, which will not be repeated here.

[0135] In the above embodiments, the description of each embodiment has its own emphasis, and the parts not described or recorded in detail in a certain embodiment can be referred to the relevant description of other embodiments.

[0136] Those skilled in the art can realize that the units and algorithm steps of each example described in combination with the disclosed embodiments of the present application can be realized by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are executed in hardware or software depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0137] In the embodiments of the present application, it should be understood that the disclosed apparatus / terminal and method can be implemented in other manners. For example, the apparatus / terminal embodiments described above are merely schematic, and the division of the modules or units is merely logical function division, and there can be another division manner in actual implementation. For example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the displayed or discussed mutual couplings or direct couplings or communication connections can be indirect couplings or communication connections through some interfaces, devices or units, and can be in electrical, mechanical or other forms.

[0138] The units described as separate components can or can not be physically separate, and the components shown as units can or can not be physical units, i.e., can be located in one place, or can be distributed on a plurality of network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the embodiment.

[0139] In addition, each functional unit in each embodiment of the present application can be integrated into a processing unit, or each unit can be physically present separately, or two or more units can be integrated into one unit. The integrated unit can be realized in the form of hardware or in the form of a software functional unit.

[0140] If the integrated module / unit is realized in the form of a software functional unit, it can be stored in a computer readable storage medium. Based on this understanding, all or part of the flow of the above-mentioned embodiment methods can also be completed by a computer program instructing related hardware, and the computer program can be stored in a computer readable storage medium. When the processor executes the computer program, the steps of each method embodiment described above can be implemented. The computer program includes computer program code, which can be in the form of source code, object code, executable file or some intermediate form. The computer readable medium can include any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal and software distribution medium, etc.

Claims

1. A physical layer security transmission method based on a graph neural network, characterized in that, The method comprises the following steps: A scenario in which a plurality of eavesdroppers and legitimate users receive downlink communication signals of a base station is constructed; Based on the given direction of arrival of the base station transmitter signal, the channel gains between the base station transmitter and the legitimate user and the eavesdropper are obtained, expressions for the received signals at the legitimate user and the eavesdropper are determined, the received signal at the legitimate user is: wherein , denotes the base station's communication channel vector to the legitimate user , is the base station's channel gain to the user , is the base station's beamforming vector to the user , denotes the complex Gaussian distributed additive white noise at the legitimate user's receiver, is the signal transmitted by the base station to the user , is the number of user eavesdroppers in the area, is the symbol for conjugate transpose and the set of users and eavesdroppers is denoted by ; Eavesdropper receives signal To: in, From base station to eavesdropper The communication channel vector, From base station to eavesdropper Channel gain, Indicates eavesdropper The receiver contains additive white noise that follows a complex Gaussian distribution. According to the obtained expressions of the received signals at the legitimate users and the eavesdroppers, a problem model is constructed, in which a transmission beamforming vector corresponding to each legitimate user is taken as a variable, and a maximum security rate of the anti-eavesdropping system is taken as an objective function; wherein is the system safety rate, is the number of user eavesdroppers in the region, is the number of legitimate users is the rate at which the target signal is received, is the number of eavesdroppers is the rate at which the target signal is received, is the maximum transmit power of the base station, is the beamforming vector of the base station to the user is the number of legitimate users is the rate at which the target signal is received is the number of eavesdroppers is the rate at which the target signal is received is: in, For base station to user Channel gain, For noise power, This is the maximum transmit power of the base station. For base station to user Beamforming vector, The number of users eavesdropping within the region. From base station to eavesdropper Channel gain; Based on the constructed scene and the obtained problem model, the nodes and connecting edges of the mapping graph of the scene and the problem are defined, in the graph neural network, a legal user and a corresponding eavesdropper thereof are set as a node, a full connection graph structure between all nodes, and the node initial features of the node are represented as: wherein is an operation that takes the real part of the beamforming vector of the base station to the user , is an operation that takes the imaginary part of the beamforming vector of the base station to the user , is an operation that takes the real part of the channel gain of the base station to the user , is an operation that takes the imaginary part of the channel gain of the base station to the user , is an operation that takes the real part of the channel gain of the base station to the eavesdropper , is an operation that takes the imaginary part of the channel gain of the base station to the eavesdropper . A graph neural network encoder is designed based on the solved problem model, and a message generation method and a message aggregation method in the graph neural network are defined; In the graph neural network encoder, set For the node In the first Feature in the message passing, the node In the first Generate the message to be passed in the second message passing , Wherein, It refers to the message generation function in the first The aggregation of neighbor messages occurs after the message passes along the edge in the graph data, the node receives the messages of multiple neighbors, and converts the message into a single fixed-dimensional aggregated information, using As an aggregation function to select the most influential message, the selected message participates in the aggregation process of neighbor messages as follows: Adopting the partial update method, part of the node features is reserved as an effective feature storage area of the original channel information, and the update variable of each layer is stored as the embedding feature of the remaining part of the node feature, Wherein, The first Term of represents the part of the node feature that stores the embedding feature, and the part is the part of the stored feature that is updated in each round of message passing process; A loss function, a learning method and parameters of the neural network are designed, the neural network is trained, and a model with generalization ability is obtained, and the loss function of the neural network is defined as a negative value of the system security rate: wherein, is the system security rate, is the legitimate user receives the target signal at a rate, is the eavesdropper receives the target signal at a rate, is the number of user-eavesdropper pairs in the region, by adjusting the parameters in multiple rounds of training, the loss function converges, and the beamforming strategy for system security communication is obtained.

2. The physical layer security transmission method based on a graph neural network according to claim 1, characterized in that, In a scenario where there are several eavesdroppers and legitimate users receiving the base station downlink communication signals, the multi-antenna base station transmits signals to uniformly randomly distributed single-antenna legitimate users in a region, and there are also one-to-one corresponding eavesdroppers in the region who simultaneously eavesdrop on the corresponding user signals.

3. The physical layer security transmission method based on a graph neural network according to claim 1, characterized in that, System safety rate For:

4. A physical layer security transmission system based on graph neural networks, characterized in that, The method for implementing the physical layer security transmission based on the graph neural network in claim 1 comprises: A scenario module, which constructs a scenario in which a plurality of eavesdroppers and legitimate users receive downlink communication signals of a base station; A signal module, which obtains channel gains between a base station transmitter and legitimate users and eavesdroppers based on given directions of signals arriving at the base station transmitter, and determines expressions of received signals at the legitimate users and the eavesdroppers; A function module, which constructs a problem model based on the obtained expressions of the received signals at the legitimate users and the eavesdroppers, in which a transmission beamforming vector corresponding to each legitimate user is taken as a variable, and a maximum security rate of the anti-eavesdropping system is taken as an objective function; A network module, which defines nodes and connecting edges of a graph mapped by the constructed scenario and the obtained problem model based on the constructed scenario and the obtained problem model, designs a graph neural network encoder based on the solved problem model, and defines a message generation method and a message aggregation method in the graph neural network; An improvement module, which designs a loss function, a learning method and parameters of the neural network, trains the neural network, and obtains a model with generalization ability.

Images