Deep learning-based intelligent reflector-assisted terahertz secure communication method and system

Abstract

The invention provides an intelligent reflector-assisted terahertz secure communication method and system based on deep learning, and belongs to the technical field of wireless communication, and the method comprises the steps: constructing a reflection beam; the intelligent reflecting surface directly reflects received information which is sent by an information source and carries privacy signals to a legal receiving end, and an eavesdropping user eavesdrops the information at a link from a base station to the intelligent reflecting surface; calculating the lower bound of the secrecy rate in an imperfect eavesdropping channel state; the method comprises the following steps: constructing an optimization objective function for maximizing the secrecy rate, taking the secrecy rate objective function as a loss function of unsupervised learning on the basis of training a large amount of channel data by a deep learning network, predicting an optimal intelligent reflecting surface shift matrix, solving an optimal transmitting beam forming matrix, and obtaining the optimal transmitting beam forming matrix. Performing information secrecy transmission according to the optimal secrecy rate obtained under the optimal solution; according to the method, the influence of the imperfect acquisition of the eavesdropping channel state information on the information security transmission is considered, and the purpose of improving the security communication of the terahertz system through the deep learning neural network is achieved.

Description

Technical Field

[0001] This invention belongs to the field of wireless communication technology and relates to a deep learning-based intelligent reflector-assisted terahertz secure communication method and system, which further improves the overall security performance of wireless communication systems. Background Technology

[0002] The continuous development of wireless communication technology has led to higher data transmission rates, lower latency, and wider coverage, making wireless communication security paramount. The openness and complexity of wireless communication also present significant security challenges, with issues such as data eavesdropping, signal interference, and spoofing attacks emerging frequently. Therefore, physical layer security, as an important supplement to information security, can significantly improve the security of information transmission. Compared to traditional encryption technologies that rely on large amounts of communication resources and complex infrastructure, physical layer security offers a more efficient and reliable solution. However, to cover large geographical areas and cope with complex environments, a large number of high-cost, high-energy-efficiency communication security devices must be deployed and maintained, inevitably leading to substantial capital and operational expenditures. Furthermore, the complexity of communication environments limits the confidentiality of traditional physical layer security methods. Therefore, overcoming these new security challenges remains a significant challenge.

[0003] Compared to traditional microwave communication, terahertz communication's narrow beamwidth offers a natural advantage in terms of concealment and security, making it difficult for eavesdroppers to capture valid signals. Existing research on terahertz secure communication systems largely focuses on channel estimation or reducing channel overhead, with limited research on terahertz security efficiency. The narrow beamwidth of terahertz communication requires precise beam alignment. Furthermore, terahertz devices require high power to operate normally, and the significant propagation loss and water molecule absorption of terahertz waves severely limit transmission distance. In complex urban environments, terahertz waves are directly blocked by buildings and moving objects. Introducing intelligent reflectors to construct virtual direct links can lead to both links being vulnerable to eavesdropping, resulting in a decrease in security efficiency. Therefore, a model that considers both improving system security and focusing on solving optimization problems efficiently is urgently needed.

[0004] Deep learning can automatically extract complex nonlinear features, which is particularly crucial for Channel State Information (CSI) analysis, eavesdropping detection, and signal recognition in communication systems. Through deep learning, terahertz communication systems can capture potential correlations and hidden patterns in multidimensional data, significantly improving algorithm efficiency. Secondly, deep learning algorithms excel at processing large-scale data, making them suitable for high-density, multi-user scenarios in terahertz communication systems. Compared to traditional optimization algorithms, deep learning not only significantly improves the security performance of terahertz communication systems but also reduces computational complexity and effectively lowers power consumption requirements, providing a feasible solution for terahertz multivariate optimization and low-power devices.

[0005] However, existing research on security gains largely relies on perfectly intercepted user channel state information, which is a highly idealistic assumption. In real communication systems, to achieve effective eavesdropping without being detected by legitimate users, eavesdropping users often remain silent or conceal their presence, and they do not actively or accurately report the intercepted channel state information to legitimate users. Therefore, it is difficult for legitimate communication users to obtain perfectly intercepted channel state information. This uncertainty leads to serious information leakage and a significant reduction in system security performance. Summary of the Invention

[0006] To address the shortcomings of existing technologies and solve the aforementioned problems, this invention proposes a deep learning-based intelligent reflector-assisted terahertz secure communication method and system. Specifically, for information security transmission systems where the eavesdropping channel state information is not perfectly known, this invention theoretically derives a closed-form expression for the confidentiality rate and a lower bound for the confidentiality rate under imperfect eavesdropping channel state information. It then proposes a robust secure transmission method based on deep learning to design the transmit beamforming matrix and the intelligent reflector phase shift matrix, maximizing the system's confidentiality rate and achieving secure information transmission.

[0007] The technical solution of the present invention is as follows: A deep learning-based intelligent reflector-assisted terahertz secure communication method includes the following steps: (1) Deploy smart reflective surfaces on the surface of buildings, and the smart reflective surfaces are equipped with multiple reflective elements; (2) The intelligent reflector constructs a reflected beam by adjusting the phase shift matrix of the reflective element; (3) By using the transmit beam and the reflect beam, the signal can be sent from the base station and reflected to the legitimate receiver through the smart reflector, and the user can be eavesdropped on receiving the information. (4) Calculate the lower bound of the confidentiality rate under the imperfect eavesdropping channel state based on the information received by the legitimate receiver and the eavesdropping user; (5) Construct an optimization objective function to maximize the security rate based on the transmit beamforming matrix and the intelligent reflector phase shift matrix. The objective function is equivalent to two sub-problems. Based on the corresponding sub-problems, design the data preprocessing method, network structure and loss function of the neural network to calculate the optimal solution. Based on the obtained optimal solution, calculate the optimal security rate and carry out information security transmission.

[0008] Preferably, in step (1), the base station sends the confidential signal to the smart reflector, establishes a terahertz channel model, and considers path fading, molecular absorption, channel gain, and arrival and departure azimuth angles: (1) (2) (3) (4) Among them, H BI H is the channel matrix from the BS to the smart reflector. IU H is the channel matrix from the smart reflector to Bob. IE H is the channel matrix from the smart reflector to Eve. BE Let L be the channel matrix from BS to Eve, where L represents the number of paths from BS to IRS, N is the number of reflecting elements, and M is the number of base station antennas. l Indicates the corresponding path, when l When =0, the corresponding path is represented as LoS Path; in addition, Indicates the communication link between BS and IRS. l The channel gain of a path includes transmission loss, molecular absorption, and path loss caused by intercepting eavesdropping devices. The first communication link between the smart reflector IRS and Bob l Complex gain of the path, For the communication link between IRS and Eve l Complex gain of the path, The first communication link between BS and Eve l Channel gain of each path, G r and G t This is the antenna gain of the transceiver, where the subscript r represents the receiver antenna and the subscript t represents the transmitter antenna. and This represents the array steering vector at the transceiver; , It is the azimuth angle at which the signal arrives. , Given the azimuth angle at which the signal departs, we will randomly generate signal reflection points at different locations to reflect the signal, while the azimuth angles at which the signal arrives and departs satisfy a certain distribution.

[0009] Terahertz signals suffer from severe molecular absorption, propagation loss, and high penetration loss, therefore we assume LoS The path loss formula is: (5) Where r is the distance between the transmitter and the receiver, and j is a complex unit. f For operating frequency, K ( f ) is the molecular absorption coefficient, and c is the speed of light; NLoS Path loss is the Fresnel reflection coefficient R ( f The combination of propagation loss and molecular absorption loss gives the path loss formula as follows: (6) in r 1 represents the distance from the transmitter to the reflecting surface. r 2 represents the distance from the reflector to the receiver; f For operating frequency, K ( f The molecular absorption coefficient () can be written as: (7) Where p and p0 are the system pressure and standard pressure, respectively, and T and T0 are the system temperature and standard temperature, respectively; The absorption cross section is represented by g, which represents the type of gas molecules. For N in the xz plane x ×N z The first USPA array antenna element, the... l The array steering vector corresponding to each path can be represented as: (8) in For wavelength, f c For carrier frequency, The unit spacing is 0 ≤ p ≤ N. x , 0≤q≤N z For the element index of the xz plane, The azimuth angle at which the signal arrives or departs. The elevation angle at which the signal arrives or departs.

[0010] Preferably, in step (4), the calculation of the lower bound of the confidentiality rate under the imperfect eavesdropping channel state information based on the information received by the legitimate receiver and the eavesdropping user specifically includes: (4-1) Determine the information received by the legitimate receiving end and the information received by the eavesdropping user; (4-2) Determine the signal-to-noise ratio of the legitimate receiver and the signal-to-noise ratio of the eavesdropping user based on the information received by the legitimate receiver and the information received by the eavesdropping user; (4-3) Considering the imperfect eavesdropping channel state information, calculate the lower bound of the confidentiality rate under the condition that the eavesdropping channel state information is imperfectly known.

[0011] Further, in step (4-1), the information received by the legitimate receiving end is: (9) Because the addition of a smart reflector allows eavesdropping users to eavesdrop on two different links, when an eavesdropping user eavesdrops at the link between the base station and the smart reflector, the information they receive is as follows: (10) in, , representing the average power of the signal, the power limit of the transmit beam matrix, and P, respectively. s The transmit power at the base station. The channel matrix from the BS to the smart reflector. Let C be the channel matrix from the smart reflector to Bob and the channel matrix from BS to Eve, respectively. C represents the complex domain, N is the number of reflector elements, and M is the number of base station antennas. r N represents the number of antennas for the user. e The number of antennas used by the eavesdropper. It is the reflection phase shift matrix at the intelligent reflector, where and It is the phase shift of the reflecting unit of the intelligent reflecting surface, where j is the complex phase part, the imaginary unit; For phase The reflection coefficient is denoted by ; w is the beamforming vector, and s is the transmitted signal. The mean is 0 and the variance is 0. Additive white Gaussian noise, The mean is 0 and the variance is 0. Additive white Gaussian noise.

[0012] Furthermore, in step (4-2), the signal-to-noise ratio of the legitimate receiver is determined based on the information received by the legitimate receiver and the information received by the eavesdropping user. Signal-to-noise ratio of the user being eavesdropped Specifically: (11) (12) in, This is the channel matrix that includes all uncertainties in the ECSI.

[0013] Further, in step (4-3), the step of considering the imperfect eavesdropping channel state information and calculating the lower bound of the secrecy rate under the imperfect eavesdropping channel state specifically involves: (4-3-1) Constructing channel state information for an imperfect eavesdropper; (4-3-2) Based on the triangle inequality, Cauchy inequality and the maximum signal-to-noise ratio at the eavesdropping user, the lower bound of the confidentiality rate under the imperfect eavesdropping channel state is obtained.

[0014] Furthermore, in step (4-3-1), considering the feedback delay and unreliability of feedback information caused by the malicious behavior of the eavesdropper, the construction of imperfect eavesdropper channel state information specifically adopts an additive uncertainty channel model with error bounds to explain the incomplete eavesdropping channel state information, that is: (13) in It is a channel matrix that includes all uncertainties in ECSI. To estimate the error between the channel matrix and the actual matrix, a continuous set Ω contains all possible CSI uncertainties, and its norm is given by... As a boundary.

[0015] Further, in step (4-3-2), the lower bound of the confidentiality rate under the imperfect eavesdropping channel state is obtained based on the triangle inequality, Cauchy inequality, and the maximum signal-to-noise ratio at the eavesdropping user, specifically as follows: By using the triangle inequality and Cauchy's inequality, for any vector... as well as The following inequality always holds true. (14) Therefore, the inequality is satisfied. ,in, The coefficients encompass all uncertain channels; Therefore, the maximum signal-to-noise ratio for the eavesdropper at the base station to the smart reflector link is: (15) in, To estimate the error between the channel matrix and the actual matrix at the eavesdropping link; With partial eavesdropper channel state information, the lower bound of the confidentiality rate between the base station and the legitimate receiver is: (16) in , This represents the maximum signal-to-noise ratio that an eavesdropping user can receive under any channel implementation. The signal-to-noise ratio at the eavesdropper's location. These represent the signal-to-noise ratios of the eavesdropper at the base station to the smart reflector link and the smart reflector to the user link, respectively.

[0016] Preferably, in step (5), an optimization objective function that maximizes the security rate is constructed based on the transmit beamforming matrix and the smart reflector phase shift matrix. Specifically, when the eavesdropper is located on the link from the base station to the smart reflector, the objective function is as follows: (17) Among the constraints This means that when Eve is near BS, Eve can directly intercept confidential signals emitted from BS; to reduce information leakage at Eve's location, the forced zero principle is used to invalidate confidential signals leaked into the eavesdropping channel; constraints This indicates the total power limit imposed on the transmitter; constraint This represents the phase shift constraint of the intelligent reflective surface.

[0017] Furthermore, in step (5), the objective function formula (17) is also equivalent to two subproblems, and the optimal solutions are calculated sequentially, specifically as follows: The objective function is equivalent to two independent subproblems, namely: ; Based on two independent sub-problems, a deep learning-based approach is used to design the optimal transmit beam and intelligent reflector phase shift. The optimal approach is obtained through training and iterative processing using randomly acquired channel data. .

[0018] Furthermore, in step (5), the CNN network structure is set to include input, two convolutional layers, three fully connected layers, a lambda layer, and output; Selecting a legitimate channel H IU H BE As the input data, since they are two-dimensional matrices, in order to improve the feature extraction capability of the CNN network, it is necessary to process the input Data1={H IU H BE Preprocessing is performed, and definitions are defined. , where g n It is H IU The nth column, It is H BE The nth line, F n This is the cascaded channel corresponding to the nth element of the IRS; therefore, the original channel can be rewritten as: (18) Specifically, the original data X=[F1,F2,…,F… N ] can be regarded as N N intelligent reflective surface units r ×M matrix, here The phase shift coefficient corresponding to each smart reflective surface unit; Since the input data for neural networks requires real numbers, but the original data composed of CSI is complex, we extract the real part, imaginary part, and absolute value of each element to form a new three-dimensional matrix, that is, according to the order of real part, imaginary part, and absolute value of X. ; The preprocessed data X passes through two convolutional layers (CL1, CL2) and three fully connected layers (FCL1, FCL2, FCL3), and is finally output by a Lambda layer. CL1 has 256 2×2 filters, CL2 has 512 2×2 filters, FCL1 and FCL2 have 64N and 16N neurons respectively, and the number of neurons in FCL3 is the same as the number of IRS reflective elements. To prevent overfitting of the neural network, a Dropout layer is inserted into each fully connected layer, and a rectified linear unit is used as the activation function. FCL3 uses a sigmoid activation function, which can compress the output z value to the range (0,1), and then the optimal phase shift matrix of the IRS is obtained through the Lambda layer. ; For any given phase shift matrix The objective function can be transformed into: (19) in , For channel H BI H IU H BE Given the transpose of the matrix, and by applying the null space constraint, transforming the problem into a generalized eigenvector problem, the optimal w is: (20) Where P s Let B be the transmit power at the base station, B be the null space of the eavesdropping channel, and q be the matrix. The eigenvector corresponding to its largest eigenvalue; Then predict Based on this, set Data1 and Data2={ Substitute into formula (20) and then calculate to produce the optimal w. opt .

[0019] Input data2 to obtain the optimal solution, and substitute the updated parameters into formula (16) to solve for the updated confidentiality rate; This method employs unsupervised learning and requires no additional annotations. Furthermore, the design of the loss function is directly related to the objective function P2, that is: (twenty one) Among them, W k Let R be the beamforming vector generated by the neural network for the k-th training sample. 1,k The covariance matrix corresponding to the k-th sample, where K is the number of samples in each batch of the training set. Meanwhile, the neural network is trained in the direction of minimizing the loss function, which corresponds exactly to the original increasing trend of the objective function.

[0020] In a second aspect, the present invention provides a deep learning-based intelligent reflector-assisted terahertz secure communication system, including an intelligent reflector-assisted terahertz communication system secure transmission model and a deep learning-based intelligent reflector-assisted terahertz secure communication system transmission module. The aforementioned secure transmission model for a smart reflector-assisted terahertz communication system includes: communication between a base station and a legitimate user is assisted by a smart reflector with N reflective elements, while an unknown eavesdropper passively eavesdrops; wherein the base station with M antennas attempts to transmit to the N-eavesdropping terahertz communication system... r Users with one antenna send confidential information, while those with N antennas send confidential information. e The eavesdropper with one antenna is passively intercepting information; The deep learning-based intelligent reflector-assisted terahertz secure communication system transmission module is configured to calculate the lower bound of the communication system's security rate under imperfect eavesdropping channel conditions based on the information received by the legitimate receiver and the eavesdropping user; construct an optimization objective function to maximize the security rate based on the transmit beamforming matrix and the intelligent reflector phase shift matrix; convert this objective function into two sub-problems; and design a neural network data preprocessing method, network structure, and loss function based on the corresponding sub-problems to calculate the optimal solution; and calculate the optimal security rate based on the obtained optimal solution and perform secure information transmission.

[0021] The beneficial effects of this invention are as follows: This invention utilizes deep learning to design the transmission beam matrix and the phase shift matrix of the intelligent reflector, establishing a system security optimization problem when an eavesdropper is located on the link between the base station and the intelligent reflector. When the eavesdropper is on the link, the forced zero principle is used to invalidate confidential signals leaked into the eavesdropping channel. The original problem is equivalent to two independent subproblems, which are solved using Cauchy's inequality and multi-layer neural networks in deep learning. By seeking a surrogate function to replace the objective function for optimization, the complexity of this method in practical communication is greatly reduced, making it easier to implement. Attached Figure Description

[0022] Figure 1 This is a flowchart of a deep learning-based intelligent reflector-assisted terahertz secure communication method according to Embodiment 1 of the present invention; Figure 2 This is a schematic diagram of the secure transmission model of the deep learning-based intelligent reflector-assisted terahertz communication system described in Embodiment 2 of the present invention. Figure 3 This is a schematic diagram of the deep learning neural network prediction phase shift module provided in Example 1; Figure 4 This is a schematic diagram of the training and verification process of deep learning neural network learning channel data provided in Example 1; where (a) is the total number of training iterations, the horizontal axis of (a) is the number of training iterations, and the vertical axis is the size of the loss function; (b) is the training round loss, the horizontal axis of (b) is the number of training rounds, and the vertical axis is the size of the loss function; the two lines in (b) are the training loss and the verification loss, respectively. Figure 5 This is a simulation diagram of the system security performance under different transmission power conditions for a smart reflective surface equipped with different numbers of reflective elements, as provided in Example 1. Detailed Implementation

[0023] The present invention will be further described below with reference to the embodiments and accompanying drawings, but is not limited thereto.

[0024] Example 1 This embodiment provides a deep learning-based intelligent reflector-assisted terahertz secure communication method, including: (1) Deploy the smart reflective surface on the surface of the building, and the smart reflective surface is equipped with multiple reflective elements; the base station sends information carrying privacy signals to the smart reflective surface, establishes a terahertz channel model and obtains a large amount of channel data; (2) The intelligent reflector constructs a reflected beam by adjusting the phase shift matrix of the reflective element; (3) By using the transmit beam and the reflect beam, signal information can be sent from the base station and reflected to the legitimate receiver through the intelligent reflector, and the user can be eavesdropped on receiving the information. (4) Calculate the lower bound of the confidentiality rate under the imperfect eavesdropping channel state based on the information received by the legitimate receiver and the eavesdropping user; (5) The eavesdropper is located at the link between the base station and the smart reflector. The eavesdropper constructs an optimization problem to maximize the security rate using the transmit beamforming matrix and the smart reflector phase shift matrix. The eavesdropper constructs an optimization objective function to maximize the security rate based on the transmit beamforming matrix and the smart reflector phase shift matrix. The objective function is equivalent to two sub-problems. The eavesdropper designs a data preprocessing method, network structure and loss function of the neural network based on the corresponding sub-problems to calculate the optimal solution. Based on the obtained optimal solution, the optimal security rate is calculated and the information is transmitted securely.

[0025] This embodiment addresses the secure transmission problem in a smart reflector-assisted terahertz communication system. Communication between a base station and a legitimate user is facilitated by a smart reflector with N reflective elements. An unknown eavesdropper passively eavesdrops. Specifically, a base station with M antennas attempts to transmit data to a terahertz communication system with N antennas. r Users with one antenna send confidential information, while those with N antennas send confidential information. e The eavesdropper with one antenna is passively intercepting information.

[0026] In step (1), the base station sends confidential signals to the smart reflector, establishes a terahertz channel model, and considers path fading, molecular absorption, channel gain, and arrival and departure azimuth angles: (1) (2) (3) (4) Among them, H IE Let L be the channel matrix from the smart reflector to Eve, where L represents the number of paths from the BS to the IRS, N is the number of reflector elements, and M is the number of base station antennas. l Indicates the corresponding path, when l When =0, the corresponding path is represented as LoS Path; in addition, Indicates the communication link between BS and IRS. l The channel gain of a path includes transmission loss, molecular absorption, and path loss caused by intercepting eavesdropping devices. The first communication link between the smart reflector IRS and Bob l Complex gain of the path, For the communication link between IRS and Eve l Complex gain of the path, The first communication link between BS and Eve l Channel gain of each path, G r and G tThis is the antenna gain of the transceiver, where the subscript r represents the receiver antenna and the subscript t represents the transmitter antenna. and This represents the array steering vector at the transceiver; , It is the azimuth angle at which the signal arrives. , Given the azimuth angle at which the signal departs, we will randomly generate signal reflection points at different locations to reflect the signal, while the azimuth angles at which the signal arrives and departs satisfy a certain distribution.

[0027] Terahertz signals suffer from severe molecular absorption, propagation loss, and high penetration loss, therefore we assume LoS The path loss formula is: (5) Where r is the distance between the transmitter and the receiver, and j is a complex unit. f For operating frequency, K ( f ) is the molecular absorption coefficient, and c is the speed of light; NLoS Path loss is the Fresnel reflection coefficient R ( f The combination of propagation loss and molecular absorption loss gives the path loss formula as follows: (6) in r 1 represents the distance from the transmitter to the reflecting surface. r 2 represents the distance from the reflector to the receiver; f For operating frequency, K ( f The molecular absorption coefficient () can be written as: (7) Where p and p0 are the system pressure and standard pressure, respectively, and T and T0 are the system temperature and standard temperature, respectively; The absorption cross section is represented by g, which represents the type of gas molecules. For N in the xz plane x ×N z The first USPA array antenna element, the... l The array steering vector corresponding to each path can be represented as: (8) in For wavelength, f c For carrier frequency, The unit spacing is 0 ≤ p ≤ N. x, 0≤q≤N z For the element index of the xz plane, The azimuth angle at which the signal arrives or departs. The elevation angle at which the signal arrives or departs.

[0028] In step (4), the calculation of the lower bound of the confidentiality rate under the imperfect eavesdropping channel state information based on the information received by the legitimate receiver and the eavesdropping user is specifically as follows: (4-1) Determine the information received by the legitimate receiving end and the information received by the eavesdropping user; With the aid of a smart reflector, the confidential signal is reflected to the designated legitimate user Bob's receiver; the signal received by Bob can then be written as: (9) Because the addition of a smart reflector allows eavesdropping users to eavesdrop on two different links, when an eavesdropping user eavesdrops at the link between the base station and the smart reflector, the information they receive is as follows: (10) in, , representing the average power of the signal, the power limit of the transmit beam matrix, and P, respectively. s The transmit power at the base station. The channel matrix from the BS to the smart reflector. Let C be the channel matrix from the smart reflector to Bob and the channel matrix from BS to Eve, respectively. C represents the complex domain, N is the number of reflector elements, and M is the number of base station antennas. r N represents the number of antennas for the user. e The number of antennas used by the eavesdropper. It is the reflection phase shift matrix at the intelligent reflector, where and It represents the phase shift of the reflecting unit of the intelligent reflective surface; j is the phase part of the complex number, the imaginary unit; For phase The reflection coefficient is denoted by ; w is the beamforming vector, and s is the transmitted signal. The mean is 0 and the variance is 0. Additive white Gaussian noise, The mean is 0 and the variance is 0. Additive white Gaussian noise.

[0029] (4-2) Then, we can obtain the signal-to-noise ratio (SNR) received by the legitimate user and the eavesdropping user, and determine the SNR of the legitimate receiver based on the information received by the legitimate receiver and the information received by the eavesdropping user. Signal-to-noise ratio of the user being eavesdropped ; Specifically: (11) (12) in, This is the channel matrix that includes all uncertainties in the ECSI.

[0030] (4-3) Considering the imperfect eavesdropping channel state information, calculate the lower bound of the secrecy rate under the condition that the eavesdropping channel state information is imperfectly known. Specifically:

[0031] (4-3-1) Constructing channel state information for an imperfect eavesdropper; Considering the feedback delay and unreliability of feedback information caused by malicious eavesdropping behavior, the construction of imperfect eavesdropping channel state information specifically employs an additive uncertainty channel model with error bounds to explain the incomplete eavesdropping channel state information, namely: (13) in It is a channel matrix that includes all uncertainties in ECSI. To estimate the error between the channel matrix and the actual matrix, a continuous set Ω contains all possible CSI uncertainties, and its norm is given by... As a boundary.

[0032] (4-3-2) Based on the triangle inequality, Cauchy inequality, and the maximum signal-to-noise ratio at the eavesdropping user, the lower bound of the secrecy rate under the imperfect eavesdropping channel condition is obtained; specifically: By using the triangle inequality and Cauchy's inequality, for any vector... as well as The following inequality always holds true. (14) Therefore, the inequality is satisfied. ,in, The coefficients encompass all uncertain channels.

[0033] Therefore, the maximum signal-to-noise ratio for the eavesdropper at the base station to the smart reflector link is: (15) in, To estimate the error between the channel matrix and the actual matrix at the eavesdropping link; With partial eavesdropper channel state information, the lower bound of the confidentiality rate between the base station and the legitimate receiver is: (16) in , This represents the maximum signal-to-noise ratio that an eavesdropping user can receive under any channel implementation. Signal-to-noise ratio at the eavesdropper's location. These represent the signal-to-noise ratios of the eavesdropper at the base station to the smart reflector link and the smart reflector to the user link, respectively.

[0034] In step (5), an optimization objective function to maximize the security rate is constructed based on the transmit beamforming matrix and the smart reflector phase shift matrix. Specifically, when the eavesdropper is located on the link from the base station to the smart reflector, the objective function is as follows: (17) Among the constraints This means that when Eve is near the BS, Eve can directly intercept confidential signals emitted from the BS. To reduce information leakage at Eve's location, the forced zero principle is used to invalidate any confidential signals leaked into the eavesdropping channel. Constraints This indicates the total power limit imposed on the transmitter. Constraint This represents the phase shift constraint of the intelligent reflective surface.

[0035] The objective function formula (17) is also equivalent to two subproblems, and the optimal solutions are calculated sequentially, as follows: The objective function is equivalent to two independent subproblems, namely: ; Based on two independent sub-problems, a deep learning-based approach is used to design the optimal transmit beam and intelligent reflector phase shift. The optimal approach is obtained through training and iterative processing using randomly acquired channel data. .

[0036] The CNN network structure includes an input, two convolutional layers, three fully connected layers, a lambda layer, and an output, as follows: Figure 3 As shown.

[0037] Selecting a legitimate channel H IU H BE As the input data, since they are two-dimensional matrices, in order to improve the feature extraction capability of the CNN network, it is necessary to process the input Data1={H IU H BE Preprocessing is performed, and definitions are defined. , where g n It is H IU The nth column, It is H BE The nth line, F n This is the cascaded channel corresponding to the nth element of the IRS; therefore, the original channel can be rewritten as: (18) Specifically, the original data X=[F1,F2,…,F… N ] can be regarded as NN intelligent reflective surface units r ×M matrix, here The phase shift coefficient corresponding to each smart reflective surface unit.

[0038] Since the input data for neural networks requires real numbers, but the original data composed of CSI is complex, we extract the real part, imaginary part, and absolute value of each element to form a new three-dimensional matrix, that is, according to the order of real part, imaginary part, and absolute value of X. .

[0039] like Figure 3 As shown, the preprocessed data X passes through two convolutional layers (CL1, CL2) and three fully connected layers (FCL1, FCL2, FCL3), and is finally output by a Lambda layer. CL1 has 256 2×2 filters, CL2 has 512 2×2 filters, and FCL1 and FCL2 have 64N and 16N neurons, respectively. The number of neurons in FCL3 is the same as the number of IRS reflective elements. To prevent overfitting of the neural network, a Dropout layer is inserted into each fully connected layer, using rectified linear units as the activation function. FCL3 uses a sigmoid activation function, which can compress the output z-value to the range (0,1), and then the optimal phase shift matrix of the IRS is obtained through the Lambda layer. .

[0040] For any given phase shift matrix The objective function can be transformed into: (19) in , For channel H BI H IU H BE Given the transpose of the matrix, and by applying the null space constraint, transforming the problem into a generalized eigenvector problem, the optimal w is: (20) Where P s Let B be the transmit power at the base station, B be the null space of the eavesdropping channel, and q be the matrix. The eigenvector corresponding to its largest eigenvalue is the unit norm eigenvector.

[0041] Then predict Based on this, set Data1 and Data2={ Substitute into formula (20) and then calculate to produce the optimal w. opt .

[0042] Figure 1In the middle, input data2 to obtain the optimal solution, and substitute the updated parameters into formula (16) to solve for the updated confidentiality rate.

[0043] This method employs unsupervised learning and requires no additional annotations. Furthermore, the design of the loss function is directly related to the objective function P2, that is: (twenty one) Among them W k Let R be the beamforming vector generated by the neural network for the k-th training sample. 1,k The covariance matrix corresponding to the k-th sample, where K is the number of samples in each batch of the training set. Meanwhile, the neural network is trained in the direction of minimizing the loss function, which corresponds exactly to the original increasing trend of the objective function.

[0044] In step (5), the entire neural network is used to calculate... Meanwhile, the input data consists of Data1 and Data2. To fully train the entire neural network, we randomly generate 8×10... 5 We used a dataset of 1000 data samples, allocating 70% for training, 20% for validation, and 10% for testing. Furthermore, we set the maximum training epochs to 2000, the batch size to 1000, the initial learning rate to 0.01, and the learning rate to decay by 0.3.

[0045] Figure 4 It can be seen that deep learning neural networks are effective for training and validating channel data, based on... Figure 5 The results show that increasing the transmitter's transmission power improves the overall security performance of the system. It also indicates that as the number of reflective elements on the smart reflector increases, the security rate also increases, with N=36 being greater than the results for N=25 and N=16.

[0046] Example 2 This embodiment provides a deep learning-based intelligent reflector-assisted terahertz secure communication system, such as... Figure 2 As shown, it includes a smart reflector-assisted terahertz secure transmission model and a deep learning-trained terahertz channel module. The smart reflector-assisted terahertz secure transmission model includes a source, a legitimate receiver, a smart reflector, and an eavesdropping user. The intelligent reflector constructs a reflected beam by adjusting the phase shift matrix of the intelligent reflective element; The smart reflector will directly reflect the information carrying privacy signals sent by the source to the legitimate receiving end, and the eavesdropping user will receive the information at the link between the source and the smart reflector. The secure transmission module is configured to calculate the lower bound of the security rate under the condition that the eavesdropping channel state information is not perfectly known, based on the information received by the legitimate receiver and the eavesdropping user. The secure transmission module is configured to construct an optimization objective function that maximizes the security rate based on beamforming and the phase shift matrix of the intelligent reflector. This objective function is equivalent to two sub-problems, and the optimal solutions are calculated sequentially. The information is then transmitted securely with the optimal security rate obtained under the optimal solution.

[0047] Communication between a base station and a legitimate user is assisted by a smart reflective surface with N reflective elements. An unknown eavesdropper is passively eavesdropping. The base station with M antennas attempts to communicate with the user with N antennas. r Users with one antenna send confidential information, while those with N antennas send confidential information. e The eavesdropper with one antenna is passively intercepting information.

[0048] The examples and application scenarios implemented by the above modules and corresponding steps are the same, but are not limited to the content disclosed in Embodiment 1 above. It should be noted that the above modules, as part of the system, can be executed in a computer system such as a set of computer-executable instructions.

[0049] The descriptions of each embodiment in the above embodiments have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0050] The proposed system can be implemented in other ways. For example, the system embodiments described above are merely illustrative, and the division of modules described above is only a logical functional division. In actual implementation, there may be other division methods. For example, multiple modules may be combined or integrated into another system, or some features may be ignored or not executed.

[0051] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of hardware embodiments, software embodiments, or embodiments combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage and optical storage) containing computer-usable program code.

[0052] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0053] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0054] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0055] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0056] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A deep learning-based intelligent reflector-assisted terahertz secure communication method, characterized in that, The steps include the following: (1) Deploy smart reflective surfaces on the surface of buildings, and the smart reflective surfaces are equipped with multiple reflective elements; (2) The intelligent reflector constructs a reflected beam by adjusting the phase shift matrix of the reflective element; (3) Using the transmit beam and the reflect beam, the signal is sent from the base station and reflected to the legitimate receiver through the smart reflector, and the user is eavesdropped on receiving the information; (4) Calculate the lower bound of the confidentiality rate under the imperfect eavesdropping channel state based on the information received by the legitimate receiver and the eavesdropping user; (5) Construct an optimization objective function to maximize the security rate based on the transmit beamforming matrix and the intelligent reflector phase shift matrix. The objective function is equivalent to two sub-problems. Based on the corresponding sub-problems, design the data preprocessing method, network structure and loss function of the neural network to calculate the optimal solution. Based on the obtained optimal solution, calculate the optimal security rate and carry out information security transmission.

2. The deep learning-based intelligent reflector-assisted terahertz secure communication method according to claim 1, characterized in that, In step (1), the base station sends the confidential signal to the smart reflector to establish a terahertz channel model, considering path fading, molecular absorption, channel gain, and arrival and departure azimuth angles: （1） （2） （3） （4） Among them, H BI H is the channel matrix from the BS to the smart reflector. IU H is the channel matrix from the smart reflector to Bob. IE H is the channel matrix from the smart reflector to Eve. BE Let L be the channel matrix from BS to Eve, where L represents the number of paths from BS to IRS, N is the number of reflecting elements, and M is the number of base station antennas. l Indicates the corresponding path, when l When =0, the corresponding path is represented as LoS Path; in addition, Indicates the communication link between BS and IRS. l The channel gain of a path includes transmission loss, molecular absorption, and path loss caused by intercepting eavesdropping devices. The first communication link between the smart reflector IRS and Bob l Complex gain of the path, For the communication link between IRS and Eve l Complex gain of the path, The first communication link between BS and Eve l Channel gain of each path, G r and G t This is the antenna gain of the transceiver, where the subscript r represents the receiver antenna and the subscript t represents the transmitter antenna. and This represents the array steering vector at the transceiver; , It is the azimuth angle at which the signal arrives. , The azimuth angle at which the signal departs; Assumption LoS The path loss formula is: （5） Where r is the distance between the transmitter and the receiver, and j is a complex unit. f For operating frequency, K ( f ) is the molecular absorption coefficient, and c is the speed of light; NLoS Path loss is the Fresnel reflection coefficient R ( f The combination of propagation loss and molecular absorption loss gives the path loss formula as follows: （6） in r 1 represents the distance from the transmitter to the reflecting surface. r 2 represents the distance from the reflector to the receiver; f For operating frequency, K ( f The molecular absorption coefficient is denoted as: （7） Where p and p0 are the system pressure and standard pressure, respectively, and T and T0 are the system temperature and standard temperature, respectively; The absorption cross section is represented by g, which represents the type of gas molecules. For N in the xz plane x ×N z The first USPA array antenna element, the... l The array steering vector corresponding to each path can be represented as: （8） in For wavelength, f c For carrier frequency, The unit spacing is 0 ≤ p ≤ N. x , 0≤q≤N z For the element index of the xz plane, The azimuth angle at which the signal arrives or departs. The elevation angle at which the signal arrives or departs.

3. The deep learning-based intelligent reflector-assisted terahertz secure communication method according to claim 1, characterized in that, In step (4), the calculation of the lower bound of the confidentiality rate under the imperfect eavesdropping channel state information based on the information received by the legitimate receiver and the eavesdropping user is specifically as follows: (4-1) Determine the information received by the legitimate receiving end and the information received by the eavesdropping user; (4-2) Determine the signal-to-noise ratio of the legitimate receiver and the signal-to-noise ratio of the eavesdropping user based on the information received by the legitimate receiver and the information received by the eavesdropping user; (4-3) Considering the imperfect eavesdropping channel state information, calculate the lower bound of the confidentiality rate under the condition that the eavesdropping channel state information is imperfectly known.

4. The deep learning-based intelligent reflector-assisted terahertz secure communication method according to claim 3, characterized in that, In step (4-1), the information received by the legitimate receiving end is: （9） When a user eavesdrops at the link between the base station and the smart reflector, the information they receive is: （10） in, , representing the average power of the signal, the power limit of the transmit beam matrix, and P, respectively. s The transmit power at the base station. The channel matrix from the BS to the smart reflector is as follows: Let C be the channel matrix from the smart reflector to Bob and the channel matrix from BS to Eve, respectively. C represents the complex domain, N is the number of reflector elements, and M is the number of base station antennas. r N represents the number of antennas for the user. e The number of antennas used by the eavesdropper. It is the reflection phase shift matrix at the intelligent reflector, where and It is the phase shift of the reflecting unit of the intelligent reflecting surface, where j is the complex phase part, the imaginary unit; For phase is The reflection coefficient is denoted by ; w is the beamforming vector, and s is the transmitted signal. The mean is 0 and the variance is 0. Additive white Gaussian noise, The mean is 0 and the variance is 0. Additive white Gaussian noise.

5. The deep learning-based intelligent reflector-assisted terahertz secure communication method according to claim 4, characterized in that, In step (4-2), the signal-to-noise ratio of the legitimate receiver is determined based on the information received by the legitimate receiver and the information received by the eavesdropping user. Signal-to-noise ratio of the user being eavesdropped Specifically: （11） （12） in, This is the channel matrix that includes all uncertainties in the ECSI.

6. The deep learning-based intelligent reflector-assisted terahertz secure communication method according to claim 3, characterized in that, In step (4-3), the step of considering the imperfect eavesdropping channel state information and calculating the lower bound of the secrecy rate under the imperfect eavesdropping channel state specifically involves: (4-3-1) Constructing channel state information for an imperfect eavesdropper; (4-3-2) Based on the triangle inequality, Cauchy inequality and the maximum signal-to-noise ratio at the eavesdropping user, the lower bound of the confidentiality rate under the imperfect eavesdropping channel state is obtained.

7. The deep learning-based intelligent reflector-assisted terahertz secure communication method according to claim 6, characterized in that, In step (4-3-1), considering the feedback delay and unreliability of feedback information caused by the malicious behavior of the eavesdropper, the construction of imperfect eavesdropper channel state information specifically adopts an additive uncertainty channel model with error bounds to explain the incomplete eavesdropping channel state information, that is: （13） in It is a channel matrix that includes all uncertainties in ECSI. To estimate the error between the channel matrix and the actual matrix, a continuous set Ω contains all possible CSI uncertainties, and its norm is given by... As a boundary; In step (4-3-2), the lower bound of the confidentiality rate under the imperfect eavesdropping channel state is obtained based on the triangle inequality, Cauchy inequality, and the maximum signal-to-noise ratio at the eavesdropping user. Specifically: By using the triangle inequality and Cauchy's inequality, for any vector... as well as The following inequality always holds true. （14） Therefore, the inequality is satisfied. ,in, The coefficients encompass all uncertain channels; Therefore, the maximum signal-to-noise ratio for the eavesdropper at the base station to the smart reflector link is: （15） in, To estimate the error between the channel matrix and the actual matrix at the eavesdropping link; With partial eavesdropper channel state information, the lower bound of the confidentiality rate between the base station and the legitimate receiver is: （16） in , This represents the maximum signal-to-noise ratio that an eavesdropping user can receive under any channel implementation. Signal-to-noise ratio at the eavesdropper's location. These represent the signal-to-noise ratios of the eavesdropper at the base station to the smart reflector link and the smart reflector to the user link, respectively.

8. The deep learning-based intelligent reflector-assisted terahertz secure communication method according to claim 1, characterized in that, In step (5), an optimization objective function to maximize the security rate is constructed based on the transmit beamforming matrix and the smart reflector phase shift matrix. Specifically, when the eavesdropper is located on the link from the base station to the smart reflector, the objective function is as follows: （17） Among the constraints This means that when Eve gets close to BS, Eve directly intercepts the confidential signals emitted from BS; The forced zero principle is used to invalidate confidential signals leaked into the eavesdropping channel; constraints This indicates the total power limit imposed on the transmitter; constraint This represents the phase shift constraint of the intelligent reflective surface; In step (5), the objective function formula (17) is also equivalent to two subproblems, and the optimal solution is calculated sequentially, as follows: The objective function is equivalent to two independent subproblems, namely: ; Based on two independent sub-problems, a deep learning-based approach is used to design the optimal transmit beam and intelligent reflector phase shift. The optimal approach is obtained through training and iterative processing using randomly acquired channel data. .

9. The deep learning-based intelligent reflector-assisted terahertz secure communication method according to claim 8, characterized in that, In step (5), the CNN network structure is set to include input, two convolutional layers, three fully connected layers, a lambda layer, and output; Selecting a legitimate channel H IU H BE As input data, for the input Data1={H IU H BE Preprocessing is performed, and definitions are defined. , where g n It is H IU The nth column, It is H BE The nth line, F n This is the cascaded channel corresponding to the nth element of the IRS; therefore, the original channel is rewritten as: （18） Specifically, the original data X=[F1,F2,…,F… N ], considered N N intelligent reflective surface units r ×M matrix, here The phase shift coefficient corresponding to each smart reflective surface unit; Extract the real part, imaginary part, and absolute value of each element to form a new three-dimensional matrix, i.e., in the order of taking the real part, imaginary part, and absolute value of X. ; The preprocessed data X passes through two convolutional layers (CL1, CL2) and three fully connected layers (FCL1, FCL2, FCL3), and is finally output by a Lambda layer. CL1 has 256 2×2 filters, CL2 has 512 2×2 filters, FCL1 and FCL2 have 64N and 16N neurons respectively, and the number of neurons in FCL3 is the same as the number of IRS reflective elements. A Dropout layer is inserted into each fully connected layer, and a rectified linear unit is used as the activation function. FCL3 uses "Sigmoid" as the activation function to compress the output z value to the range (0,1), and then the optimal phase shift matrix of the IRS is obtained through the Lambda layer. ; For any given phase shift matrix The objective function is transformed into: （19） in , For channel H BI H IU H BE Given the transpose of the matrix, and by applying the null space constraint, transforming the problem into a generalized eigenvector problem, the optimal w is: （20） Where P s Let B be the transmit power at the base station, B be the null space of the eavesdropping channel, and q be the matrix. The eigenvector corresponding to its largest eigenvalue; Then predict Based on this, set Data1 and Data2={ Substitute into formula (20) and then calculate to produce the optimal w. opt ; Find the optimal solution, and substitute the updated parameters into formula (16) to calculate the updated confidentiality rate; Unsupervised learning is used, and no additional annotation is required. Furthermore, the design of the loss function is directly related to the objective function P2, that is: （21） Among them, W k Let R be the beamforming vector generated by the neural network for the k-th training sample. 1,k The covariance matrix corresponding to the k-th sample, where K is the number of samples in each batch of the training set. Meanwhile, the neural network is trained in the direction of minimizing the loss function, which corresponds exactly to the original increasing trend of the objective function.

10. A deep learning-based intelligent reflector-assisted terahertz secure communication system, characterized in that, This includes a secure transmission model for a smart reflector-assisted terahertz communication system and a deep learning-based transmission module for a smart reflector-assisted terahertz secure communication system. The aforementioned secure transmission model for a smart reflector-assisted terahertz communication system includes: communication between a base station and a legitimate user is assisted by a smart reflector with N reflective elements, while an unknown eavesdropper passively eavesdrops; wherein the base station with M antennas attempts to transmit to the N-eavesdropping terahertz communication system... r Users with one antenna send confidential information, while those with N antennas send confidential information. e The eavesdropper with one antenna is passively intercepting information; The deep learning-based intelligent reflector-assisted terahertz secure communication system transmission module executes the method described in any one of claims 1-9. It is configured to calculate the lower bound of the communication system's security rate under imperfect eavesdropping channel conditions based on the information received by the legitimate receiver and the eavesdropping user; construct an optimization objective function that maximizes the security rate based on the transmit beamforming matrix and the intelligent reflector phase shift matrix; convert the objective function into two sub-problems; design a neural network data preprocessing method, network structure, and loss function based on the corresponding sub-problems to calculate the optimal solution; and calculate the optimal security rate based on the obtained optimal solution and perform secure information transmission.

Images