RIS-assisted secure communication method and system giving consideration to downstream task and communication security

Abstract

The invention relates to an RIS-assisted secure communication method considering downstream tasks and communication security. According to the method, a sending end and a target signal receiver complete a picture classification communication task with the help of N RIS units; comprising the following steps: a sending stage: a sending end firstly extracts a feature vector from an input picture by using a feature encoder, and then converts the feature vector into a signal form which can be transmitted by a channel by means of a channel pre-coding matrix; in the transmission stage, the N RIS units carry out reflection, amplitude modulation and phase shift operation on the sent signals; and a classification stage: after the signals reflected by the N RIS units are received, integrating the signals into feature vectors of the picture, and calculating through a classifier to obtain categories, thereby completing a classification task. According to the method, higher reasoning accuracy is obtained, and task-oriented wireless communication is realized; and the low classification accuracy of the eavesdropping end Eve is kept, and the communication information security is ensured.

Description

Technical Field

[0001] The present invention relates to a secure communication method and system assisted by RIS that takes into account both downstream tasks and communication security, and belongs to the technical field of wireless communication. Background Art

[0002] Reconfigurable intelligent surfaces (RIS) are software-controlled metasurfaces equipped with low-complexity passive reflective elements. They have shown significant promise in improving wireless communication performance [Ahmed M, Raza S, Soofi AA, et al. Active reconfigurable intelligent surfaces: Expanding the frontiers of wireless communication-a survey [J]. IEEE Communications Surveys & Tutorials, 2024.]. While RIS improves signal reception performance, it also increases the likelihood of communication information leakage. Recently, RIS designed for communication security has attracted widespread attention in the field of physical layer security [Khoshafa MH, Maraqa O, Moualeu JM, et al. RIS-assisted physical layer security in emerging RF and optical wireless communication systems: A comprehensive survey [J]. IEEE Communications Surveys & Tutorials, 2024.]. Their primary application is to enhance secure transmission in multi-antenna systems. The literature [Wu Y, Luo J, Chen W, et al. Passive secure communications based on reconfigurable intelligent surface [J]. IEEE Communications Letters, 2022, 27(2): 472-476.] studies a passive secure communication scheme using RIS, where RIS allocates power from the input signal to simultaneously transmit confidential information and generate passive jamming signals. The literature [Dong L, Wang HM, Bai J. Active reconfigurable intelligent surface aided secure transmission [J]. IEEE Transactions on Vehicular Technology, 2021, 71(2): 2181-2186.] designs an active RIS-assisted communication system by maximizing the confidentiality rate.The paper [Pan C, Ren H, Wang K, et al. Reconfigurable intelligent surfaces for 6G systems: Principles, applications, and research directions [J]. IEEE Communications Magazine, 2021, 59(6): 14-20.] proposes an active RIS-assisted system with two goals: minimizing the transmission power of the drone-mounted base station and improving the achievable confidentiality rate. The authors of the paper [Niu H, Lin Z, Chu Z, et al. Joint beamforming design for secure RIS-assisted IoT networks [J]. IEEE internet of things journal, 2022, 10(2): 1628-1641.] adopt a two-pronged approach. They use active RIS to enhance confidentiality signals while using passive RIS to improve user confidentiality in the presence of multiple eavesdroppers.

[0003] However, these previous works mainly focus on transmission-level security metrics such as confidentiality rate, without considering the ultimate purpose of many wireless transmissions, which is to enable the receiver to complete specific tasks such as classification or intelligent reasoning. To this end, task-oriented communication has emerged as a promising paradigm [Xie H, Qin Z, Tao X, et al. Task-oriented multi-user semantic communications [J]. IEEE Journal on Selected Areas in Communications, 2022, 40 (9): 2584-2597.], which optimizes communication system resources according to the needs of task execution and ensures reliable task completion [Gündüz D, Qin Z, Aguerri IE, et al. Beyond transmitting bits: Context, semantics, and task-oriented communications [J]. IEEE Journal on Selected Areas in Communications, 2022, 41 (1): 5-41.], especially under harsh communication conditions. Despite the growing interest in this field, most research aims to improve the task performance of the receiver, while security issues in task-oriented communication remain underexplored. Summary of the Invention

[0004] In response to the shortcomings of the existing technology, the present invention provides a secure communication method assisted by RIS that takes into account both downstream tasks and communication security. On the one hand, the present invention proposes an optimization method for RIS signal improvement and a high correlation between the transmission data and the task, thereby ensuring that the receiver can complete the communication task; on the other hand, the information carried by the signal reflected by the RIS proposed by the present invention cannot be identified by eavesdroppers, thereby ensuring communication security. Specifically, the present invention proposes a method based on maximum coding rate reduction (MCR) 2 ) system objective function. First, take MCR 2 The feature encoder is pre-trained for the objective function to ensure that the feature vectors extracted by the feature encoder are highly relevant to the task. Secondly, the present invention proposes a RIS optimization objective (STOR) that simultaneously satisfies task orientation and ensures communication security. Taking into account transmit power limitations and RIS reflection unit amplitude constraints, the present invention formulates the system optimization problem as an alternating optimization problem (AO) of the channel precoder and the RIS reflection vector, and proposes the BGP and RGP methods to solve the two sub-problems, respectively.

[0005] The present invention also provides a secure communication system assisted by RIS that takes into account both downstream tasks and communication security.

[0006] The technical solution of the present invention is:

[0007] A secure communication method assisted by RIS that balances downstream tasks and communication security. This method involves a transmitter and a target signal receiver completing an image classification communication task with the help of N RIS units. Specifically, it includes:

[0008] Transmission phase: The transmitter first applies a feature encoder to the input image to extract feature vectors, and then uses the channel precoding matrix to convert the feature vectors into a signal form that can be transmitted by the channel;

[0009] Transmission phase: N RIS units reflect, amplitude modulate, and phase shift the transmitted signal;

[0010] Classification stage: The target signal receiver is equipped with multiple receiving antennas. After receiving the signal reflected by N RIS units, it integrates the signal into a feature vector of the image and then calculates the category through a classifier, thus completing the classification task.

[0011] According to a preferred embodiment of the present invention, the transmitting end first applies a feature encoder to the input image to extract a feature vector; including:

[0012] Perform a series of preprocessing operations on the input image;

[0013] The feature encoder gradually extracts the features of the image through convolution, pooling, and multi-layer stacking operations, and converts them into a one-dimensional feature vector s.

[0014] According to the present invention, the feature vector Adjust to plural form

[0015]

[0016] In formula (I), represents the imaginary unit, assuming D = 2F, where F is half the dimension of the eigenvector s, [s] 1:F Refers to the first half of s, [s] F+1:D It's the second half.

[0017] According to the preferred embodiment of the present invention, the channel precoder of the transmitting end is defined The output signal is N t is the number of transmitting antennas at the transmitter, x satisfies the transmit power constraint shown in formula (II):

[0018]

[0019] In formula (II), E(·) represents the mean operator, ‖·‖2 represents the two-norm operator, express The covariance matrix, P bs represents the maximum signal transmission power of the transmitter. The superscript H represents the conjugate transpose of the matrix. tr() refers to the trace operator of the matrix, which calculates the sum of the diagonal elements of the matrix.

[0020] According to the preferred embodiment of the present invention, a wireless communication channel is established, and the communication signal transmission process of the target signal receiver and the eavesdropper is expressed as formula (III):

[0021]

[0022] In formula (III), y B and y E They represent the eigenvectors of Bob and Eve respectively fused according to the reflected signals they received; represents Bob’s transmission channel matrix, where represents the linear channel matrix between Alice and Bob, represents the reflection channel matrix between RIS and Bob, represents the transmission channel matrix between Alice and RIS; represents Eve’s transmission channel matrix, where represents the linear channel matrix between Alice and Eve, represents the reflection channel matrix between RIS and Eve; Represents the RIS phase shift vector is a diagonal matrix, And θ i ∈(0,2π] represents the phase shift caused by the i-th RIS unit; for passive RIS, the amplification factor ρ i =1, while for active RIS, the amplification factor satisfies 1<ρ i <ρ max ; N r is the number of receiving antennas of Bob and Eve; represents the Gaussian noise vector, where δ represents the noise power, I represents the identity matrix, Alice represents the sender, Bob represents the target signal receiver, and Eve represents the eavesdropper.

[0023] Further preferably, the wireless communication channel is a Rayleigh channel.

[0024] According to the present invention, preferably, when the RIS is a passive RIS, the phase shift vector of the RIS unit satisfies the constraint condition of formula (IV):

[0025]

[0026] When the RIS is an active RIS, it is subject to the power constraint of the RIS, which can be expressed as:

[0027]

[0028] Among them, δ r is the noise power added by the active RIS, represents the matrix binorm, P ris is the maximum power of the active RIS.

[0029] According to the preferred embodiment of the present invention, the objective function MCR of the pre-trained feature encoder 2 , expressed as:

[0030]

[0031] In formula (VI), for passive RIS, For active RIS, in represents the signal-to-noise ratio, ε is the lossy decoding rate; and Represents the covariance matrix of the eigenvector and the covariance matrix of the eigenvector of category j, p jis the prior probability of the feature vector received by the edge server and related to category j.

[0032] Preferably, according to the present invention, the optimization problem STOR of the channel precoder and the passive / active RIS is established as formula (VII):

[0033]

[0034] In formula (VII), Represents the vector form of the channel precoding matrix, represents the Kronecker product, T represents the matrix transpose; the coefficient β is used to mediate the weight between the communication task and security.

[0035] According to the present invention, the multivariable optimization problem in formula (VII) is preferably transformed into an alternating optimization problem with single variable constraints, specifically comprising:

[0036] 1) Fix the RIS phase shift vector φ and transform Equation (VII) into Equation (VIII):

[0037] max V G(V)=(1-β)Ψ B (V)-βΛ E (V)(VIII)

[0038] stv H Mv≤P bs

[0039] 2) Fix Alice’s channel precoding matrix V and transform Equation (VII) into Equation (IX) for passive / active RIS:

[0040]

[0041] Further preferably, solving the optimal channel precoding matrix V using a BGP algorithm includes:

[0042] 3) Calculate the gradient of the objective function of formula (VIII) with respect to V:

[0043]

[0044] in,

[0045]

[0046] in,

[0047] 4) When the transmission power of the transmitting end Alice satisfies v H Mv=P bsWhen , the maximum value of G(v) is obtained, so The projection is:

[0048]

[0049] In formula (XII),

[0050] 5) Update v by formula (XIII):

[0051]

[0052] In formula (XIII), By finding a θ∈[0,π / 2) that satisfies θ=argmaxG(V), the optimal v and V are obtained based on formula (XII);

[0053] Repeat the above optimization process, i.e., step 3) to step 5) until convergence, and obtain the optimal precoding matrix V with a fixed RIS phase shift vector φ.

[0054] Further preferably, the RGP algorithm is used to solve the RIS optimization problem, including:

[0055] For passive RIS, the constraint in Equation (IX) is transformed into a convex constraint:

[0056] tr(φφ H )=N and||φ|| ∞ ≤1(XIV);

[0057] 6) Use Approximate infinity norm l ∞ ; Using the potential barrier method, using the logarithmic potential barrier function Integrate the non-negative constraints to approximate the penalty for violating the constraints, where t is a constant used to adjust the penalty; the objective function in Equation (IX) can be rewritten as:

[0058] max φ Υ(φ,h)=G(φ)+I(1-‖φ‖ h )(XV);

[0059] 7) Calculate the gradient of the objective function of formula (XV) with respect to φ:

[0060]

[0061] In formula (XV), at the same time in

[0062]

[0063] 8) Project the search direction to formula (XVIII):

[0064]

[0065] Among them, <, > represents the inner product, by finding a satisfy And update φ using formula (XIX):

[0066]

[0067] 9) Use equation (XX) to project φ to satisfy the constraints:

[0068]

[0069] Repeat the above optimization process, i.e., step 6) to step 10) until convergence, and obtain the optimal RIS phase shift vector φ that meets the constraints under the given precoding matrix V;

[0070] For active RIS, rewrite the constraint as:

[0071]

[0072] Formula (XXI) is a non-convex constraint, and further introduces the penalty function T(u) = λ(e u -1) to convert the constraint term into a penalty term to punish the violation of the constraint condition, where λ is used to adjust the penalty value; and the objective function is expressed as:

[0073] max φ Υ(φ)=Ψ(φ)-T(g(φ))=Ψ(φ)-λ(e g(φ) -1)(XXII).

[0074] g(φ) is used as parameter u to pass into the penalty function T(u) for calculation.

[0075] A computer device includes a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, it implements the steps of the above-mentioned secure communication method assisted by RIS that takes into account both downstream tasks and communication security.

[0076] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the secure communication method assisted by RIS that takes into account both downstream tasks and communication security.

[0077] A secure communication system assisted by RIS that takes into account both downstream missions and communication security, including:

[0078] The sending module is configured as follows: the sending end first applies a feature encoder to the input image to extract a feature vector, and then uses a channel precoding matrix to convert the feature vector into a signal form that can be transmitted by the channel;

[0079] The transmission module is configured as follows: N RIS units perform reflection, amplitude modulation, and phase shift operations on the transmitted signal;

[0080] The classification module is configured as follows: the target signal receiver is equipped with multiple receiving antennas. After receiving the signal reflected by N RIS units, it integrates the signal into the feature vector of the image, and then calculates the category through the classifier to complete the classification task.

[0081] The beneficial effects of the present invention are:

[0082] In view of the current demand for task-related data in wireless communications and the fact that RIS, which is designed with physical layer security as its design goal, ignores specific task requirements, this paper proposes a RIS-assisted task-oriented secure communication system design method. The design goal of the feature encoder in the pre-trained model is MCR. 2 , and proposed STOR as the joint optimization objective function of the channel precoder and RIS phase shift vector to solve the maximization problem under the constraints of transmission power and phase shift vector amplitude; proposed BGP and RGP methods to solve the channel precoder and RIS phase shift vector respectively. Under the alternating optimization strategy, the final designed scheme achieved higher inference accuracy at the target receiver Bob in a low channel environment (SNR is -20dB to 10dB) compared with the communication system without RIS assistance, realizing task-oriented wireless communication; and maintained the low classification accuracy of the eavesdropping end Eve to ensure the security of communication information; in addition, the designed scheme is also applicable to active RIS.

[0083] Under the same SNR (signal-to-noise ratio) range, the task-oriented and communication-secure RIS design method proposed in this invention was verified on the ModelNet10 dataset against a communication system without RIS. Specifically, when SNR = -10dB and MAP is used as the receiving end, the classification accuracy of the target receiving end Bob with the assistance of RIS is 35% higher than that of Bob without RIS. While improving the target receiving end Bob, the RIS design method proposed in this invention also ensures the security of communication information. Specifically, the classification accuracy achieved by the eavesdropper Eve in the communication system equipped with RIS is almost the same as that in the system without RIS, and both are approximately equal to 10%, which is equivalent to the accuracy of randomly selected classification categories. In addition, the design method proposed in this invention is applicable not only to passive RIS but also to active RIS, which fully demonstrates the versatility of this invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0084] Figure 1 It is a schematic diagram of a secure communication method assisted by RIS according to the present invention, which takes into account both downstream tasks and communication security;

[0085] Figure 2 This is a comparison chart of the classification accuracy of the communication system of the present invention with passive RIS assistance and without RIS assistance under different maximum a posteriori probability classifiers (MAP);

[0086] Figure 3 This is a comparison chart of the classification accuracy of the communication system of the present invention with passive RIS assistance and without RIS assistance using different K-nearest neighbor classifiers (KNN);

[0087] Figure 4 This is a comparison chart of the classification accuracy of the communication system of the present invention with passive RIS assistance and without RIS assistance using different neural network classifiers (NN);

[0088] Figure 5 This is a comparison chart of the classification accuracy of the K-nearest neighbor classifier (KNN) assisted by active RIS in the present invention;

[0089] Figure 6 This is a comparison chart of the classification accuracy of the K-nearest neighbor classifier (MAP) assisted by active RIS in the present invention;

[0090] Figure 7 This is a comparison chart of the classification accuracy of the K nearest neighbor classifier (NN) assisted by active RIS in the present invention;

[0091] Figure 8 This is a relationship diagram between the weight coefficient β and the classification accuracy and communication security protection under the MAP classifier and SNR=-10dB in the method of the present invention. DETAILED DESCRIPTION

[0092] The present invention will be further described below with reference to the accompanying drawings and embodiments, but is not limited thereto.

[0093] Example 1

[0094] A secure communication method assisted by RIS that takes into account both downstream tasks and communication security; it means that the sender and the target signal receiver complete a communication task of image classification with the help of N RIS units; Figure 1 As shown, specifically including:

[0095] Transmission phase: The transmitter first applies a feature encoder to the input image to extract feature vectors, and then uses the channel precoding matrix to convert the feature vectors into a signal form that can be transmitted by the channel;

[0096] Transmission stage: N RIS units are embedded on surrounding building surfaces to reflect, amplitude modulate, and phase shift the transmitted signal;

[0097] Classification stage: The target signal receiver is equipped with multiple receiving antennas. After receiving the signal reflected by N RIS units, it integrates the signal into a feature vector of the image and then calculates the category through a classifier, thus completing the classification task.

[0098] At the same time, the eavesdropper attempts to intercept information based on the reflected signal, but the RIS design method proposed in the present invention can ensure that the eavesdropper fails to intercept the information.

[0099] A RIS design method that takes into account both downstream tasks and communication security, the specific steps of the method include:

[0100] (1) Establish a task-oriented secure communication system, signal transmission model, and channel model;

[0101] (2) MCR 2 The feature encoder is pre-trained for the objective function, and the joint optimization objective of the pre-encoder and RIS is STOR;

[0102] (3) Convert the optimization problem with multivariable constraints into an alternating optimization problem with single variable constraints for the precoder and RIS;

[0103] (4) Using the BGP algorithm to solve the precoder optimization problem;

[0104] (5) Solve the optimization problem of RIS using the RGP algorithm;

[0105] (6) Use the alternating iterative algorithm to solve the optimization problem of the entire system.

[0106] First, the transmitter (Alice) extracts features from the image to be sent using a feature encoder. The channel precoder then converts the features into a data format suitable for wireless channel transmission. The transmitted signal is reflected by the RIS, increasing both transmission distance and communication performance. The reflected signal from the RIS is received by the target signal receiver (Bob) and intercepted by the eavesdropper (Eve). The target signal receiver and the eavesdropped on use a classifier to extract information from the received signal and obtain a classification result.

[0107] Example 2

[0108] The difference between the RIS-assisted secure communication method for balancing downstream tasks and communication security described in Example 1 is that:

[0109] The sending end first applies a feature encoder to the input image to extract a feature vector; this includes:

[0110] Alice, the sender, is equipped with a feature encoder composed of a neural network (a computational model that mimics the structure and function of biological neural networks and updates parameters through forward propagation and gradient backpropagation). Before the image is fed into the feature encoder, it undergoes a series of preprocessing operations (such as resizing) to ensure the consistency and quality of the image data.

[0111] The feature encoder gradually extracts image features through convolution, pooling, and multi-layer stacking operations, converting them into a one-dimensional feature vector s. This feature vector contains important information about the image. For example, if an input image of a cat is input, the resulting feature vector may contain features that cats have but not dogs. Features are stored as numbers that only a trained feature encoder can distinguish.

[0112] The feature vector Adjust to plural form For wireless channel transmission:

[0113]

[0114] In formula (I), represents the imaginary unit, assuming D = 2F, where F is half the dimension of the eigenvector s, [s] 1:F Refers to the first half of s, [s] F+1:D It's the second half.

[0115] Define the channel precoder at the transmitter The output signal is N t is the number of transmitting antennas at the transmitter, x satisfies the transmit power constraint shown in formula (II):

[0116]

[0117] In formula (II), E(·) represents the mean operator, ‖·‖2 represents the two-norm operator, express The covariance matrix, P bs represents the maximum signal transmission power of the transmitter. The superscript H represents the conjugate transpose of the matrix. tr() refers to the trace operator of the matrix, which calculates the sum of the diagonal elements of the matrix.

[0118] Once a wireless communication channel is established, the target signal receiver and the eavesdropper's receiving communication signal transmission process can be expressed as formula (III):

[0119]

[0120] In formula (III), y B and yE They represent the eigenvectors of Bob and Eve respectively fused according to the reflected signals they received; represents Bob’s transmission channel matrix, where represents the linear channel matrix between Alice and Bob, represents the reflection channel matrix between RIS and Bob, represents the transmission channel matrix between Alice and RIS; represents Eve’s transmission channel matrix, where represents the linear channel matrix between Alice and Eve, represents the reflection channel matrix between RIS and Eve; Represents the RIS phase shift vector is a diagonal matrix, And θ i ∈(0,2π] represents the phase shift caused by the i-th RIS unit; for passive RIS, the amplification factor ρ i =1, while for active RIS, the amplification factor satisfies 1<ρ i <ρ max ; N r is the number of receiving antennas of Bob and Eve; represents the Gaussian noise vector, where δ represents the noise power, I represents the identity matrix, Alice represents the sender, Bob represents the target signal receiver, and Eve represents the eavesdropper.

[0121] The wireless communication channel is a Rayleigh channel.

[0122] When the RIS is a passive RIS, the phase shift vector of the RIS unit satisfies the constraint condition of formula (IV):

[0123]

[0124] When the RIS is an active RIS, it is no longer subject to the amplitude limiting constraint of formula (IV) because it is equipped with a signal amplification circuit, but is subject to the power constraint of the RIS, which is expressed as:

[0125]

[0126] Among them, δ r is the noise power added by the active RIS, represents the matrix binorm, P ris is the maximum power of the active RIS.

[0127] For a fair comparison, the present invention assumes that Bob and Eve are equipped with the same classifier, and the classification labels can be obtained by calculating the classifier based on their respective received signals.

[0128] Objective function MCR of pre-trained feature encoder 2 , expressed as:

[0129]

[0130] In formula (VI), for passive RIS, For active RIS, in represents the signal-to-noise ratio (SNR), ε is the lossy decoding rate; and Represents the covariance matrix of the eigenvector and the covariance matrix of the eigenvector of category j, p j is the prior probability of the feature vector received by the edge server and related to category j.

[0131] According to the literature [Zhou J, Guo S, Ye J, et al. Maximal Coding Rate Reduction: A Unified Approach for Task-Oriented RIS and Communication [J]. IEEE Transactions on Vehicular Technology, 2024.], and combined with the constraints, the optimization problem STOR for the channel precoder and passive / active RIS is established as formula (VII):

[0132]

[0133] In formula (VII), Represents the vector form of the channel precoding matrix, where β represents the Kronecker product and T represents the matrix transpose. The coefficient β is used to adjust the weight between the communication task and security. In extreme communication environments, a larger β is set to ensure communication security; conversely, in benign environments, a smaller β is set to ensure the completion of the communication task.

[0134] The multivariable optimization problem in formula (VII) is transformed into an alternating optimization problem with single variable constraints, specifically including:

[0135] 1) Fix the RIS phase shift vector φ and transform Equation (VII) into Equation (VIII):

[0136] max V G(V)=(1-β)Ψ B (V)-βΛ E (V)(VIII)

[0137] stv H Mv≤P bs

[0138] 2) Fix Alice’s channel precoding matrix V and transform Equation (VII) into Equation (IX) for passive / active RIS:

[0139]

[0140] The optimal channel precoding matrix V is solved using the BGP algorithm; including:

[0141] 3) Calculate the gradient of the objective function of formula (VIII) with respect to V:

[0142]

[0143] in,

[0144]

[0145] in,

[0146] 4) When the transmission power of the transmitting end Alice satisfies v H Mv=P bs When , the maximum value of G(v) is obtained, so The projection is:

[0147]

[0148] In formula (XII),

[0149] 5) Update v by formula (XIII):

[0150]

[0151] In formula (XIII), By finding a θ∈[0,π / 2) that satisfies θ=argmaxG(V), the optimal v and V are obtained based on formula (XII);

[0152] Repeat the above optimization process, i.e., step 3) to step 5) until convergence, and obtain the optimal precoding matrix V with a fixed RIS phase shift vector φ.

[0153] The RGP algorithm is used to solve the RIS optimization problem, according to the literature [Ye J, Guo S, Alouini M S. Joint reflecting and precoding designs for SER minimization in reconfigurable intelligent surfaces assisted MIMO systems [J]. IEEE Transactions on Wireless Communications, 2020, 19(8): 5561-5574.], including:

[0154] For passive RIS, the constraint in Equation (IX) is transformed into a convex constraint:

[0155] tr(φφ H )=N and||φ|| ∞ ≤1(XIV);

[0156] 6) Use Approximate infinity norm e ∞ ; Using the potential barrier method, using the logarithmic potential barrier function Integrate the non-negative constraints to approximate the penalty for violating the constraints, where t is a constant used to adjust the penalty; the objective function in Equation (IX) can be rewritten as:

[0157] max φ Υ(φ,h)=G(φ)+I(1-‖φ‖ h )(XV);

[0158] 7) Calculate the gradient of the objective function of formula (XV) with respect to φ:

[0159]

[0160] In formula (XV), at the same time in

[0161]

[0162] 8) Project the search direction to formula (XVIII):

[0163]

[0164] Among them, <,> represents the inner product, by finding a satisfy And update φ using formula (XIX):

[0165]

[0166] 9) Use equation (XX) to project φ to satisfy the constraints:

[0167]

[0168] Repeat the above optimization process, i.e., step 6) to step 10) until convergence, and obtain the optimal RIS phase shift vector φ that meets the constraints under the given precoding matrix V;

[0169] For active RIS, rewrite the constraint as:

[0170]

[0171] Formula (XXI) is a non-convex constraint, and further introduces the penalty function T(u) = λ(e u -1) to convert the constraint term into a penalty term to punish the violation of the constraint condition, where λ is used to adjust the penalty value; and the objective function is expressed as:

[0172] max φ Υ(φ)=Ψ(φ)-T(g(φ))=Ψ(φ)-λ(e g(φ) -1)(XXII).

[0173] g(φ) is used as parameter u to pass into the penalty function T(u) for calculation.

[0174] The optimization problem of the entire system can be solved by using the alternating iterative algorithm as follows:

[0175] 1) Initialize the precoding matrix V to satisfy v H Mv=P bs , for passive RIS, initialize the phase shift vector φ to satisfy |φ i |=1, and for active RIS, randomly initialize 1<|φ i |≤10;

[0176] 2) Given a phase shift vector φ, solve the precoding matrix V: Calculate according to formula (X) Calculated according to formula (XII) And according to Find one Used to update V and v according to formula (XIII), repeat the above steps until convergence;

[0177] 3) Given the precoding matrix V, solve the phase shift vector φ: Calculate according to formula (XVI) Project it according to formula (XVIII) and according to Find one Update φ by formula (XIX). For passive RIS, project φ according to formula (XX) to satisfy the constraint;

[0178] Repeat steps 2) and 3) until convergence.

[0179] The output results are the channel precoding matrix V and the RIS phase shift vector φ obtained by solution.

[0180] Figure 2 This is a comparison chart of the classification accuracy of the communication system of the present invention with passive RIS assistance and without RIS assistance under different maximum a posteriori probability classifiers (MAP); Figure 3 This is a comparison chart of the classification accuracy of the communication system of the present invention with passive RIS assistance and without RIS assistance using different K-nearest neighbor classifiers (KNN); Figure 4 This is a comparison chart of the classification accuracy of the communication system of the present invention with passive RIS assistance and without RIS assistance using different neural network classifiers (NN);

[0181] Figure 5 This is a comparison chart of the classification accuracy of the K-nearest neighbor classifier (KNN) assisted by active RIS in the present invention; Figure 6 This is a comparison chart of the classification accuracy of the K-nearest neighbor classifier (MAP) assisted by active RIS in the present invention; Figure 7 This is a comparison chart of the classification accuracy of the K nearest neighbor classifier (NN) assisted by active RIS in the present invention;

[0182] Figures 2 to 7 In the figure, the horizontal axis is the signal-to-noise ratio (SNR), and the vertical axis is the classification accuracy of the receiving end (including the target receiver Bob and the eavesdropper Eve).

[0183] Figures 2 to 4 The classification accuracy of Bob and Eve using the optimization method proposed in this invention is described, including a comparison between a communication system equipped with RIS and a communication system without RIS. The results show that in the system with RIS, Bob achieves higher classification accuracy with all three classifiers, while ensuring that Eve's classification accuracy is consistent with that in the system without RIS (close to 10%, equivalent to the result of random classification). This proves that the method of this invention can improve the accuracy of the target task while ensuring communication security. Figures 2 to 4 The classification accuracy of Bob and Eve in systems equipped with active and passive RIS is compared. The results show that in the system equipped with active RIS, Bob can achieve a higher classification accuracy, while Eve's accuracy remains at around 10%, indicating that the present invention is adaptable to both active and passive RIS designs.

[0184] Figure 8 This is a relationship diagram between the weight coefficient β and the classification accuracy and communication security protection under the MAP classifier and SNR=-10dB in the method of the present invention. Figure 8In the figure, the horizontal axis is the weight coefficient β, and the vertical axis is the classification accuracy. Figure 8 The classification accuracy of Bob and Eve is described under different weight coefficients β. The larger the weight coefficient β (closer to 1), the more the system tends to improve the quality of communication security, thereby sacrificing the performance of the communication task (manifested as a decrease in Bob's classification accuracy). Conversely, the smaller β (closer to 0), the more the system tends to improve the classification accuracy of the target receiver, thereby sacrificing communication security (manifested as an increase in Eve's classification accuracy).

[0185] Example 3

[0186] A computer device includes a memory and a processor. The memory stores a computer program. When the processor executes the computer program, the steps of the RIS-assisted secure communication method described in embodiment 1 or 2 are implemented, which takes into account both downstream tasks and communication security.

[0187] Example 4

[0188] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the secure communication method assisted by RIS, which takes into account both downstream tasks and communication security as described in embodiment 1 or 2.

[0189] Example 5

[0190] A secure communication system assisted by RIS that takes into account both downstream missions and communication security, including:

[0191] The sending module is configured as follows: the sending end first applies a feature encoder to the input image to extract a feature vector, and then uses a channel precoding matrix to convert the feature vector into a signal form that can be transmitted by the channel;

[0192] The transmission module is configured as follows: N RIS units are embedded on the surrounding building surfaces to reflect, amplitude modulate, and phase shift the transmitted signals;

[0193] The classification module is configured as follows: the target signal receiver is equipped with multiple receiving antennas. After receiving the signal reflected by N RIS units, it integrates the signal into the feature vector of the image, and then calculates the category through the classifier to complete the classification task.

Claims

1. A secure communication method assisted by RIS that takes into account both downstream tasks and communication security. The feature is that the transmitter and the target signal receiver complete a communication task of image classification with the help of N RIS units. Specifically, it includes: Transmission phase: The transmitter first applies a feature encoder to the input image to extract feature vectors, and then uses the channel precoding matrix to convert the feature vectors into a signal form that can be transmitted by the channel; Transmission phase: N RIS units reflect, amplitude modulate, and phase shift the transmitted signal; Classification stage: The target signal receiver is equipped with multiple receiving antennas. After receiving the signal reflected by N RIS units, it integrates the signal into a feature vector of the image and then calculates the category through a classifier, thus completing the classification task.

2. A secure communication method assisted by RIS that takes into account both downstream tasks and communication security according to claim 1, characterized in that: The feature vector Adjust to plural form In formula (I), represents the imaginary unit, assuming D = 2F, where F is half the dimension of the eigenvector s, [s] 1:F Refers to the first half of s, [s] F+1:D It's the second half.

3. The secure communication method assisted by RIS that takes into account both downstream tasks and communication security according to claim 1, characterized in that: Define the channel precoder at the transmitter The output signal is N t is the number of transmitting antennas at the transmitter, x satisfies the transmit power constraint shown in formula (II): In formula (II), E(·) represents the mean operator, ‖·‖2 represents the two-norm operator, express The covariance matrix, P bs represents the maximum signal transmission power of the transmitter. The superscript H represents the conjugate transpose of the matrix. tr() refers to the trace operator of the matrix, which calculates the sum of the diagonal elements of the matrix.

4. The secure communication method assisted by RIS that takes into account both downstream tasks and communication security according to claim 1, characterized in that: Once a wireless communication channel is established, the target signal receiver and the eavesdropper's receiving communication signal transmission process can be expressed as formula (III): In formula (III), y B and y E They represent the eigenvectors of Bob and Eve respectively fused according to the reflected signals they received; represents Bob’s transmission channel matrix, where represents the linear channel matrix between Alice and Bob, represents the reflection channel matrix between RIS and Bob, represents the transmission channel matrix between Alice and RIS; represents Eve’s transmission channel matrix, where represents the linear channel matrix between Alice and Eve, represents the reflection channel matrix between RIS and Eve; Represents the RIS phase shift vector is a diagonal matrix, And θ i ∈(0,2π] represents the phase shift caused by the i-th RIS unit; for passive RIS, the amplification factor ρ i =1, while for active RIS, the amplification factor satisfies 1<ρ i <ρ max ; N r is the number of receiving antennas of Bob and Eve; represents the Gaussian noise vector, where δ represents the noise power, I represents the identity matrix, Alice represents the sender, Bob represents the target signal receiver, and Eve represents the eavesdropper; Further preferably, the wireless communication channel is a Rayleigh channel.

5. The secure communication method assisted by RIS that takes into account both downstream tasks and communication security according to claim 1, characterized in that: When the RIS is a passive RIS, the phase shift vector of the RIS unit satisfies the constraint condition of formula (IV): When the RIS is an active RIS, it is subject to the power constraint of the RIS, which can be expressed as: Among them, δ r is the noise power added by the active RIS, represents the matrix binorm, P ris is the maximum power of the active RIS.

6. A secure communication method assisted by RIS according to any one of claims 1 to 5, wherein: Objective function MCR of pre-trained feature encoder 2 , expressed as: In formula (VI), for passive RIS, For active RIS, in represents the signal-to-noise ratio, ε is the lossy decoding rate; and Represents the covariance matrix of the eigenvector and the covariance matrix of the eigenvector of category j, p j is the prior probability of the feature vector received by the edge server and related to category j; Further preferably, the optimization problem STOR of the channel precoder and the passive / active RIS is established as formula (VII): In formula (VII), Represents the vector form of the channel precoding matrix, represents the Kronecker product, T represents the matrix transpose; the coefficient β is used to mediate the weight between the communication task and security.

7. A secure communication method assisted by RIS that takes into account both downstream tasks and communication security according to claim 6, characterized in that: The multivariable optimization problem in formula (VII) is transformed into an alternating optimization problem with single variable constraints, specifically including: 1) Fix the RIS phase shift vector φ and transform Equation (VII) into Equation (VIII): max V G(V)=(1-β)Ψ B (V)-βΛ E (V)(VIII) s.t.v H Mv≤P bs 2) Fix Alice’s channel precoding matrix V and transform Equation (VII) into Equation (IX) for passive / active RIS: Further preferably, solving the optimal channel precoding matrix V using a BGP algorithm includes: 3) Calculate the gradient of the objective function of formula (VIII) with respect to V: in, in, 4) When the transmission power of the transmitting end Alice satisfies v H Mv=P bs When , the maximum value of G(v) is obtained, so The projection is: In formula (XII), 5) Update v by formula (XIII): In formula (XIII), By finding a θ∈[0,π / 2) that satisfies θ=argmaxG(V), the optimal v and V are obtained based on formula (XII); Repeat the above optimization process, i.e., step 3) to step 5) until convergence, and obtain the optimal precoding matrix V for a fixed RIS phase shift vector φ; Further preferably, the RGP algorithm is used to solve the RIS optimization problem, including: For passive RIS, the constraint in Equation (IX) is transformed into a convex constraint: tr(φφ H )=N and||φ|| ∞ ≤1(XIV); 6) Use Approximate infinity norm l ∞ ; Using the potential barrier method, using the logarithmic potential barrier function Integrate the non-negative constraints to approximate the penalty for violating the constraints, where t is a constant used to adjust the penalty; the objective function in Equation (IX) can be rewritten as: max φ Y(φ,h)=G(φ)+I(1-‖φ‖ h (XV); 7) Calculate the gradient of the objective function of formula (XV) with respect to φ: In formula (XV), at the same time in 8) Project the search direction to formula (XVIII): Among them, <,> represents the inner product, by finding a satisfy And update φ using formula (XIX): 9) Use equation (XX) to project φ to satisfy the constraints: Repeat the above optimization process, i.e., step 6) to step 10) until convergence, and obtain the optimal RIS phase shift vector φ that meets the constraints under the given precoding matrix V; For active RIS, rewrite the constraint as: Formula (XXI) is a non-convex constraint, and further introduces the penalty function T(u) = λ(e u -1) to convert the constraint term into a penalty term to punish the violation of the constraint condition, where λ is used to adjust the penalty value; and the objective function is expressed as: max φ Y(φ)=Ψ(φ)-T(g(φ))=Ψ(φ)-λ(e g(φ) -1)(XXII); g(φ) is used as parameter u to pass into the penalty function T(u) for calculation.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the secure communication method assisted by RIS, which takes into account both downstream tasks and communication security, are implemented as described in any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the secure communication method assisted by RIS, which takes into account both downstream tasks and communication security, are implemented as described in any one of claims 1 to 7.

10. A secure communication system assisted by RIS that takes into account both downstream tasks and communication security, characterized in that: include: The sending module is configured as follows: the sending end first applies a feature encoder to the input image to extract a feature vector, and then uses a channel precoding matrix to convert the feature vector into a signal form that can be transmitted by the channel; The transmission module is configured as follows: N RIS units perform reflection, amplitude modulation, and phase shift operations on the transmitted signal; The classification module is configured as follows: the target signal receiver is equipped with multiple receiving antennas. After receiving the signal reflected by N RIS units, it integrates the signal into the feature vector of the image, and then calculates the category through the classifier to complete the classification task.

Images