Intelligent metasurface secure transmission optimization method based on supervised deep neural network

Abstract

The application relates to an intelligent metasurface security transmission optimization method based on a supervised deep neural network. The method comprises the following steps: acquiring a signal of a transmitting integrated RIS, modeling an RIS-to-legal user channel, combining the signal and the channel, and calculating a legal user received signal-to-interference-and-noise ratio and an achievable rate; meanwhile, corresponding indexes at an eavesdropper and an achievable rate of an eavesdropping link are calculated. Then, a multicast system security capacity is designed according to a minimum value of the achievable rate of the legal user and the achievable rate of the eavesdropping link. In order to achieve the target, an optimization model is constructed by combining the achievable rate of the legal user, RIS power and a perceived signal-to-noise ratio constraint. A supervised deep neural network and a training set are constructed, an offline training, online inference and parameter updating are carried out by using an Adam optimizer according to a preset loss function, and a trained network is obtained. Finally, the optimization model is solved by using the network, and RIS phase shift and power allocation coefficients are acquired. The method can reduce the calculation complexity and improve the real-time performance.

Description

Technical Field

[0001] This application relates to the field of wireless communication technology, and in particular to an intelligent metasurface secure transmission optimization method based on supervised deep neural networks. Background Technology

[0002] In recent years, Integrated Sensing and Communication (ISAC) technology has become increasingly important in Intelligent Transportation Systems (ITS), aiming to simultaneously achieve high-precision environmental perception and reliable communication. However, traditional optimization methods face challenges such as high computational complexity and poor real-time performance when dealing with resource allocation problems in ISAC systems, especially in dynamic on-board environments where rapid channel changes make traditional methods difficult to adapt. Reconfigurable Smart Surfaces (RIS), as an emerging technology, provide a new way to improve the performance of ISAC systems by controlling the propagation characteristics of electromagnetic signals. However, the high-dimensional optimization variables and complex constraints introduced by RIS further exacerbate the difficulty of real-time optimization. The development of deep learning (DL) technology offers a new approach to solving this problem; its powerful nonlinear mapping capabilities and real-time inference characteristics promise to break through the bottlenecks of traditional optimization methods.

[0003] In existing research, the optimization of RIS-based ISAC systems largely relies on mathematical optimization methods, which are computationally burdensome in large-scale scenarios and fail to meet the ultra-low latency requirements of ITS. Although methods such as reinforcement learning (RL) have been explored for dynamic optimization, their slow convergence speed and sensitivity to initial conditions limit their application in real-time demanding automotive scenarios. Summary of the Invention

[0004] Therefore, it is necessary to provide an intelligent metasurface secure transmission optimization method based on supervised deep neural networks that can reduce computational complexity and improve real-time performance to address the aforementioned technical problems.

[0005] A method for optimizing secure transmission on intelligent metasurfaces based on supervised deep neural networks is proposed. This method is applied to an ISAC system assisted by a RIS (Intelligent Transmission Integrated System) including a transmission integrated RIS (Reflection System) with configured reflection units, a legitimate user, a sensing target, and an eavesdropper. The method includes:

[0006] Acquire the transmitted signal of the integrated RIS and model the channel from the RIS to the legitimate user; calculate the signal-to-interference-plus-noise ratio (SIR) received at the legitimate user and the reachable rate of the legitimate user based on the transmitted signal of the integrated RIS and the channel from the RIS to the legitimate user; calculate the SIR received at the legitimate user end and the reachable rate of the eavesdropping link based on the communication signal received at the eavesdropper.

[0007] The security capacity of the multicast system is designed based on the minimum reachable rate of legitimate users and the reachable rate of the eavesdropping link. With security capacity as the objective function, a RIS-assisted ISAC system optimization model is constructed using pre-set reachable rate constraints of legitimate users, RIS amplification power constraints, and perceived signal-to-noise ratio constraints.

[0008] Construct a supervised deep neural network and a training set. Perform offline training and online inference on the supervised deep neural network based on the training set and a pre-set loss function. Use the Adam optimizer to iteratively update the network parameters to obtain a trained supervised deep neural network.

[0009] The RIS-assisted ISAC system optimization model is solved using a trained supervised deep neural network to obtain the RIS phase shift and power allocation coefficients.

[0010] The aforementioned intelligent metasurface secure transmission optimization method based on supervised deep neural networks significantly reduces computational complexity and improves real-time response capabilities by transforming the high-dimensional and complex resource optimization problem into an end-to-end deep learning mapping. Utilizing the nonlinear mapping capability of DNNs avoids the iterative solution process of traditional optimization methods, enabling resource allocation decisions to be completed within milliseconds, making it suitable for rapid channel changes in vehicular environments. Through codebook-driven supervised training, the model automatically learns the optimal solution that satisfies CRB and rate constraints, reducing manual parameter tuning overhead. The physical constraint processing mechanism integrated into the network architecture ensures that the output RIS phase shift and power allocation coefficients strictly conform to the requirements of the actual system. In dynamic vehicular scenarios, the offline-trained DNN can adapt to different channel environments by updating the dataset, and responds to real-time CSI changes without additional computation during the online inference phase. This achieves coordinated optimization of security efficiency, sensing accuracy, and communication rate, providing an efficient and feasible technical solution for real-time ISAC applications in ITS. Attached Figure Description

[0011] Figure 1 This is a schematic diagram of a RIS-assisted ISAC system in one embodiment;

[0012] Figure 2 This is a flowchart illustrating an intelligent metasurface secure transport optimization method based on a supervised deep neural network in one embodiment.

[0013] Figure 3 This is a schematic diagram of a deep neural network architecture in one embodiment. Detailed Implementation

[0014] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0015] The intelligent metasurface secure transport optimization method based on supervised deep neural networks provided in this application can be applied to, for example... Figure 1 The RIS-assisted ISAC system shown includes a transmit integrated RIS (TX-RIS) with M reflector elements, K legitimate single-antenna users, L sensing targets, and a single-antenna eavesdropper. The TX-RIS maximizes the security capacity while meeting the sensing signal-to-noise ratio and rate constraints by adjusting its phase shift matrix and power allocation coefficient.

[0016] In one embodiment, such as Figure 2 As shown, a method for optimizing secure transport on intelligent metasurfaces based on supervised deep neural networks is provided, including the following steps:

[0017] Step 202: Obtain the transmission signal of the integrated RIS and model the channel from the RIS to the legitimate user; based on the transmission signal of the integrated RIS and the channel from the RIS to the legitimate user, calculate the signal-to-interference-plus-noise ratio (SIR) received at the legitimate user and the reachable rate of the legitimate user; based on the communication signal received at the eavesdropper, calculate the SIR received at the legitimate user end and the reachable rate of the eavesdropping link.

[0018] Traditional methods require real-time matrix inversion and iterative updates of the channel from the RIS to legitimate users and eavesdroppers, with the computational load increasing exponentially with the number of legitimate users. This new process, however, standardizes signal parameters and maps them to channel characteristics, transforming the calculation of metrics such as signal-to-interference-plus-noise ratio (SINR) and reachable rate into fixed formulas, thus avoiding redundancy in matrix operations during dynamic iteration. The specific process includes:

[0019] The RIS-assisted ISAC system comprises a TX-RIS with M reflector elements, employing a direct-fed architecture that integrates a digital phased array and an RF power distribution layer, enabling simultaneous transmission of communication and sensing signals. Assuming the strongly fed RIS has M arrays, the phase shift vector of the RIS can be expressed as:

[0020] ;

[0021] in, This represents the phase shift coefficient of the m-th element in the RIS array.

[0022] The transmit signal of TX-RIS is represented as follows:

[0023] ;

[0024] in, , These represent the communication signals transmitted by the RIS to legitimate users and the sensing signals transmitted to the target, respectively. , These represent the transmit power allocated to legitimate users and targets by the RF-fed antenna, respectively. , These represent the phase shift vectors used by the RIS for communication and sensing, respectively, and both satisfy Equation 1. The TX-RIS optimizes its phase shift vector and power allocation coefficients based on the channel between the RIS and legitimate communication users, eavesdroppers, and the sensing signal-to-noise ratio constraints.

[0025] In real-world scenarios, legitimate users are typically located beyond the Rayleigh distance of the transmitter. Since both LoS and NLoS components exist in far-field plane wave propagation, the channel from RIS to legitimate user k can be modeled as a Rician fading channel, as expressed below:

[0026] ;

[0027] in, This represents the path loss at a reference distance of 1 meter. This represents the corresponding path loss index. This represents the distance between RIS and the legitimate user k. The Rician factor represents the RIS-legal user link. It is a deterministic Loss component. It is modeled as the NLoS component of Rayleigh fading. The array response vector (ARV) can be represented as:

[0028] ;

[0029] in, and These represent the pitch angle and azimuth angle, respectively. For the spacing between array elements, For carrier wavelength, Represents the Kronecker product. Eavesdropper channel. and target channel Expression derivation .

[0030] The signal received by legitimate user k can be represented as:

[0031] ;

[0032] in, The noise power at point k for legitimate user is... Additive white Gaussian noise (AWGN).

[0033] In the The signal-to-interference-plus-noise ratio (SIR) received at a legitimate multicast client can be expressed as:

[0034] ;

[0035] The reachable rate for the kth legitimate user can be expressed as:

[0036] .

[0037] In multicast communication scenarios, the minimum communication rate is defined as the set of legitimate users. The achievable rate for the legitimate user with the worst channel conditions. Since multicast transmission requires all legitimate users to correctly receive the same public information, system performance is limited by the legitimate user with the worst link quality. Therefore, the minimum communication rate directly reflects the minimum communication performance of the multicast system and is a key indicator for ensuring fairness and reliability in communication among multiple legitimate users. The minimum communication rate is... The minimum reachable rate for a number of legitimate users can be expressed as:

[0038] ;

[0039] The communication signal received by the eavesdropper can be represented as:

[0040] ;

[0041] in, Indicates noise power as AWGN. satisfy ,in Represents the Boltzmann constant. Indicates the noise temperature. Indicates the noise bandwidth.

[0042] The signal-to-interference-plus-noise ratio (SIR) received by a legitimate user terminal during eavesdropping can be expressed as:

[0043] ;

[0044] The achievable rate of an eavesdropping link can be expressed as:

[0045] ;

[0046] Similarly, the target echo received by RIS is represented as:

[0047] ;

[0048] in, The AWGN at RIS is represented by the following distribution: , This represents noise power.

[0049] The echo signal-to-noise ratio (SNR) is expressed as:

[0050] .

[0051] Step 204: Design the security capacity of the multicast system based on the minimum reachable rate of legitimate users and the reachable rate of the eavesdropping link. Using the security capacity as the objective function, construct a RIS-assisted ISAC system optimization model using pre-set reachable rate constraints of legitimate users, RIS amplification power constraints, and perceived signal-to-noise ratio constraints.

[0052] In the proposed integrated reconnaissance and communication system, to achieve a balance between system performance and power consumption, the security capacity is set as the optimization objective. Under the premise of satisfying the reachability requirements of legitimate multicast users, the RIS amplification power constraint, and the sensing signal-to-noise ratio constraint, the system's security capacity is maximized by jointly optimizing the RIS phase shift vector and the power allocation coefficient of the feed antenna. Therefore, the optimization problem can be formulated as:

[0053] ;

[0054] ;

[0055] ;

[0056] ;

[0057] ;

[0058] Among them, constraint (C1) limits the power of RIS transmission, and constraint (C2) guarantees that the minimum reachable rate of legitimate multicast user k must not be lower than the rate threshold. Constraint (C3) represents the phase shift angle constraint and unit mode constraint for each element of the RIS array, and constraint (C4) represents the sensing signal-to-noise ratio constraint.

[0059] Step 206: Construct a supervised deep neural network and a training set. Perform offline training and online inference on the supervised deep neural network based on the training set and a pre-set loss function. Use the Adam optimizer to iteratively update the network parameters to obtain a trained supervised deep neural network.

[0060] The introduction of supervised deep neural networks is key to the efficiency leap: In the offline training phase, by covering massive channel scenarios with the training set, and leveraging the adaptive learning rate characteristics of the Adam optimizer, the optimization logic of high-dimensional RIS phase shift and power allocation is solidified into a nonlinear mapping relationship of the network, replacing the convex optimization solution that needs to be performed for each round of channel change in traditional methods. During online inference, after inputting real-time channel state information, the network directly outputs the optimization result, eliminating the complete computation chain from modeling to solving in traditional methods, compressing the response time to the millisecond level, perfectly adapting to the characteristics of rapidly changing channels in the vehicle environment. The physical constraint processing mechanism is embedded in the network layer structure, satisfying constraints such as RIS power and perceived signal-to-noise ratio when the output result is generated, eliminating the need for constraint verification and parameter readjustment after solving, as in traditional methods, further reducing redundant computation.

[0061] Step 208: Solve the RIS-assisted ISAC system optimization model using the trained supervised deep neural network to obtain the RIS phase shift and power allocation coefficients.

[0062] The aforementioned intelligent metasurface secure transmission optimization method based on supervised deep neural networks significantly reduces computational complexity and improves real-time response capabilities by transforming the high-dimensional and complex resource optimization problem into an end-to-end deep learning mapping. Utilizing the nonlinear mapping capability of DNNs avoids the iterative solution process of traditional optimization methods, enabling resource allocation decisions to be completed within milliseconds, making it suitable for rapid channel changes in vehicular environments. Through codebook-driven supervised training, the model automatically learns the optimal solution that satisfies CRB and rate constraints, reducing manual parameter tuning overhead. The physical constraint processing mechanism integrated into the network architecture ensures that the output RIS phase shift and power allocation coefficients strictly conform to the requirements of the actual system. In dynamic vehicular scenarios, the offline-trained DNN can adapt to different channel environments by updating the dataset, and responds to real-time CSI changes without additional computation during the online inference phase. This achieves coordinated optimization of security efficiency, sensing accuracy, and communication rate, providing an efficient and feasible technical solution for real-time ISAC applications in ITS.

[0063] In one embodiment, the channel from the RIS to the legitimate user is modeled, including:

[0064] The channel from RIS to legitimate user k is modeled as follows:

[0065] ;

[0066] in, This represents the path loss at a reference distance of 1 meter. This represents the corresponding path loss index. Indicates RIS and legitimate usersk The distance between them The Rician factor represents the RIS-legal user link. It is a deterministic Loss component. It is modeled as the NLoS component of Rayleigh fading. It is the array response vector for the k-th user. This represents the total number of legitimate users. This represents the pitch angle of the k-th user. This represents the azimuth angle of the k-th user. This represents an M×M dimensional identity matrix.

[0067] In one embodiment, calculating the signal-to-interference-plus-noise ratio (SINNR) received at the legitimate user and the reachable rate of the legitimate user based on the transmitted signal of the transmission integrated RIS and the channel from the RIS to the legitimate user includes:

[0068] Based on the transmitted signal of the integrated RIS and the channel calculation from the RIS to the legitimate user, the signal-to-interference-plus-noise ratio (SIR) received at the legitimate user and the reachable rate of the legitimate user are as follows:

[0069] ;

[0070] in, Let S be the signal-to-interference-plus-noise ratio (SIR) received at the k-th legitimate multicast client. The superscript H represents the conjugate transpose of the channel from RIS to legitimate user k. , These represent the transmit power allocated to legitimate users and targets by the RF-fed antenna, respectively. , These represent the phase shift vectors used by RIS for communication and sensing, respectively. The noise power at point k is for legitimate users.

[0071] The reachable rate of the kth legal user is then calculated as follows:

[0072] .

[0073] In one embodiment, the communication signal received by the eavesdropper is:

[0074] ;

[0075] in, , These represent the communication signals transmitted by the RIS to legitimate users and the sensing signals transmitted to the target, respectively. , These represent the transmit power allocated to legitimate users and targets by the RF-fed antenna, respectively. This represents the phase shift vector used by RIS for communication. Indicates noise power as AWGN, This represents the eavesdropper's channel, and the superscript H indicates the conjugate transpose.

[0076] In one embodiment, calculating the signal-to-interference-plus-noise ratio (SIR) received by the legitimate user terminal and the achievable speed of the eavesdropping link based on the communication signals received at the eavesdropper includes:

[0077] Based on the communication signals received at the eavesdropper's location, the signal-to-interference-plus-noise ratio (SIR) received by the legitimate user terminal and the achievable speed of the eavesdropping link are calculated as follows:

[0078] ;

[0079] ;

[0080] in, , These represent the transmit power allocated to legitimate users and targets by the RF-fed antenna, respectively. , These represent the phase shift vectors used by RIS for communication and sensing, respectively. eavesdropper Noise power at the location, This indicates the channel for the eavesdropper. This indicates the signal-to-interference-plus-noise ratio (SIR) received by a legitimate user terminal during eavesdropping.

[0081] In one embodiment, using security capacity as the objective function, a RIS-assisted ISAC system optimization model is constructed using pre-set reachability constraints for legitimate users, RIS amplification power constraints, and perceived signal-to-noise ratio constraints, including:

[0082] Using safe capacity as the objective function, a RIS-assisted ISAC system optimization model is constructed using pre-set reachability constraints for legitimate users, RIS amplification power constraints, and sensing signal-to-noise ratio constraints:

[0083] ;

[0084] ;

[0085] ;

[0086] ;

[0087] ;

[0088] Where C1 represents the power limit for RIS transmission, and C2 represents the minimum reachable rate that guarantees legitimate multicast user k must not be lower than the rate threshold. C3 represents the phase shift angle constraint and unit mode constraint for each element of the RIS array, and C4 represents the sensing signal-to-noise ratio constraint. , These represent the transmit power allocated to legitimate users and targets by the RF-fed antenna, respectively. , These represent the phase shift vectors used by RIS for communication and sensing, respectively. Indicates unit modulus phase shift, This represents the minimum reachable rate for legitimate users. Indicates the rate of eavesdropping. This indicates the actual power emitted by the RIS. This represents the reachable rate of the kth legitimate user. The minimum threshold representing the echo signal-to-noise ratio of the target. Indicates the number of perceived targets. Indicates the number of RIS array elements. This indicates the maximum threshold for transmission power. This represents the echo signal-to-noise ratio of the I-th reconnaissance target.

[0089] In one embodiment, constructing a supervised deep neural network includes:

[0090] like Figure 3 As shown, a supervised deep neural network is designed using a 6-layer fully connected hidden layer architecture, with the number of neurons configured in the order of 512→512→256→128→256→512. Each layer integrates a ReLU activation function and L2 regularization to suppress overfitting. The input layer receives preprocessed channel features, including legitimate user channel data. Eavesdropper Channel With the target channel The real and imaginary parts of are composed of dimensions (K+L+1). The input vector for 2M is:

[0091] ;

[0092] in, Indicates the real part, Indicates the imaginary part. and The discrete power allocation parameters are represented by K, where K represents the total number of users and L represents the number of sensing targets. The output layer is divided into a phase shift prediction branch and a power allocation branch. The phase shift prediction branch generates a 2M-dimensional real-valued vector, with the first M dimensions corresponding to the communication phase shift. The subsequent M-dimensional corresponding sensing phase shift The output is mapped to [-1, 1] using the tanh activation function, and then converted to a unit modulus phase shift using Euler's formula:

[0093] ;

[0094] The power allocation branch outputs a 2D vector, which, through a sigmoid activation function and scaling operation, generates the power allocation coefficients as follows:

[0095] ;

[0096] in, Generated by the softmax function. Ensure the power factor is within the range of (0, 1). Represents the power allocation factor. This represents the actual distributable power.

[0097] In one embodiment, the process of constructing the training set includes:

[0098] exist N phase-shift codebooks are generated by uniform sampling within the range. and Each codebook element is an M-dimensional phase shift vector; for each codebook pair Traverse discrete power allocation parameters Calculate the corresponding safety capacity; select configurations that meet the sensing signal-to-noise ratio and rate constraints, and record the CSI and resource configuration corresponding to the maximum safety capacity as training samples. , among which the tags .

[0099] In one embodiment, the pre-set loss function is:

[0100] ;

[0101] in, This is the network prediction value. For real labels, The regularization coefficient is . Let n be the weight matrix of the nth layer. This indicates the mini-batch size during the training phase. Indicates the number of network layers.

[0102] In one embodiment, offline training includes collecting CSI data in an in-vehicle environment and preprocessing it into a network input format; generating labels by searching the codebook to construct a training dataset; initializing DNN parameters, iteratively training through the Adam optimizer, and updating network weights; periodically evaluating validation set performance and saving the model with the best generalization ability; the online inference stage includes real-time acquisition of CSI data from TX-RIS to legitimate users and eavesdroppers; preprocessing CSI data into network input, generating phase shift and power allocation prediction post-processing output results through forward propagation of the trained DNN, generating resource configurations that satisfy unity modulus constraints and power budgets; verifying whether the resource configurations satisfying unity modulus constraints and power budgets meet the perceptual signal-to-noise ratio and rate requirements, and initiating a grid search fallback mechanism if they do not.

[0103] In a specific embodiment, the Adam optimizer is used to iteratively update the network parameters, and the update formula is as follows:

[0104] ;

[0105] ;

[0106] Among them, momentum coefficient Learning rate , To prevent division-by-zero error, monitor and validate the loss during training. If there is no improvement after 30 consecutive rounds, stop early to avoid overfitting.

[0107] It should be understood that, although Figure 2 The steps in the flowchart are shown sequentially as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order in which these steps are executed, and they can be performed in other orders. Furthermore, Figure 2 At least some of the steps in the process may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least some of the sub-steps or stages of other steps.

[0108] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0109] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for optimizing secure transport on intelligent metasurfaces based on supervised deep neural networks, characterized in that, The method is applied to an ISAC system including a transmit integrated RIS with a configured reflection unit, a legitimate user, a sensing target, and an eavesdropper; the method includes: The system acquires the transmission signal of the integrated transmission RIS and models the channel from the RIS to the legitimate user; based on the transmission signal of the integrated transmission RIS and the channel from the RIS to the legitimate user, it calculates the signal-to-interference-plus-noise ratio (SIR) received at the legitimate user and the reachable rate of the legitimate user; based on the communication signal received at the eavesdropper, it calculates the SIR received at the legitimate user end and the reachable rate of the eavesdropping link. The security capacity of the multicast system is designed based on the minimum reachable rate of the legitimate users and the reachable rate of the eavesdropping link. The security capacity is used as the objective function, and a RIS-assisted ISAC system optimization model is constructed using pre-set reachable rate constraints of legitimate users, RIS amplification power constraints, and perceived signal-to-noise ratio constraints. A supervised deep neural network and a training set are constructed. The supervised deep neural network is trained offline and inferred online based on the training set and a pre-set loss function. The network parameters are iteratively updated using the Adam optimizer to obtain a trained supervised deep neural network. The RIS-assisted ISAC system optimization model is solved using the trained supervised deep neural network to obtain the RIS phase shift and power allocation coefficients. Building supervised deep neural networks includes: A supervised deep neural network is designed using a 6-layer fully connected hidden layer architecture, with the number of neurons configured in the order of 512→512→256→128→256→512. Each layer integrates a ReLU activation function and L2 regularization to suppress overfitting. The input layer receives preprocessed channel features, including legitimate user channel data. Eavesdropper Channel With the target channel The real and imaginary parts of are composed of dimensions (K+L+1). The input vector is 2M, where K represents the total number of users and L represents the number of perceived targets; The output layer is divided into a phase shift prediction branch and a power allocation branch. The phase shift prediction branch generates a 2M-dimensional real-valued vector, with the first M dimensions corresponding to the communication phase shift. The subsequent M-dimensional corresponding sensing phase shift The output is mapped to [-1, 1] using the tanh activation function, and then converted to a unit-mode phase shift using Euler's formula. The power allocation branch outputs a 2D vector, which generates power allocation coefficients using the sigmoid activation function and scaling operation. in, Generated by the softmax function. Ensure the power factor is within the range of (0, 1). Represents the power allocation factor. This represents the actual distributable power; The process of constructing a training set includes: exist N phase-shift codebooks are generated by uniform sampling within the range. and Each codebook element is an M-dimensional phase shift vector; for each codebook pair Traverse discrete power allocation parameters Calculate the corresponding safety capacity; select configurations that meet the sensing signal-to-noise ratio and rate constraints, and record the CSI and resource configuration corresponding to the maximum safety capacity as training samples. , among which the tags .

2. The method according to claim 1, characterized in that, Modeling the channel from RIS to legitimate users includes: Model the channel from RIS to legitimate user k as follows in, This represents the path loss at a reference distance of 1 meter. This represents the corresponding path loss index. Indicates RIS and legitimate users k The distance between them The Rician factor represents the RIS-legal user link. It is a deterministic Loss component. It is modeled as the NLoS component of Rayleigh fading. It is the array response vector for the k-th user. This represents the total number of legitimate users. This represents the pitch angle of the k-th user. This represents the azimuth angle of the k-th user. This represents an M×M dimensional identity matrix.

3. The method according to claim 1, characterized in that, Based on the transmitted signal of the integrated transmission RIS and the channel from the RIS to the legitimate user, the signal-to-interference-plus-noise ratio (SIR) received at the legitimate user and the reachable rate of the legitimate user are calculated, including: Based on the transmitted signal of the integrated RIS and the channel from the RIS to the legitimate user, the signal-to-interference-plus-noise ratio (SIR) received at the legitimate user and the achievable rate of the legitimate user are calculated as follows: in, Let S be the signal-to-interference-plus-noise ratio (SIR) received at the k-th legitimate multicast client. The superscript H represents the conjugate transpose of the channel from RIS to legitimate user k. , These represent the transmit power allocated to legitimate users and targets by the RF-fed antenna, respectively. , These represent the phase shift vectors used by RIS for communication and sensing, respectively. For legitimate users Noise power at the location; The reachable rate of the kth legal user is then calculated as follows: 。 4. The method according to claim 1, characterized in that, The communication signal received by the eavesdropper is in, , These represent the communication signals transmitted by the RIS to legitimate users and the sensing signals transmitted to the target, respectively. , These represent the transmit power allocated to legitimate users and targets by the RF-fed antenna, respectively. This represents the phase shift vector used by RIS for communication. Indicates noise power as AWGN, This represents the eavesdropper's channel, and the superscript H indicates the conjugate transpose.

5. The method according to claim 4, characterized in that, Calculate the signal-to-interference-plus-noise ratio (SIR) received by the legitimate user terminal and the achievable speed of the eavesdropping link based on the communication signals received at the eavesdropper's location, including: Based on the communication signals received at the eavesdropper's location, the signal-to-interference-plus-noise ratio (SIR) received by the legitimate user terminal and the achievable speed of the eavesdropping link are calculated as follows: in, , These represent the transmit power allocated to legitimate users and targets by the RF-fed antenna, respectively. , These represent the phase shift vectors used by RIS for communication and sensing, respectively. eavesdropper Noise power at the location, This indicates the channel for the eavesdropper. This indicates the signal-to-interference-plus-noise ratio (SIR) received by a legitimate user terminal during eavesdropping.

6. The method according to claim 1, characterized in that, Using the security capacity as the objective function, a RIS-assisted ISAC system optimization model is constructed using pre-set reachability constraints for legitimate users, RIS amplification power constraints, and sensing signal-to-noise ratio constraints, including: Using the security capacity as the objective function, a RIS-assisted ISAC system optimization model is constructed using pre-set reachability constraints for legitimate users, RIS amplification power constraints, and sensing signal-to-noise ratio constraints. Where C1 represents the power limit for RIS transmission, and C2 represents the minimum reachable rate that guarantees legitimate multicast user k must not be lower than the rate threshold. C3 represents the phase shift angle constraint and unit mode constraint for each element of the RIS array, and C4 represents the sensing signal-to-noise ratio constraint. , These represent the transmit power allocated to legitimate users and targets by the RF-fed antenna, respectively. , These represent the phase shift vectors used by RIS for communication and sensing, respectively. Indicates unit modulus phase shift, This represents the minimum reachable rate for legitimate users. Indicates the rate of eavesdropping. This indicates the actual power emitted by the RIS. This represents the reachable rate of the kth legitimate user. The minimum threshold representing the echo signal-to-noise ratio of the target. Indicates the number of perceived targets. Indicates the number of RIS array elements. This indicates the maximum threshold for transmission power. Indicates the first l The signal-to-noise ratio of the echo of a reconnaissance target.

7. The method according to claim 1, characterized in that, The pre-set loss function is in, This is the network prediction value. For real labels, The regularization coefficient is . Let n be the weight matrix of the nth layer. This indicates the mini-batch size during the training phase. Indicates the number of network layers.

8. The method according to claim 1, characterized in that, The offline training includes collecting CSI data in a vehicle environment, preprocessing it into a network input format, generating labels by searching the codebook, and constructing a training dataset. Initialize the DNN parameters, iteratively train using the Adam optimizer, and update the network weights; Regularly evaluate the performance on the validation set and save the model with the best generalization ability; The online inference phase includes real-time acquisition of TX-RIS data from legitimate users to the CSI of eavesdroppers; The preprocessed CSI is used as the network input. The phase shift and power allocation prediction postprocessing output is generated through forward propagation of the trained DNN to generate a resource configuration that satisfies the unity modulus constraint and power budget. The resource configuration that satisfies the unity modulus constraint and power budget is verified to meet the sensing signal-to-noise ratio and rate requirements. If it does not meet the requirements, the grid search fallback mechanism is activated.

Images