IRS-NOMA system security resource optimization method, system and device based on graph neural network and medium

Abstract

An IRS-NOMA system security resource optimization method, system and device based on a graph neural network, and a medium, the method comprising: a base station first sends a signal to be transmitted, the signal is received by a legal user through a direct link and an indirect link, and the transmission rate of the legal user signal is calculated; the method comprises the steps that firstly, an IRS-assisted NOMA system is established, then the eavesdropping rate of an external eavesdropper or an internal eavesdropper in the IRS-assisted NOMA system for eavesdropping and receiving signals is calculated, then the communication process is analyzed and deduced to obtain a secrecy rate formula, an optimization objective function of the IRS-assisted NOMA system is determined, and finally performance optimization is conducted on the optimization objective function through a neural network. An optimization result is obtained; systems, devices, and media for implementing the methods thereof; the method has the advantages of being low in communication cost, low in design complexity, comprehensive in design, simple and easy to optimize, high in optimization efficiency, high in optimization adaptability and flexibility and high in secrecy rate.

Description

Technical Field

[0001] The present invention relates to the fields of wireless communication security, intelligent reflecting surface and non-orthogonal multiple access technology, and in particular to a method, system, device and medium for optimizing security resources of an IRS-NOMA system based on a graph neural network. Background Art

[0002] With the advancement of science and technology, the demand for high capacity and low latency in wireless communications has increased dramatically, driving continuous technological innovation. Non-Orthogonal Multiple Access (NOMA) technology has attracted much attention in academia and industry for its efficient spectrum utilization, wide connectivity, and enhanced user fairness. However, while NOMA improves spectrum efficiency and user fairness, it also faces security threats such as eavesdropping, which challenges transmission confidentiality and integrity and poses a severe test to the security of wireless communication systems. Physical layer security (PLS) technology uses wireless channel characteristics (such as attenuation, multipath, and interference) to enhance information confidentiality and integrity. Among them, the Intelligent Reflecting Surface (IRS) technology is particularly critical. It can dynamically adjust the signal phase, enhance the legitimate user's reception signal through optimized configuration, and weaken the eavesdropper's interception ability, opening up a new way to improve the transmission performance and security of the NOMA system.

[0003] Combining NOMA and IRS technology, an efficient and secure wireless communication system can be built, which not only meets the needs of high capacity and low latency, but also effectively resists eavesdropping and ensures information confidentiality and integrity. However, there are currently few studies on resource optimization and performance analysis of IRS-assisted NOMA systems to achieve secure communication, and there are some problems. First, many studies have a one-sided consideration of security performance, only considering malicious external eavesdroppers, while ignoring possible rebellious internal eavesdroppers, and cannot provide a comprehensive security analysis. In addition, many studies need to use additional artificial noise to interfere with eavesdroppers to suppress the eavesdroppers' acquisition of signals. The introduction of noise increases the cost and complexity of the system. At the same time, most studies use traditional optimization methods to obtain resource allocation and performance results. These methods not only have low optimization efficiency, but also have weak adaptability and flexibility.

[0004] The Chinese patent application with publication number CN115002802A discloses a method for maximizing the security rate of an IRS-assisted NOMA drone network. However, since the patent application only considers external eavesdroppers and uses traditional optimization methods, the system is not comprehensive and the optimization efficiency is low.

[0005] Wei Wang et al. published Beamforming and Jamming Optimization for IRS-Aided Secure NOMA Networks (W. Wang et al., "Beamforming and Jamming Optimization for IRS-Aided Secure NOMA Networks," in IEEE Transactions on Wireless Communications, vol. 21, no. 3, pp. 1557-1569, March 2022, doi: 10.1109 / TWC.2021.3104856.). The paper proposes to enhance the secure communication of the IRS-assisted NOMA system with the help of artificial noise. The base station needs to combine the NOMA signal with the artificial noise and send it to the user together. The user can separate the artificial noise to obtain the NOMA signal, while the eavesdropper is interfered by the artificial noise and cannot obtain the NOMA signal. However, since this design requires more communication resources, it has the disadvantages of high complexity and high communication cost. Summary of the invention

[0006] In order to overcome the shortcomings of the above-mentioned prior art, the purpose of the present invention is to provide an IRS-NOMA system security resource optimization method, system, device and medium based on graph neural network. By designing external and internal eavesdropper models and constructing a graph neural network, a resource optimization and performance analysis method of an IRS-assisted NOMA system is realized, which has the characteristics of low communication cost, low design complexity, comprehensive design, simple and easy optimization, high optimization efficiency, strong adaptability and flexibility of optimization, and high confidentiality rate.

[0007] In order to achieve the above object, the technical solution adopted by the present invention is:

[0008] A security resource optimization method for an IRS-NOMA system based on a graph neural network includes the following steps:

[0009] Step 1: The base station sends the signal to be transmitted, and the signal is received by the legitimate user through the direct link and the indirect link, and the transmission rate of the legitimate user signal is calculated. The direct link is the base station directly broadcasting to the user, and the indirect link is the base station transmitting to the IRS, which is decoded by the IRS and then broadcast to the user;

[0010] Step 2: Calculate the eavesdropping rate of the external eavesdropper or the internal eavesdropper in the IRS-assisted NOMA system to eavesdrop on the signal broadcast in step 1;

[0011] Step 3: The confidentiality rate formula is obtained by analyzing and deriving the communication process from step 1 to step 2, and then the optimization objective function of the IRS-assisted NOMA system is clarified;

[0012] Step 4: Use a neural network to optimize the performance of the optimization objective function analyzed and derived in step 3 to obtain the optimization result.

[0013] The step 1 specifically includes:

[0014] All channel coefficients are set to exhibit small-scale fading with path loss, following the Rayleigh fading model. The coefficient of the channel from the base station to the IRS is The coefficient of the channel from the base station to user k is The coefficient of the channel from the base station to the external eavesdropper is The coefficient of the channel from the base station to the internal eavesdropper is The coefficient of the channel from IRS to user k is The coefficient of the channel from IRS to the external eavesdropper is The coefficient of the channel from IRS to the internal eavesdropper is set up is the signal to be transmitted, then among all legal users, the signal that the kth user can receive is:

[0015]

[0016] Among them, K is the number of users, k is the user, T is the matrix transpose operation, and w k is the beamforming matrix, a k is the NOMA power allocation factor, x k is the signal to be transmitted, P t is the transmission power of the base station, is the IRS phase shift matrix, where θ n is the nth reflection coefficient of the phase shift matrix, n k is Gaussian white noise;

[0017] In the IRS-assisted NOMA system, users use the Successive Interference Cancellation (SIC) technology to decode signals, and the order of user demodulation is ||h1|| 2 ≥||h2|| 2 ≥…≥||h K || 2, where the kth ranked user corresponds to the kth strongest channel. The kth strongest user decodes the signal by treating the signals of other weaker users as interference. Therefore, the signal-to-interference-plus-noise ratio (SINR) of the kth user when decoding the signal is expressed as:

[0018]

[0019] Among them, a k is the NOMA power allocation factor, w k is the beamforming matrix, σ k is the noise at user k;

[0020] Then the transmission rate of the kth legal user is expressed as:

[0021] R k =log2(1+γ k )(1-3).

[0022] The step 2 specifically includes:

[0023] The signal received by the external eavesdropper is derived as:

[0024]

[0025] Where e is an external eavesdropper, K is the number of users, k is a user, T is a matrix transpose operation, and h is b,e is the coefficient of the channel from the base station to the external eavesdropper, h r,e is the coefficient of the channel from IRS to the external eavesdropper, G is the coefficient of the channel from the base station to IRS, is the IRS phase shift matrix, where θ n is the nth reflection coefficient of the phase shift matrix, w k is the beamforming matrix, P t is the transmission power of the base station, a k is the NOMA power allocation factor, x k is the signal to be transmitted, n e is the Gaussian white noise at the eavesdropper;

[0026] The SINR of the k-th user signal eavesdropped by an external eavesdropper is expressed as:

[0027]

[0028] Among them, a k is the NOMA power allocation factor, w k is the beamforming matrix, σ k is the noise at user k;

[0029] Then the eavesdropping rate of the external eavesdropper is expressed as:

[0030] R e→k =log2(1+γ e→k ) (2-3)

[0031] The signal received by the internal eavesdropper is expressed as:

[0032]

[0033] Among them, f is the internal eavesdropper, n f is the Gaussian white noise at the eavesdropper, h b,f is the coefficient of the channel from the base station to the internal eavesdropper, h r,k is the coefficient of the channel from IRS to user k;

[0034] The SINR of the internal eavesdropper eavesdropping on the kth user is expressed as:

[0035]

[0036] Among them, w f is the beamforming at the internal eavesdropper f, σ f is the noise at the eavesdropper's location;

[0037] Then the eavesdropping rate of the external eavesdropper can be expressed as:

[0038] R f→k =log2(1+γ f→k ) (2-6).

[0039] The step 3 specifically includes:

[0040] The formula for calculating the confidentiality rate of the IRS-assisted NOMA system is expressed as:

[0041]

[0042] Among them, R k is the transmission rate of the kth user, is the eavesdropping rate of the eavesdropper eavesdropping on user k, e is the external eavesdropper, and f is the internal eavesdropper;

[0043] Through the mapping function g, the input signal Y is mapped to the optimized beamforming, power allocation factor and phase shift matrix. The optimization objective function is written as:

[0044]

[0045] Among them, a and a k is the NOMA power allocation factor, w and w kis the beamforming matrix, Φ is the IRS phase shift matrix, K is the number of users, k is the number of users, h bk is the coefficient of the channel from the base station to user k, h rk is the coefficient of the channel from IRS to user k, P t is the transmission power of the base station, is the element in the IRS phase shift matrix, where θ n is the nth reflection coefficient of the phase shift matrix.

[0046] The step 4 specifically includes:

[0047] Step 4.1: Build the network;

[0048] The use of collaborative graph neural network (CO-GNN) aims to learn the received signal Y through a multi-layer architecture consisting of an input layer, two message passing layers and an output layer. The received signal as input contains the information required for the joint optimization of the initial input values. These inputs are propagated through two message passing layers containing aggregation and combination functions. Finally, at the output layer, a fully connected layer directly converts the representation into the optimized beamforming w, power allocation factor a and phase shift matrix Φ;

[0049] (1) Input layer: The input layer receives the feature vector Y of each user’s signal k , where k = 1,…,K. For user nodes, the feature representation of the input layer is:

[0050]

[0051] in, It is a multilayer perceptron (MLP) with two fully connected hidden layers, and T is the matrix transpose operation;

[0052] IRS is represented by the features of the input layer as:

[0053]

[0054] Among them, K is the number of users, k is the user;

[0055] (2) Message passing layer: The message passing layer uses neighbor aggregation and combination strategies to extract information from nodes to generate a representation vector containing beamforming, power allocation factors, and phase shift matrices to achieve joint optimization. The aggregation function of the user node is expressed as:

[0056]

[0057] Among them, i is the i-th aggregation layer, N(k) represents the set of neighboring nodes near node k, Θ is the mean pooling function, is a fully connected hidden layer;

[0058] The combined function of the user node is expressed as:

[0059]

[0060] The representation vector of an IRS node is:

[0061]

[0062] Combining aggregation and composition, the message passing layer can be expressed as:

[0063]

[0064] Among them, Δ is the maximum pooling function;

[0065] (3) Output layer: After completing I information transfer layers, the output layer generates a jointly optimized phase shift matrix Beamforming and power allocation factor a∈R K ;

[0066] The phase shift matrix is ​​derived from the IRS representation vector:

[0067]

[0068] The beamforming and power allocation factors are derived from the representation vector of the user node:

[0069]

[0070] Step 4.2: Train and optimize the network built in step 4.1;

[0071] Step 4.2.1: Set the loss function for network training optimization to the optimization objective function Negative value of

[0072] Step 4.2.2: Randomly generate the coefficients of the channel from the base station to the IRS: The coefficient of the channel from the base station to user k is The coefficient of the channel from the base station to the external eavesdropper is The coefficient of the channel from the base station to the internal eavesdropper is The coefficient of the channel from IRS to user k is The coefficient of the channel from IRS to the external eavesdropper is The coefficient of the channel from IRS to the internal eavesdropper is and the input of the network, i.e. the received signal Y;

[0073] Step 4.2.3: Set the number of training optimization rounds of the network to 100 rounds, input the channel coefficient and the received signal Y into the network, and start training;

[0074] Step 4.2.4: Based on the network input and training optimization parameters set in steps 4.2.1 to 4.2.3, perform 100 rounds of training optimization on the input layer, message passing layer, and output layer of the network built in step 4.1 to obtain the optimization results of beamforming, power allocation factor, and phase shift matrix;

[0075] Step 4.2.5: Calculate the optimized objective function based on the beamforming, power allocation factor and phase shift matrix obtained in step 4.2.4 result.

[0076] A resource optimization system for an IRS-assisted NOMA system with eavesdroppers based on graph neural networks, comprising:

[0077] Security model module: The base station sends the signal to be transmitted, and the signal is received by the legitimate user through the direct link and the indirect link, and the transmission rate of the legitimate user signal is calculated, and the eavesdropping rate of the signal eavesdropped by the external eavesdropper or the internal eavesdropper in the IRS-assisted NOMA system is calculated. Among them, the direct link is the base station directly broadcasting to the user, and the indirect link is the base station transmitting to the IRS, and the IRS decodes and then broadcasts to the user;

[0078] Optimization target confirmation module: derives the confidentiality rate formula by analyzing and deducing the communication process, and then clarifies the optimization target function of the IRS-assisted NOMA system;

[0079] Network training module: Use neural network to optimize the performance of the optimization objective function and obtain the optimization result.

[0080] A security resource optimization device for an IRS-NOMA system based on a graph neural network, comprising:

[0081] Memory: used to store a computer program for implementing the IRS-NOMA system security resource optimization method based on graph neural network;

[0082] Processor: used to implement the IRS-NOMA system security resource optimization method based on graph neural network when executing the computer program.

[0083] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of a method for optimizing security resources of an IRS-NOMA system based on a graph neural network.

[0084] Compared with the prior art, the beneficial effects of the present invention are:

[0085] 1. By analyzing the signal receiving behaviors of legitimate users, external eavesdroppers and internal eavesdroppers, the present invention realizes the IRS-assisted NOMA system design with eavesdroppers. It not only provides a comprehensive and systematic perspective for analysis and design, but also has low communication cost and design complexity.

[0086] 2. The present invention realizes the optimization of various resources of the IRS-assisted NOMA system by constructing a collaborative graph neural network (CO-GNN), which is simple and easy to implement, has high optimization efficiency, and strong adaptability and flexibility.

[0087] 3. The present invention maximizes the total confidentiality rate by establishing the optimization target of the IRS-assisted NOMA system and using CO-GNN for network optimization training, and can provide a resource optimization performance analysis solution with a high confidentiality rate.

[0088] In summary, the present invention has the effects of low communication cost, low design complexity, comprehensive design, simple and easy optimization, high optimization efficiency, strong optimization adaptability and flexibility, and high confidentiality rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0089] Figure 1 This is a diagram of the NOMA system security model assisted by IRS based on graph neural network.

[0090] Figure 2 This is a simulation experiment result diagram of the relationship between the confidentiality rate and the transmission power in the external eavesdropping scenario in this embodiment.

[0091] Figure 3 This is a simulation experiment result diagram of the relationship between confidentiality rate and transmission power in the internal eavesdropping scenario in this embodiment.

[0092] Figure 4 This is a simulation experiment result diagram of the relationship between the confidentiality rate and the number N of reflective elements equipped with the IRS in the external eavesdropping scenario in this embodiment.

[0093] Figure 5 This is a simulation experiment result diagram of the relationship between the confidentiality rate and the number N of reflective elements equipped with the IRS in the internal eavesdropping scenario in this embodiment.

[0094] Figure 6 4 is a simulation experiment result diagram of the relationship between the confidentiality rate and the number of antennas M equipped in the base station in the external eavesdropping scenario in this embodiment.

[0095] Figure 7 4 is a simulation experiment result diagram of the relationship between the confidentiality rate and the number of antennas M equipped in the base station in the internal eavesdropping scenario in this embodiment. DETAILED DESCRIPTION

[0096] The present invention will be described in detail below in conjunction with the accompanying drawings.

[0097] Example

[0098] See also Figure 1 , a method for optimizing security resources of IRS-NOMA system based on graph neural network is applied to communication in wireless network scenario. The communication adopts IRS-assisted NOMA system. The system includes a base station, an IRS, K users, and two types of eavesdroppers: a malicious external eavesdropper or a "rebellious" internal eavesdropper. The base station is equipped with M antennas, the IRS is equipped with N reflective elements, and the users and eavesdroppers are equipped with a single antenna. The following steps are included:

[0099] Step 1: The base station sends the signal to be transmitted, and the signal is received by the legitimate user through the direct link and the indirect link, and the transmission rate of the legitimate user signal is calculated. The direct link is the base station directly broadcasting to the user, and the indirect link is the base station transmitting to the IRS, which is decoded by the IRS and then broadcast to the user;

[0100] Step 2: Calculate the eavesdropping rate of the signal eavesdropped by an external eavesdropper or an internal eavesdropper in the IRS-assisted NOMA system;

[0101] Step 3: The confidentiality rate formula is obtained by analyzing and deriving the communication process from step 1 to step 2, and then the optimization objective function of the IRS-assisted NOMA system is clarified;

[0102] Step 4: Use a neural network to optimize the performance of the optimization objective function analyzed and derived in step 3 to obtain the optimization result.

[0103] The step 1 specifically includes:

[0104] All channel coefficients are set to exhibit small-scale fading with path loss, following the Rayleigh fading model. The coefficient of the channel from the base station to the IRS is The coefficient of the channel from the base station to user k is The coefficient of the channel from the base station to the external eavesdropper is The coefficient of the channel from the base station to the internal eavesdropper is The coefficient of the channel from IRS to user k is The coefficient of the channel from IRS to the external eavesdropper is The coefficient of the channel from IRS to the internal eavesdropper is set up is the signal to be transmitted, then among all legal users, the signal that the kth user can receive is:

[0105]

[0106] Among them, K is the number of users, k is the user, T is the matrix transpose operation, and w k is the beamforming matrix, a k is the NOMA power allocation factor, x k is the signal to be transmitted, P t is the transmission power of the base station, is the IRS phase shift matrix, where θ n is the nth reflection coefficient of the phase shift matrix, n k is Gaussian white noise;

[0107] In the IRS-assisted NOMA system, users use the Successive Interference Cancellation (SIC) technology to decode signals. The demodulation order in SIC is determined according to the equivalent channel gain of the combined channel. The user with the strongest channel gain demodulates first, while the user with weaker channel gain directly receives the information. The order of user demodulation is ||h1|| 2 ≥||h2|| 2 ≥…≥||h K || 2 , where the kth ranked user corresponds to the kth strongest channel. The kth strongest user decodes the signal by treating the signals of other weaker users as interference. Therefore, the signal-to-interference-plus-noise ratio (SINR) of the kth user when decoding the signal is expressed as:

[0108]

[0109] Among them, a k is the NOMA power allocation factor, w k is the beamforming matrix, σ k is the noise at user k;

[0110] Then the transmission rate of the kth legal user is expressed as:

[0111] R k =log2(1+γ k ) (1-3).

[0112] The step 2 specifically includes:

[0113] Considering the adverse effects of complex electromagnetic environment on secure communications through exposed radio signals, external and internal eavesdropping scenarios are set up to analyze the security and confidentiality performance of the IRS-assisted NOMA system.

[0114] The external eavesdropper is an unauthorized attacker who is not a member of the NOMA pairing. The broadcast characteristics of the wireless network enable the external eavesdropper to eavesdrop on the signal broadcast by the IRS base station. Therefore, the signal received by the external eavesdropper is derived as follows:

[0115]

[0116] Where e is an external eavesdropper, K is the number of users, k is a user, T is a matrix transpose operation, and h is b,e is the coefficient of the channel from the base station to the external eavesdropper, h r,e is the coefficient of the channel from IRS to the external eavesdropper, G is the coefficient of the channel from the base station to IRS, is the IRS phase shift matrix, where θ n is the nth reflection coefficient of the phase shift matrix, w k is the beamforming matrix, P t is the transmission power of the base station, a k is the NOMA power allocation factor, x k is the signal to be transmitted, n e is the Gaussian white noise at the eavesdropper;

[0117] The external eavesdropper decodes the information of the kth user by applying serial interference cancellation. Therefore, the SINR of the kth user signal eavesdropped by the external eavesdropper is expressed as:

[0118]

[0119] Among them, a k is the NOMA power allocation factor, w k is the beamforming matrix, σ k is the noise at user k;

[0120] Then the eavesdropping rate of the external eavesdropper is expressed as:

[0121] R e→k =log2(1+γ e→k ) (2-3)

[0122] The internal eavesdropper is a legitimate user that can perform NOMA pairing, but because its channel condition is relatively weak compared to other users, it can act as an internal eavesdropper to eavesdrop on other legitimate users. The signal received by the internal eavesdropper is expressed as:

[0123]

[0124] Among them, f is the internal eavesdropper, n f is the Gaussian white noise at the eavesdropper, h b,f is the coefficient of the channel from the base station to the internal eavesdropper, h r,kis the coefficient of the channel from IRS to user k;

[0125] The SINR of the internal eavesdropper eavesdropping on the kth user is expressed as:

[0126]

[0127] Among them, w f is the beamforming at the internal eavesdropper f, σ f is the noise at the eavesdropper's location;

[0128] Then the eavesdropping rate of the external eavesdropper can be expressed as:

[0129] R f→k =log2(1+γ f→k ) (2-6).

[0130] The step 3 specifically includes:

[0131] The goal of the present invention is to maximize the total confidentiality rate of the IRS-assisted NOMA system by jointly optimizing the base station beamforming w and power allocation factor a, and the IRS phase shift matrix Φ in the IRS-NOMA network security model, and obtain the corresponding performance analysis. By directly mapping the received signal Y to the optimized beamforming, power allocation factor and phase shift matrix to maximize the utility, the channel estimation can be bypassed and the optimized transmission strategy can be obtained more efficiently. The formula for calculating the confidentiality rate of the IRS-assisted NOMA system is expressed as:

[0132]

[0133] Among them, R k is the transmission rate of the kth user, is the eavesdropping rate of the eavesdropper eavesdropping on user k, e is the external eavesdropper, and f is the internal eavesdropper;

[0134] Through the mapping function g, the input signal Y is mapped to the optimized beamforming, power allocation factor and phase shift matrix. The optimization objective function is written as:

[0135]

[0136] Among them, a and a k is the NOMA power allocation factor, w and w k is the beamforming matrix, Φ is the IRS phase shift matrix, K is the number of users, k is the number of users, h bk is the coefficient of the channel from the base station to user k, h rk is the coefficient of the channel from IRS to user k, P t is the transmission power of the base station, is the element in the IRS phase shift matrix, where θn is the nth reflection coefficient of the phase shift matrix.

[0137] Due to the non-convexity of this objective function, solving it computationally is very challenging. To solve this problem efficiently and bypass the need for channel estimation, the mapping function g(·) is visualized through a neural network and the received signal Y is used to train the network parameters to achieve the best overall confidentiality rate.

[0138] The step 4 specifically includes:

[0139] Due to the non-convexity of the objective function, the use of traditional optimization methods is often challenging. At the same time, it is usually impractical to apply traditional optimization techniques in dynamically changing environments to formulate effective security joint optimization strategies. Based on this, the present invention introduces a neural network, the collaborative graph neural network (CO-GNN), to simulate the mapping function g(·). Using the collaborative graph neural network (CO-GNN), the received signal Y can be directly mapped to the base station beamforming w, the power allocation factor a and the smart reflection surface phase shift matrix Φ. Through continuous training and optimization, the total confidentiality rate of the system can be maximized. By using the opposite of the objective function as the loss function of the network, joint optimization can be achieved by suppressing malicious eavesdroppers according to the instructions of the objective function, thereby ensuring that the signal is transmitted in a direction that is beneficial to legitimate users. Specifically, this step mainly includes the network construction and network training process:

[0140] Step 4.1: Build the network;

[0141] The use of collaborative graph neural network (CO-GNN) aims to learn the received signal Y through a multi-layer architecture consisting of an input layer, two message passing layers and an output layer. The received signal as input contains the information required for the joint optimization of the initial input values. These inputs are propagated through two message passing layers containing aggregation and combination functions. Finally, at the output layer, a fully connected layer directly converts the representation into the optimized beamforming w, power allocation factor a and phase shift matrix Φ;

[0142] (1) Input layer: The input layer receives the feature vector Y of each user’s signal k , where k = 1,…,K, the signal Y contains rich information related to beamforming, power allocation and phase shift matrix, which is beneficial to the feature learning and joint optimization of the CO-GNN network. Since the signal contains real and imaginary parts, for the user node, the feature representation through the input layer is:

[0143]

[0144] in, It is a multilayer perceptron (MLP) with two fully connected hidden layers, and T is the matrix transpose operation;

[0145] For IRS, since all signals are reflected by IRS, it is necessary to ensure that the received signals contain equal information related to IRS. Therefore, the feature representation of IRS through the input layer is:

[0146]

[0147] Among them, K is the number of users, k is the user;

[0148] (2) Message passing layer: The message passing layer uses neighbor aggregation and combination strategies to extract information from nodes to generate a representation vector containing beamforming, power allocation factors, and phase shift matrices to achieve joint optimization. The aggregation function compiles the features of neighboring nodes into a message vector and passes it to the central node. The combination function updates the representation of the node at the current moment and merges the current representation of the node with the message collected from the aggregation function, which is actually equivalent to an information transmission process. The aggregation function of the user node is expressed as:

[0149]

[0150] Among them, i is the i-th aggregation layer, N(k) represents the set of neighboring nodes near node k, Θ is the mean pooling function, is a fully connected hidden layer;

[0151] The combined function of the user node is expressed as:

[0152]

[0153] The representation vector of an IRS node is:

[0154]

[0155] Combining aggregation and composition, the message passing layer can be expressed as:

[0156]

[0157] Among them, Δ is the maximum pooling function;

[0158] (3) Output layer: After completing I information transfer layers, the output layer generates a jointly optimized phase shift matrix Beamforming and power allocation factor a∈R K ;

[0159] The phase shift matrix is ​​derived from the IRS representation vector:

[0160]

[0161] The beamforming and power allocation factors are derived from the representation vector of the user node:

[0162]

[0163] Step 4.2: Train and optimize the network built in step 4.1;

[0164] Step 4.2.1: Set the loss function for network training optimization to the optimization objective function Negative value of

[0165] Step 4.2.2: Randomly generate the coefficients of the channel from the base station to the IRS: The coefficient of the channel from the base station to user k is The coefficient of the channel from the base station to the external eavesdropper is The coefficient of the channel from the base station to the internal eavesdropper is The coefficient of the channel from IRS to user k is The coefficient of the channel from IRS to the external eavesdropper is The coefficient of the channel from IRS to the internal eavesdropper is and the input of the network, i.e. the received signal Y;

[0166] Step 4.2.3: Set the number of training optimization rounds of the network to 100 rounds, input the channel coefficient and the received signal Y into the network, and start training;

[0167] Step 4.2.4: Based on the network input and training optimization parameters set in steps 4.2.1 to 4.2.3, perform 100 rounds of training optimization on the input layer, message passing layer, and output layer of the network built in step 4.1 to obtain the optimization results of beamforming, power allocation factor, and phase shift matrix;

[0168] Step 4.2.5: Calculate the optimized objective function based on the beamforming, power allocation factor and phase shift matrix obtained in step 4.2.4 result.

[0169] In network training, the loss function is defined as the negative value of the total confidentiality rate. In each iteration of training, the loss is iteratively fed back to the network to guide the optimization of beamforming, power allocation factor and phase shift matrix, so as to maximize the transmission rate of legitimate users while minimizing the eavesdropping rate. As the value of the loss function decreases, the total confidentiality rate performance of the system also improves synchronously, and finally the joint optimization of parameters can be achieved to obtain the optimal total confidentiality rate. The network training is conducted separately for external eavesdropping scenarios and internal eavesdropping scenarios.

[0170] A security resource optimization method system for an IRS-NOMA system based on a graph neural network, comprising:

[0171] Security model module: The base station sends the signal to be transmitted, and the signal is received by the legitimate user through the direct link and the indirect link, and the transmission rate of the legitimate user signal is calculated, and the eavesdropping rate of the signal eavesdropped by the external eavesdropper or the internal eavesdropper in the IRS-assisted NOMA system is calculated. Among them, the direct link is the base station directly broadcasting to the user, and the indirect link is the base station transmitting to the IRS, and the IRS decodes and then broadcasts to the user, which is used to implement steps 1 to 2 of a security resource optimization method for an IRS-NOMA system based on a graph neural network;

[0172] Optimization target confirmation module: The confidentiality rate formula is obtained by analyzing and deducing the communication process, and then the optimization objective function of the IRS-assisted NOMA system is clarified, which is used to implement step 3 of a security resource optimization method for an IRS-NOMA system based on a graph neural network;

[0173] Network training module: The performance of the optimization objective function is optimized through a neural network to obtain the optimization result, which is used to implement step 4 of a security resource optimization method for an IRS-NOMA system based on a graph neural network.

[0174] A security resource optimization device for an IRS-NOMA system based on a graph neural network, comprising:

[0175] Memory: used to store a computer program for implementing the IRS-NOMA system security resource optimization method based on graph neural network;

[0176] Processor: used to implement the IRS-NOMA system security resource optimization method based on graph neural network when executing the computer program.

[0177] The present invention also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the steps of the IRS-NOMA system security resource optimization method based on graph neural network.

[0178] Simulation Analysis

[0179] Figure 2The relationship between the confidentiality rate and the transmission power using the method proposed in the present invention and other different optimization situations in the presence of external eavesdroppers is described. The specific simulation conditions are as follows: the number of antennas equipped in the base station is M=5, the number of reflective elements equipped in the IRS is N=100, and the number of legal users is K=2. As can be seen from the figure, with the increase of the transmission power, the confidentiality rate continues to increase. This is because in the process of joint optimization of beamforming, power allocation factor and phase shift matrix, the increase in power can make the transmission channel biased towards legal users and avoid eavesdropping by external illegal eavesdroppers as much as possible. At the same time, the method proposed in the present invention achieved the highest confidentiality rate among all the schemes, which illustrates the rationality and effectiveness of the optimization scheme.

[0180] Figure 3 The relationship between the confidentiality rate and the transmission power under different optimization conditions using the proposed method in the scenario of the presence of an internal eavesdropper is described. The specific simulation conditions are as follows: the number of antennas equipped in the base station is M = 5, the number of reflective elements equipped in the IRS is N = 100, and the number of legal users is K = 2. Figure 2 In comparison, the confidentiality rate in the presence of internal eavesdropping is higher than that in the external eavesdropping scenario. This is due to the characteristics of the internal eavesdropping node itself - it used to be a legitimate node, and the neural network can more effectively improve its transmission performance. So when it exists as an eavesdropper, the neural network can also correspondingly better suppress its transmission to obtain higher system security performance.

[0181] Figure 4 The relationship between the confidentiality rate and the number of reflective elements N equipped in the IRS using the method proposed in this invention and other different optimization conditions in the presence of an external eavesdropper is described. The specific simulation conditions are as follows: the number of antennas equipped in the base station is M = 5, the number of legal users is K = 2, and the transmission power P t = 30. Except for the “random IRS phase shift matrix” scheme, whose rate remains relatively stable and is insensitive to changes in the number of IRS elements, all other schemes with IRS assistance significantly improve security performance as N increases. By deploying an IRS with more reflection elements, user links can be effectively enhanced and external eavesdropping can be suppressed, thereby ensuring the security of legitimate users.

[0182] Figure 5 The relationship between the confidentiality rate and the number of reflective elements N equipped in the IRS using the method proposed in this invention and other different optimization conditions in the presence of an internal eavesdropper is described. The specific simulation conditions are as follows: the number of transmission antennas is M = 5, the number of legal users is K = 2, and the transmission power P is t =30. The scheme proposed in this invention shows the highest confidentiality rate. In addition, the trend observed in this figure is consistent with Figure 3The curves depicted in are very similar, which shows that in the internal eavesdropping scenario, the transmission power P t The effect of the number of reflective elements equipped with IRS on the confidentiality rate is similar.

[0183] Figure 6 and Figure 7 The relationship between the confidentiality rate and the number of transmission antennas M using the method proposed in this invention and other different optimization situations in the presence of external and internal eavesdroppers is described. The specific simulation conditions are as follows: the number of legitimate users is K = 2, the number of reflective elements equipped with the IRS is N = 100, and the transmission power P t =30. As can be seen from the figure, the confidentiality rate increases monotonically with the increase in the number of transmitting antennas. This is because the greater the number of antennas, the stronger the main link can be, providing a more centralized approach to confuse potential eavesdroppers by optimizing beamforming, power allocation, and phase shift matrices. Even when the number of antennas M is limited, the confidentiality rate in all experimental scenarios remains high. This robustness is attributed to the combined effect of the specified power level and the sufficient number of reflective elements N = 100 in the IRS, which together ensure the strong security of the system.

[0184] Obviously, those skilled in the art should understand that the above modules or steps of the present invention can be implemented by a general computing device, they can be concentrated on a single computing device, or distributed on a network composed of multiple computing devices, and optionally, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. Thus, the present invention is not limited to any specific combination of hardware and software.

[0185] The embodiments described above are merely illustrative. If a unit described as a separate component is involved, it may or may not be physically separated; if a component displayed as a unit is involved, it may or may not be a physical unit, that is, it may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the embodiment. Those skilled in the art can understand and implement it without creative work.

[0186] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention is described in detail with reference to the above embodiments, a person skilled in the art should understand that the technical solutions described in the above embodiments can still be modified, or some of the technical features can be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

[0187] The present invention is not limited to the above optional implementations, and anyone can derive other various forms of products under the enlightenment of the present invention. The above specific implementations should not be understood as limiting the scope of protection of the present invention. The scope of protection of the present invention should be based on the definition in the claims, and the description can be used to interpret the claims.

Claims

1. A security resource optimization method for IRS-NOMA system based on graph neural network, characterized in that: The following steps are involved: Step 1: The base station sends the signal to be transmitted, and the signal is received by the legitimate user through the direct link and the indirect link, and the transmission rate of the legitimate user signal is calculated. The direct link is the base station directly broadcasting to the user, and the indirect link is the base station transmitting to the IRS, which is decoded by the IRS and then broadcast to the user; Step 2: Calculate the eavesdropping rate of the external eavesdropper or the internal eavesdropper in the IRS-assisted NOMA system to eavesdrop on the signal broadcast in step 1; Step 3: The confidentiality rate formula is obtained by analyzing and deriving the communication process from step 1 to step 2, and then the optimization objective function of the IRS-assisted NOMA system is clarified; Step 4: Use a neural network to optimize the performance of the optimization objective function analyzed and derived in step 3 to obtain the optimization result.

2. According to claim 1, a method for optimizing security resources of an IRS-NOMA system based on a graph neural network is characterized in that: The step 1 specifically includes: All channel coefficients are set to exhibit small-scale fading with path loss, following the Rayleigh fading model. The coefficient of the channel from the base station to the IRS is The coefficient of the channel from the base station to user k is The coefficient of the channel from the base station to the external eavesdropper is The coefficient of the channel from the base station to the internal eavesdropper is The coefficient of the channel from IRS to user k is The coefficient of the channel from IRS to the external eavesdropper is The coefficient of the channel from IRS to the internal eavesdropper is set up is the signal to be transmitted, then among all legal users, the signal that the kth user can receive is: Among them, K is the number of users, k is the user, T is the matrix transpose operation, and w k is the beamforming matrix, a k is the NOMA power allocation factor, x k is the signal to be transmitted, P t is the transmission power of the base station, is the IRS phase shift matrix, where θ n is the nth reflection coefficient of the phase shift matrix, n k is Gaussian white noise; In the IRS-assisted NOMA system, users use the Successive Interference Cancellation (SIC) technology to decode signals, and the order of user demodulation is ||h1|| 2 ≥||h2|| 2 ≥…≥||h K || 2 , where the kth ranked user corresponds to the kth strongest channel. The kth strongest user decodes the signal by treating the signals of other weaker users as interference. Therefore, the signal-to-interference-plus-noise ratio (SINR) of the kth user when decoding the signal is expressed as: Among them, a k is the NOMA power allocation factor, w k is the beamforming matrix, σ k is the noise at user k; Then the transmission rate of the kth legal user is expressed as: R k =log2(1+γ k (1-3).

3. According to claim 1, a method for optimizing security resources of an IRS-NOMA system based on a graph neural network is characterized in that: The step 2 specifically includes: The signal received by the external eavesdropper is derived as: Where e is an external eavesdropper, K is the number of users, k is a user, T is a matrix transpose operation, and h is b,e is the coefficient of the channel from the base station to the external eavesdropper, h r,e is the coefficient of the channel from IRS to the external eavesdropper, G is the coefficient of the channel from the base station to IRS, is the IRS phase shift matrix, where θ n is the nth reflection coefficient of the phase shift matrix, w k is the beamforming matrix, P t is the transmission power of the base station, a k is the NOMA power allocation factor, x k is the signal to be transmitted, n e is the Gaussian white noise at the eavesdropper; The SINR of the k-th user signal eavesdropped by an external eavesdropper is expressed as: Among them, a k is the NOMA power allocation factor, w k is the beamforming matrix, σ k is the noise at user k; The eavesdropping rate of the external eavesdropper is expressed as: R e→k =log2(1+γ e→k ) (2-3) The signal received by the internal eavesdropper is expressed as: Among them, f is the internal eavesdropper, n f is the Gaussian white noise at the eavesdropper, h b,f is the coefficient of the channel from the base station to the internal eavesdropper, h r,k is the coefficient of the channel from IRS to user k; The SINR of the internal eavesdropper eavesdropping on the kth user is expressed as: Among them, w f is the beamforming at the internal eavesdropper f, σ f is the noise at the eavesdropper's location; Then the eavesdropping rate of the external eavesdropper can be expressed as: R f→k =log2(1+γ f→k ) (2-6).

4. According to claim 1, a method for optimizing security resources of an IRS-NOMA system based on a graph neural network is characterized in that: The step 3 specifically includes: The formula for calculating the confidentiality rate of the IRS-assisted NOMA system is expressed as: Among them, R k is the transmission rate of the kth user, is the eavesdropping rate of the eavesdropper eavesdropping on user k, e is the external eavesdropper, and f is the internal eavesdropper; Through the mapping function g, the input signal Y is mapped to the optimized beamforming, power allocation factor and phase shift matrix. The optimization objective function is written as: Among them, a and a k is the NOMA power allocation factor, w and w k is the beamforming matrix, Φ is the IRS phase shift matrix, K is the number of users, k is the number of users, h bk is the coefficient of the channel from the base station to user k, h rk is the coefficient of the channel from IRS to user k, P t is the transmission power of the base station, is the element in the IRS phase shift matrix, where θ n is the nth reflection coefficient of the phase shift matrix.

5. According to claim 1, a method for optimizing security resources of an IRS-NOMA system based on a graph neural network is characterized in that: The step 4 specifically includes: Step 4.1: Build the network; The use of collaborative graph neural network (CO-GNN) aims to learn the received signal Y through a multi-layer architecture consisting of an input layer, two message passing layers and an output layer. The received signal as input contains the information required for the joint optimization of the initial input values. These inputs are propagated through two message passing layers containing aggregation and combination functions. Finally, at the output layer, a fully connected layer directly converts the representation into the optimized beamforming w, power allocation factor a and phase shift matrix Φ; (1) Input layer: The input layer receives the feature vector Y of each user’s signal k , where k = 1,…,K. For user nodes, the feature representation of the input layer is: in, It is a multilayer perceptron (MLP) with two fully connected hidden layers, and T is the matrix transpose operation; IRS is represented by the features of the input layer as: Among them, K is the number of users, k is the user; (2) Message passing layer: The message passing layer uses neighbor aggregation and combination strategies to extract information from nodes to generate a representation vector containing beamforming, power allocation factors, and phase shift matrices to achieve joint optimization. The aggregation function of the user node is expressed as: Among them, i is the i-th aggregation layer, N(k) represents the set of neighboring nodes near node k, Θ is the mean pooling function, is a fully connected hidden layer; The combined function of the user node is expressed as: The representation vector of an IRS node is: Combining aggregation and composition, the message passing layer can be expressed as: Among them, Δ is the maximum pooling function; (3) Output layer: After completing I information transfer layers, the output layer generates a jointly optimized phase shift matrix Beamforming and power allocation factor a∈R K ; The phase shift matrix is ​​derived from the IRS representation vector: The beamforming and power allocation factors are derived from the representation vector of the user node: Step 4.2: Train and optimize the network built in step 4.1; Step 4.2.1: Set the loss function for network training optimization to the optimization objective function Negative value of Step 4.2.2: Randomly generate the coefficients of the channel from the base station to the IRS: The coefficient of the channel from the base station to user k is The coefficient of the channel from the base station to the external eavesdropper is The coefficient of the channel from the base station to the internal eavesdropper is The coefficient of the channel from IRS to user k is The coefficient of the channel from IRS to the external eavesdropper is The coefficient of the channel from IRS to the internal eavesdropper is and the input of the network, i.e. the received signal Y; Step 4.2.3: Set the number of training optimization rounds of the network to 100 rounds, input the channel coefficient and the received signal Y into the network, and start training; Step 4.2.4: Based on the network input and training optimization parameters set in steps 4.2.1 to 4.2.3, perform 100 rounds of training optimization on the input layer, message passing layer, and output layer of the network built in step 4.1 to obtain the optimization results of beamforming, power allocation factor, and phase shift matrix; Step 4.2.5: Calculate the optimized objective function based on the beamforming, power allocation factor and phase shift matrix obtained in step 4.2.4 result.

6. A system based on a method for optimizing security resources of an IRS-NOMA system based on a graph neural network according to any one of claims 1 to 5, characterized in that: include: Security model module: The base station sends the signal to be transmitted, and the signal is received by the legitimate user through the direct link and the indirect link, and the transmission rate of the legitimate user signal is calculated, and the eavesdropping rate of the signal eavesdropped by the external eavesdropper or the internal eavesdropper in the IRS-assisted NOMA system is calculated. Among them, the direct link is the base station directly broadcasting to the user, and the indirect link is the base station transmitting to the IRS, and the IRS decodes and then broadcasts to the user; Optimization target confirmation module: derives the confidentiality rate formula by analyzing and deducing the communication process, and then clarifies the optimization target function of the IRS-assisted NOMA system; Network training module: Use neural network to optimize the performance of the optimization objective function and obtain the optimization result.

7. A security resource optimization device for IRS-NOMA system based on graph neural network, characterized in that: include: Memory: used to store a computer program for implementing a method for optimizing security resources of an IRS-NOMA system based on a graph neural network as described in any one of claims 1 to 5; Processor: used to implement the IRS-NOMA system security resource optimization method based on graph neural network as described in any one of claims 1 to 5 when executing the computer program.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of a method for optimizing security resources of an IRS-NOMA system based on a graph neural network as described in any one of claims 1 to 5 are implemented.

Images