A prediction-action based isac active safety transmission method

Abstract

The application belongs to the technical field of wireless communication security and signal processing, and discloses an ISAC active security transmission method based on prediction-action, which comprises the following steps: 1, establishing a multi-user ISAC system; 2, establishing a tracking model; 3, formulating a key performance index, and evaluating the multi-user ISAC system constructed in step 1 from the aspects of communication, sensing and double-layer security; 4, reconstructing an Eve eavesdropping channel based on the predicted state and geometric relationship in step 2; 5, jointly designing a data beamformer and an AN beamformer; 6, receiving a new echo signal from the direction of an unknown attacker Eve, updating the parameters of the constructed tracking model and entering the next round of iteration. The application not only improves the sensing accuracy and double-layer communication security of the ISAC system, but also effectively deals with eavesdropping and transmission detection attacks, and has important practical significance.

Description

Technical Field

[0001] This application belongs to the field of wireless communication security and signal processing technology, specifically relating to an ISAC proactive secure transmission method based on prediction-action. Background Technology

[0002] Integrated Sensing and Communication (ISAC) has recently emerged as a key enabling technology for future sixth-generation (6G) wireless networks. By integrating sensing and communication functions onto a unified platform with a shared waveform, ISAC can significantly improve spectral efficiency and reduce hardware costs. Therefore, it holds promise for supporting visionary applications with high-capacity communication and high-precision sensing requirements, such as virtual reality, vehicle-to-everything (V2X) and digital twins. However, the dual-purpose nature of ISAC—embedding confidential information into broadcast signals used for sensing—provides potential adversaries with excellent opportunities for information interception and sensitive location detection. Therefore, designing a robust security framework to ensure the secure operation of relevant ISAC systems is crucial.

[0003] Currently, Physical Layer Security (PLS) has attracted considerable attention as a promising paradigm due to its significant advantages, such as reduced computational burden and lower time overhead, and can complement technologies based on upper-layer cryptography. The principle of PLS ​​is to utilize the inherent characteristics of wireless channels to design secure transmission strategies that prevent eavesdropping. Recently, some preliminary research has focused on deploying PLS in ISAC systems. Although some work has made significant contributions to ensuring the security of ISAC systems, some challenges remain unresolved.

[0004] Most existing works are based on static eavesdroppers with fully known eavesdropping channels or employ reactive perception-then-action paradigms. The inherent decision delays in the reaction loop mean that the final security strategy is optimized for outdated adversary states, making this one-step-back defense mismatched with the real-time dynamics of highly mobile adversaries. Previous work has often focused on single security guarantees, thus lacking a comprehensive analysis of the performance trade-offs of secure ISACs. The fundamental interactions between two-layer security, sensing accuracy, and communication quality in ISAC systems have not been investigated. The performance of existing engineering projects has been validated primarily through simulations based on idealized channel models. This leaves a significant gap between theoretical analysis and practical feasibility, indicating the need for validation with real-world channel data to demonstrate the deployment potential of any proposed security framework. Therefore, ensuring the physical layer security of ISAC systems is a crucial direction for current research. Summary of the Invention

[0005] To address the aforementioned technical issues, this application provides a prediction-action-based proactive secure transmission method for ISAC. This method overcomes existing technical bottlenecks, designs a secure transmission strategy for the ISAC system, maximizes collaborative awareness performance, and simultaneously ensures two-layer communication security for all legitimate users against the mobile adversary, namely confidentiality and covertness.

[0006] To achieve the above objectives, this application employs the following technical solution:

[0007] This application discloses a prediction-action-based ISAC proactive secure transmission method, which is implemented through the ISActive framework. The ISAC proactive secure transmission method specifically includes the following steps:

[0008] Step 1: ISAC signal design. Establish a multi-user ISAC system, which includes a base station (BS), users (User), friendly passive sensing receivers (SR), and a decision fusion center (DFC). The base station (BS) uses the ISAC waveform to track and interfere with potential unknown attackers (Eve). The base station (BS) receives echo signals from unknown attackers (Eve) to achieve secure beamforming.

[0009] Step 2: Establish a tracking model, which includes a state transition model and a measurement model. To establish the state transition model, use the Extended Kalman Filter (EKF) to calculate the predicted state of Eve based on the posterior state estimate of the previous time step. To establish the measurement model, obtain the current measurement value through matched filtering, substitute the current measurement value into the EKF update rule, and calculate the posterior state estimate and posterior error covariance matrix at the current time step.

[0010] Step 3: Develop key performance indicators to evaluate the multi-user ISAC system built in Step 1 from the perspectives of communication performance, perception performance, and two-layer security performance.

[0011] Step 4: Reconstruct the Eve eavesdropping channel based on the predicted state and geometric relationship of Eve calculated in Step 2. Model the real eavesdropping channel as the sum of the norm of the predicted eavesdropping channel and the prediction error, providing prior knowledge for subsequent secure transmission design.

[0012] Step 5: Jointly design a data beamformer and an AN beamformer to suppress eavesdropping while ensuring legitimate communication, thereby obtaining the optimal beamformer and the optimal AN beamformer.

[0013] Step 6: Receive new echo signals from the direction of the unknown attacker Eve, update the parameters of the tracking model constructed in Step 2, and proceed to the next iteration.

[0014] A further improvement of this application is that, in step 1, the multi-user ISAC system includes a device equipped with Base station BS with root antenna The index is Single-antenna legitimate user User The index is of Antenna-friendly and passive sensing receiver (SR) and a decision fusion center (DFC), base station (BS) simultaneously send... Each user transmits K confidential data streams, i.e., composite signals, and actively senses the Region of Interest (RoI). Simultaneously, echo signals from targets within the RoI are captured by M distributed friendly passive sensing receivers (SRs). Then, all signals from base stations (BSs) and... The echo information collected by the distributed friendly passive sensing receiver (SR) is aggregated at the decision fusion center (DFC). The DFC performs data fusion to make a final detection decision. The base station (BS) uses the ISAC waveform to track and interfere with the potential unknown attacker, Eve, to achieve secure beamforming. Eve is an experienced adversary with two attack strategies: 1. Eavesdropping: Eve's primary goal is to compromise data confidentiality by intercepting and decoding the information stream; 2. Adversary detection: Eve also attempts to detect the base station (BS) and... Continuous transmission between users to infer propagation geometry includes the following steps:

[0015] Step 1.1: Establish a communication signal model in a multi-user ISAC system, for each time slot. Depend on It consists of symbols, at each symbol time. The composite signal transmitted by the base station (BS) consists of a linear superposition of K confidential data streams and an AN waveform. for:

[0016]

[0017] in, User The confidential information symbols are the symbols of the K confidential data streams in the composite signal. This is the symbol for dual-purpose artificial noise (AN). AN simultaneously weakens the decoding capabilities of the unknown attacker Eve and acts as a probe signal to sense the unknown attacker Eve. It is a data beamformer. It is an AN beamformer;

[0018] Assuming each user's data beamformer With AN beamformer In the time slot The internal structure is constant and adaptively designed for subsequent time slots based on predicted eavesdropping channel states, transmitting composite signals. covariance matrix Represented as:

[0019]

[0020] in, For the conjugate transpose of the data beamformer, This is the conjugate transpose of the AN beamformer. It is the conjugate transpose;

[0021] Step 1.2: Establish a channel model from base station BS to the nth time slot. The channel vectors of the friendly passive sensing receivers (SRs) are:

[0022]

[0023] in, This indicates Eve, a potential unknown attacker. , Indicates the large-scale path loss gain. It is distance Reference channel gain at that location, It is the base station BS and the first The distance between the three friendly passive sensing receivers (SRs) is α, where α is the path loss exponent. Indicates the emission angle AoD Corresponding transmit array steering vector: ,in, This indicates the phase difference between adjacent antennas;

[0024] Step 1.3: Establish the received signal model. A friendly passive sensing receiver SR at any time The received signal is:

[0025]

[0026] in, It is complex Gaussian noise. Channel vector The conjugate transpose of;

[0027] Step 1.4: Establish a sensing signal model for cooperative echo reception in target detection. To detect targets in the region of interest (RoI), the multi-user ISAC system utilizes echoes from the monostable link (BS-Target-BS) and the M-bistable link (BS-Target-SR) for cooperative sensing. For any receiving node... Where m=0 represents BS, SR represents the echo signal vector. The model is as follows:

[0028]

[0029] in It is the complex reflection coefficient and , This represents complex Gaussian sensing noise; Indicates the emission angle AoD The conjugate transpose of the corresponding transmit array steering vector, Indicates the angle of arrival AoA The corresponding transmit array steering vector, where yes =1 indicates the corresponding AoA at SR; assuming, for a single-station link pointing to a point target, the AoA at BS is... With the emission angle AoD same;

[0030] Step 1.5: Define the echo signal collected at time t, let... By superimposing the echo signals from all M+1 friendly passive sensing receivers (SRs), The echo signal collected at time t is:

[0031] ,

[0032] in, It is a block diagonal matrix, vector This represents the collected complex gain vector. This represents the collected noise vector;

[0033] Decision Fusion Center (DFC) in time slots In Collect echo signals at each time point, and let The entire time slot The collected echo signals are compressed as follows:

[0034]

[0035] in, Represents the aggregated noise vector;

[0036] Step 1.6: Define the round-trip channel matrix between base station BS and unknown attacker Eve:

[0037]

[0038] in, For complex gain, The reflection coefficient, Radar cross-section, For dynamic distance; Indicates AoA The corresponding transmission array steering vector, Indicates the angle of AoD or AoA. Indicates AoD The conjugate transpose of the corresponding transmit array steering vector;

[0039] Step 1.7: The echo signal received by BS from the unknown attacker Eve is:

[0040]

[0041] in, In order to combine the delay of propagation, For Doppler frequency shift, The phase rotation caused by the target motion This is complex Gaussian sensing noise.

[0042] A further improvement in this application is that step 2 specifically includes the following steps:

[0043] Step 2.1: Establish the state model: Define the unknown attacker Eve in the state... The actual state in each time slot is:

[0044]

[0045] in, For location, For speed, The representative moment is The x-coordinate of the attacker Eve is currently unknown. The representative moment is The attacker Eve's ordinate position is currently unknown. The representative moment is The attacker Eve's horizontal velocity is currently unknown. The representative moment is The attacker Eve's vertical velocity is currently unknown.

[0046] From time slot Time slot The state transition model is represented as:

[0047]

[0048] in It is the state transition matrix. Indicates the duration of a time slot. It is state evolution noise. Let be the covariance matrix of the state evolution noise;

[0049] Step 2.2, the prediction phase of EKF utilizes the previous time slot. The posterior state estimate is used to calculate the current time slot. The prior estimate, therefore, Eve's predicted state and its corresponding prediction error covariance matrix They are respectively:

[0050] ,

[0051] in, Indicates time slot The posterior error covariance matrix in This is the state transition matrix;

[0052] Step 2.3: Establish a measurement model; the base station starts from the time slot. echo signal To obtain new measurements, a matched filter MF is applied to utilize the known AN waveform. Obtain an estimate time slot The output signal-to-noise ratio (SNR) of the matched filter MF is written as:

[0053]

[0054] in, Indicates MF gain. It is distance Reference channel gain at that location, Indicates the emission angle AoD The corresponding transmit array steering vector conjugate transpose, This represents the receiver's total noise power. Represents the fourth power of the dynamic distance;

[0055] Modeling the relationship between the true state and the measurement vector in steps 2.4 and 2.1: True State and measurement vector The relationship between them is modeled as follows:

[0056]

[0057]

[0058]

[0059] in, Represents the speed of light. Indicates the carrier frequency. Indicates the target angle parameter sine value Measurement error, Indicates Doppler frequency shift Measurement error, Indicates combined propagation delay Measurement error, , and Modeled as a zero-mean Gaussian random variable with variances of , and , For location, For base station (BS) coordinates, The horizontal coordinate represents the fixed position of the base station (BS). This represents the fixed vertical coordinate position of the base station (BS).

[0060] Step 2.5: Update the measurement model and define a nonlinear function. To encapsulate , and The new measurement model is written as:

[0061]

[0062] in, It is a zero-mean Gaussian measurement noise vector. Let covariance matrix be the variance matrix. , and Modeled as , , , All values ​​are constants and depend on the specific estimation method; a new measurement model is used. Predicted state for Eve Make corrections, due to the measurement function It is non-linear, a non-linear function. First, by predicting the state around Eve. Linearizing the first-order Taylor expansion, we get:

[0063]

[0064] in, yes relative to Eve's predicted state Surrounding The Jacobian matrix, using a linearized model, is obtained through Kalman gain. Posterior state estimation and posterior error covariance Three consecutive calculations to update the state

[0065]

[0066]

[0067]

[0068] in, This represents a 4×4 identity matrix.

[0069] A further improvement in this application is that step 3 specifically includes the following steps:

[0070] Step 3.1, Communication Performance Modeling: Communication performance is evaluated using the Quality of Service (QoS) for each legitimate user: The Signal-to-Noise Ratio (SINR) of the kth legitimate user is expressed as:

[0071]

[0072] in, In time slot The conjugate matrix of the channel vector from BS to user k receiver. For users Data beamformer, For AN beamformer, For users Data beamformer;

[0073] Step 3.2, Perception Performance Modeling: The Decision Fusion Center (DFC) processes the aggregated data across the entire time slot. Collected echo signals To make a testing decision, this is represented as a binary hypothesis test:

[0074]

[0075] in, This indicates that there is no target in the RoI. This indicates that the target exists;

[0076] We employ the generalized likelihood ratio test (GLRT) framework to establish a detector with hypothesis testing capabilities: Let... The decision threshold is represented by the GLRT detector. :

[0077]

[0078] in, and They are and Next moment Collected echo signals The probability density function (PDF);

[0079] Analysis of GLRT detectors The detection performance, characterizing the test statistic. The distribution exist and The distribution converges to different forms of non-central chi-square distributions:

[0080]

[0081] in, This indicates parameters with degrees of freedom (DoF) and non-centrality. Non-central chi-square distribution, non-centrality parameter Represented as:

[0082]

[0083] in, Represents the trace of a matrix. , The average power of the target reflected signal; given a threshold Collaborative sensing solution The detection probability assessment is as follows: ,in It has 2 degrees of freedom and the non-central parameter is The right-tailed probability function of the non-central chi-square variable;

[0084] Step 3.3, Two-Layer Security Performance Modeling: To prevent information interception and exposure of sensitive locations, a two-layer security model was established. This model is quantified by two performance indicators: Eve's information leakage rate, ensuring data confidentiality, and Eve's detection probability, ensuring communication concealment. Specifically, it includes the following steps:

[0085] Step 3.3.1, Eve's information leakage rate: Eve's worst-case information leakage rate per user is:

[0086]

[0087] in, In time slot The conjugate matrix of the channel vector from BS to Eve receiver. This represents the noise power received by the attacker.

[0088] Step 3.3.2, Eve's detection probability: The communication stealth is evaluated by using Eve's detection probability performance. Similarly, The formula for the binary hypothesis test is:

[0089]

[0090] in, For the friendly passive sensing receiver SR at time Received signal, null hypothesis Instruct the base station (BS) to remain silent, and exist The following distribution is in Then, conversely, the alternative hypothesis This indicates that there is communication activity between the base station (BS) and the legitimate user, therefore ;

[0091] For the sake of simplicity of symbols, define Suppose Eve uses the Neyman-Pearson criterion to design a function with a preset threshold. The optimal LRT detector:

[0092] in, ,

[0093] Utilizing the established optimal LRT detector To model the detection probability of Eve and quantify the stealth of communication, the following steps are taken:

[0094] Given a preset threshold Eve detection probability The assessment is as follows:

[0095] in, .

[0096] A further improvement in this application is that step 4 specifically includes the following steps:

[0097] Step 4.1, Prediction Phase: Eavesdropping Channel Reconstruction, first using the previous Eve prediction state. To construct the first A time-slot eavesdropping channel based on Eve's predicted state and BS coordinates. Predicted distance and launch angle :

[0098]

[0099]

[0100] Then the predicted eavesdropping channel Reconstructed as:

[0101]

[0102] Due to unavoidable noise interference, the actual eavesdropping channel will be... Modeling as a predictive eavesdropping channel and the norm of prediction error The sum:

[0103]

[0104] Wherein, the norm of the prediction error is From the known constant Definition, that is , This is a known upper bound for the channel prediction error energy.

[0105] A further improvement in this application is that step 5 specifically includes the following steps:

[0106] Step 5.1: To maximize the accuracy of collaborative perception To simultaneously meet the Quality of Service (QoS) requirements of legitimate users, eavesdropper tracking accuracy, and dual-layer security under channel prediction uncertainty, all performance metrics are omitted for the sake of symbol simplicity. The explicit dependency means that the optimization problem is formulated as follows:

[0107]

[0108] in, Minimum Quality of Service (QoS) for each legitimate user. This represents the probability of Eve being detected. This serves as a unified upper bound for the probability of Eve's detection. Eve's information leakage rate; This serves as a unified upper bound for Eve's information leakage rate; This represents the total transmission power.

[0109] Step 5.2: A three-step optimization algorithm was developed to solve the problem. This includes: problem restatement, beamforming design and randomization based on semi-definite relaxation (SDR): where,

[0110] Problem restatement: To improve the tractability of optimization problems, we first need to improve the accuracy of collaborative perception. Performing an equivalent transformation, for any fixed threshold and degrees of freedom (DoF), the function... Is it a non-centrality parameter? It is a monotonically increasing function; therefore, maximizing the accuracy of collaborative sensing... Equivalent to maximizing the noncentrality parameter The noncentrality parameter is then simplified to maximizing the quadratic term. ;

[0111] Step 5.2, Solve non-convex constraints: Let , The Quality of Service (QoS) constraints in the code are rewritten as follows:

[0112]

[0113] Step 5.3, the concealment constraint is And rewrite the concealment constraint as follows:

[0114] ;

[0115] The original implicit constraint is equivalently transformed into:

[0116] ,in It is to satisfy The value;

[0117] Step 5.4: For the sake of simplicity, the index has been omitted. The original hidden constraint equation is restated as follows:

[0118]

[0119] Step 5.5, similarly, combining Bounded prediction error constraints Rephrased as:

[0120]

[0121] Step 5.6: Apply the original concealment constraints Convert to a linear matrix inequality (LMI):

[0122]

[0123] in, ;

[0124] Similarly, the constraint on bounded prediction error is expressed as:

[0125]

[0126] in, ,and ;

[0127] Original tracking accuracy constraints It is transformed into the following set of linear matrix inequalities (LMIs):

[0128] .

[0129] in, It is a selection vector where the l-th position is 1 and the other positions are 0;

[0130] Based on the above transformation, the problem The equivalent restatement is as follows:

[0131]

[0132] Step 5.7, Beamforming Design Based on Semi-Deterministic Relaxation (SDR): Using SDR technology to address the problem... The relaxation is a semi-deterministic relaxation (SDR) beam, specifically:

[0133] Define the following matrix: and

[0134] constraint and So, the question is... Equivalent to

[0135]

[0136] in ,

[0137]

[0138] Will Convert to a convex semidefinite program (SDP);

[0139] Step 5.8, Randomization: The solution obtained from the convex semidefinite procedure (SDP). It cannot be guaranteed to be rank one if the obtained matrix If the rank is 1, then the optimal beamformer can be directly recovered through eigenvalue decomposition. and the best AN beamformer Otherwise, a standard Gaussian randomization process is used to generate a high-quality rank-one solution.

[0140] A further improvement in this application is that step 6 specifically includes the following steps:

[0141] Step 6.1: Obtain the optimal beamformer and the best AN beamformer Then, the base station BS transmits the actively designed ISAC signal at the beginning of time slot n;

[0142] Step 6.2: Capture the echo signal reflected by Eve, the unknown attacker. The measurement vector for time slot n is generated through MF processing. ;

[0143] Step 6.3: Utilize this measurement vector Obtained according to EKF update rules and and will and As a subsequent time slot The key inputs in the prediction phase complete the proactive safety cycle.

[0144] The beneficial effects of this application are:

[0145] This application proposes an active security framework that integrates an extended Kalman filter (EKF) with a dual-purpose active beamforming (AN) strategy. This integration enables the prediction of the future state of a mobile adversary, facilitating active beamforming design to protect subsequent transmissions from data and location privacy breaches.

[0146] This application provides a theoretical model for two-layer security from the perspective of confidentiality and concealment, realizing two-layer security modeling. These models enable the establishment of a clear relationship between beamforming design and two-layer security performance.

[0147] This application proposes an efficient algorithm based on semidefinite relaxation (SDR) to solve the non-convex optimization problem.

[0148] This application is based on channel measurement verification. Simulation results based on real-world channel measurements demonstrate that, compared to existing work, ISActive's resource-efficient secure transmission can achieve more refined cooperative sensing performance, and also proves its practical feasibility in vehicle-to-infrastructure (V2I) scenarios. Attached Figure Description

[0149] Figure 1 This is an example diagram of the security ISAC scenario in this application.

[0150] Figure 2 This is a schematic diagram of the ISActive framework of this application.

[0151] Figure 3 It is a simulation topology diagram.

[0152] Figure 4 This is Eve's tracking performance graph.

[0153] Figure 5 It is the average tracking normalized mean square error and the transmit power budget P max The trade-off diagram.

[0154] Figure 6 It is the collaborative awareness performance P in actual V2I scenarios. D and tolerable leakage rate R E Performance comparison chart between them.

[0155] Figure 7 It is the collaborative awareness performance P in actual V2I scenarios. D And Eve P D,E Performance comparison chart of tolerable detection probabilities at each location.

[0156] Figure 8 In real-world V2I scenarios, QoS requirements Rmin for P D A diagram illustrating the impact.

[0157] Figure 9 In a real-world V2I scenario, the transmit power budget P max For P D A diagram illustrating the impact.

[0158] Figure 10 This is a flowchart illustrating the ISAC active secure transmission method of this application. Detailed Implementation

[0159] The embodiments of the present invention will be disclosed below with reference to the drawings. For clarity, many practical details will be described in the following description. However, it should be understood that these practical details are not intended to limit the present invention. That is, in some embodiments of the present invention, these practical details are not essential. In addition, for the sake of simplicity, some conventional structures and components will be shown in the drawings in a simple schematic manner.

[0160] This application presents a prediction-action-based ISAC proactive secure transmission method, implemented using the ISActive framework, such as... Figure 2 As shown, the ISActive framework includes an input phase, a prediction phase, an action phase, and a perception and update phase, wherein...

[0161] During the input phase, the predicted state of Eve in time slot n-1. That is, the posterior state estimate and the posterior error covariance matrix. As input;

[0162] In the prediction phase, the predicted state of Eve is used as input. That is, the posterior state estimate and the posterior error covariance matrix are used to predict the current time slot through an extended Kalman filter (EKF). Predicted distance and launch angle Based on the predicted target distance and angle, reconstruct the eavesdropping channel between BS and Eve;

[0163] During the operation phase, active beamforming is generated based on the reconstructed eavesdropping channel, and the optimal data beamformer and the optimal AN beamformer are generated to achieve secure ISAC transmission.

[0164] During the sensing and updating phase, the BS receives the echo signal from Eve, performs sensing measurements through a matched filter to obtain the measurement value of the current time slot, and substitutes the measurement value into the Extended Kalman Filter (EKF) update rule to calculate the current time slot. Eve's predicted state That is, the posterior state estimate and the posterior error covariance matrix. This will be used as the input for the subsequent step 1.

[0165] like Figure 10 As shown, the ISAC active secure transmission method of this application specifically includes the following steps:

[0166] Step 1: ISAC signal design. A multi-user ISAC system is established, comprising a base station (BS), users (Users), a friendly passive sensing receiver (SR), and a decision fusion center (DFC). The base station (BS) uses the ISAC waveform to track and interfere with a potential unknown attacker (Eve). The base station (BS) receives the echo signal from the unknown attacker (Eve), thus achieving secure beamforming. In Step 1, the multi-user ISAC system includes a system equipped with… Base station BS with root antenna The index is Single-antenna legitimate user User The index is of Antenna-friendly and passive sensing receiver (SR) and a decision fusion center (DFC), base station (BS) simultaneously send... Each user transmits K confidential data streams, i.e., composite signals, and actively senses the Region of Interest (RoI). Simultaneously, echo signals from targets within the RoI are captured by M distributed friendly passive sensing receivers (SRs). Then, all signals from base stations (BSs) and... The echo information collected by the distributed friendly passive sensing receiver (SR) is aggregated at the decision fusion center (DFC). The DFC performs data fusion to make a final detection decision. The base station (BS) uses the ISAC waveform to track and interfere with the potential unknown attacker Eve, achieving secure beamforming. Eve is an experienced adversary with two attack strategies: 1. Eavesdropping: Eve's primary goal is to compromise data confidentiality by intercepting and decoding the information stream; 2. Adversary detection: Eve also attempts to detect the base station (BS) and... Continuous transmission between users to infer propagation geometry includes the following steps:

[0167] Step 1.1: Establish a communication signal model in a multi-user ISAC system, for each time slot. Depend on It consists of symbols, at each symbol time. The composite signal transmitted by the base station (BS) consists of a linear superposition of K confidential data streams and an AN waveform. for:

[0168]

[0169] in, User The confidential information symbols are the symbols of the K confidential data streams in the composite signal. This is the symbol for dual-purpose artificial noise (AN). AN simultaneously weakens the decoding capabilities of the unknown attacker Eve and acts as a probe signal to sense the unknown attacker Eve. It is a data beamformer. It is an AN beamformer;

[0170] Assuming each user's data beamformer With AN beamformer In the time slot The internal structure is constant and adaptively designed for subsequent time slots based on predicted eavesdropping channel states, transmitting composite signals. covariance matrix Represented as:

[0171]

[0172] in, For the conjugate transpose of the data beamformer, This is the conjugate transpose of the AN beamformer. It is the conjugate transpose;

[0173] Step 1.2: Establish a channel model, a geometry-based channel model, which has a dominant line-of-sight propagation environment. Then, in time slot n, from base station BS to the... The channel vectors of the friendly passive sensing receivers (SRs) are:

[0174]

[0175] in, This indicates Eve, a potential unknown attacker. , Indicates the large-scale path loss gain. It is distance Reference channel gain at that location, It is the base station BS and the first The distance between the three friendly passive sensing receivers (SRs) is α, where α is the path loss exponent. Indicates the emission angle AoDc Corresponding transmit array steering vector: ,in, This indicates the phase difference between adjacent antennas;

[0176] Step 1.3: Establish the received signal model. A friendly passive sensing receiver SR at any time The received signal is:

[0177]

[0178] in, It is complex Gaussian noise. Channel vector The conjugate transpose of;

[0179] Step 1.4: Establish a sensing signal model for cooperative echo reception in target detection. To detect targets within the region of interest (RoI), the multi-user ISAC system utilizes echoes from both the monostable link (BS-Target-BS) and the M-bistable link (BS-Target-SR) for cooperative sensing. For any receiving node... Where m=0 represents BS, This indicates SR, meaning that when m=0, there is only BS and no SR; SR only occurs when m>0, and the echo signal vector... The model is as follows:

[0180]

[0181] in It is the complex reflection coefficient and , This represents complex Gaussian sensing noise; Indicates the emission angle AoD The conjugate transpose of the corresponding transmit array steering vector, Indicates the angle of arrival AoA The corresponding transmit array steering vector, where yes =1 indicates the corresponding AoA at SR; assuming, for a single-station link pointing to a point target, the AoA at BS is... With the emission angle AoD same;

[0182] Step 1.5: Define the echo signal collected at time t, let... By superimposing the echo signals from all M+1 friendly passive sensing receivers (SRs), The echo signal collected at time t is:

[0183] ,

[0184] in, It is a block diagonal matrix, vector This represents the collected complex gain vector. This represents the collected noise vector;

[0185] Decision Fusion Center (DFC) in time slots In Collect echo signals at each time point, and let The entire time slot The collected echo signals are compressed as follows:

[0186]

[0187] in, Represents the aggregated noise vector;

[0188] Step 1.6: Define the round-trip channel matrix between base station BS and unknown attacker Eve:

[0189]

[0190] in, For complex gain, The reflection coefficient, Radar cross-section, For dynamic distance; Indicates AoA The corresponding transmission array steering vector, Indicates the angle of AoD or AoA. Indicates AoD The conjugate transpose of the corresponding transmit array steering vector;

[0191] Step 1.7: The echo signal received by BS from the unknown attacker Eve is:

[0192]

[0193] in, In order to combine the delay of propagation, For Doppler frequency shift, The phase rotation caused by the target motion This is complex Gaussian sensing noise.

[0194] Step 2: Establish a tracking model, which includes a state transition model and a measurement model. The state transition model is established by using an Extended Kalman Filter (EKF) to calculate the predicted state of Eve based on the posterior state estimate from the previous time step. The measurement model is established by obtaining the current measurement value through matched filtering. The current measurement value is then substituted into the EKF update rule to calculate the posterior state estimate and the posterior error covariance matrix at the current time step. Specifically, this includes the following steps:

[0195] Step 2.1: Establish the state model: Define the unknown attacker Eve in the state... The actual state in each time slot is:

[0196]

[0197] in, For location, For speed, The representative moment is The x-coordinate of the attacker Eve is currently unknown. The representative moment is The attacker Eve's ordinate position is currently unknown. The representative moment is The attacker Eve's horizontal velocity is currently unknown. The representative moment is The attacker Eve's vertical velocity is currently unknown.

[0198] From time slot Time slot The state transition model is represented as:

[0199]

[0200] in It is the state transition matrix. Indicates the duration of a time slot. It is state evolution noise. Let be the covariance matrix of the state evolution noise;

[0201] Step 2.2, the prediction phase of EKF utilizes the previous time slot. The posterior state estimate is used to calculate the current time slot. The prior estimate, therefore, Eve's predicted state and its corresponding prediction error covariance matrix They are respectively:

[0202] ,

[0203] in, Indicates time slot The posterior error covariance matrix in This is the state transition matrix;

[0204] Step 2.3: Establish a measurement model; the base station starts from the time slot. echo signal To obtain new measurements, a matched filter MF is applied to utilize the known AN waveform. Get an estimate time slot The output signal-to-noise ratio (SNR) of the matched filter MF is written as:

[0205]

[0206] in, Indicates MF gain. It is distance Reference channel gain at that location, Indicates the emission angle AoD The corresponding transmit array steering vector conjugate transpose, This represents the receiver's total noise power. Represents the fourth power of the dynamic distance;

[0207] Modeling the relationship between the true state and the measurement vector in steps 2.4 and 2.1: True State and measurement vector The relationship between them is modeled as follows:

[0208]

[0209]

[0210]

[0211] in, Represents the speed of light. Indicates the carrier frequency. Indicates the target angle parameter sine value Measurement error, Indicates Doppler frequency shift Measurement error, Indicates combined propagation delay Measurement error, , and Modeled as a zero-mean Gaussian random variable with variances of , and , For location, For base station (BS) coordinates, The horizontal coordinate represents the fixed position of the base station (BS). This represents the fixed vertical coordinate position of the base station (BS).

[0212] Step 2.5: Update the measurement model and define a nonlinear function. To encapsulate , and The new measurement model is written as:

[0213]

[0214] in, It is a zero-mean Gaussian measurement noise vector. Let covariance matrix be the variance matrix. , and Modeled as , , , All values ​​are constants and depend on the specific estimation method; a new measurement model is used. Predicted state for Eve Make corrections, due to the measurement function It is non-linear, a non-linear function. First, by predicting the state around Eve. Linearizing the first-order Taylor expansion, we get:

[0215]

[0216] in, yes relative to Eve's predicted state Surrounding The Jacobian matrix, using a linearized model, is obtained through Kalman gain. Posterior state estimation and posterior error covariance Three consecutive calculations to update the state

[0217]

[0218]

[0219]

[0220] in, This represents a 4×4 identity matrix.

[0221] Step 3: Define key performance indicators (KPIs) to evaluate the multi-user ISAC system constructed in Step 1, considering communication performance, perception performance, and two-layer security performance. This includes the following steps:

[0222] Step 3.1, Communication Performance Modeling: Communication performance is evaluated through the Quality of Service (QoS) of each legitimate user: The QoS of the k-th user is determined by its achievable rate Rk[n]. Based on the communication signal model, the signal-to-noise ratio (SINR) of the k-th legitimate user is first expressed as:

[0223]

[0224] in, In time slot The conjugate matrix of the channel vector from BS to user k receiver. For users Data beamformer, For AN beamformer, For users Data beamformer;

[0225] Step 3.2, Perception Performance Modeling: The Decision Fusion Center (DFC) processes the aggregated data across the entire time slot. Collected echo signals To make a testing decision, this is represented as a binary hypothesis test:

[0226]

[0227] in, This indicates that there is no target in the RoI. This indicates that the target exists;

[0228] We employ the generalized likelihood ratio test (GLRT) framework to establish a detector with hypothesis testing capabilities: Let... The decision threshold is represented by the GLRT detector. :

[0229]

[0230] in, and They are and Next moment Collected echo signals The probability density function (PDF);

[0231] To design a GLRT detector, the first step is to... Maximum likelihood (ML) estimation of the unknown parameter vector β:

[0232]

[0233] in, The echo signal collected at time t;

[0234] Analysis of GLRT detectors The detection performance, characterizing the test statistic. The distribution, for a sufficient number of observations, exist and The distribution converges to different forms of non-central chi-square distributions:

[0235]

[0236] in, This indicates parameters with degrees of freedom (DoF) and non-centrality. Non-central chi-square distribution, non-centrality parameter Represented as:

[0237]

[0238] in, Represents the trace of a matrix. , The average power of the target reflected signal; given a threshold Collaborative sensing solution The detection probability assessment is as follows: ,in It has 2 degrees of freedom and the non-central parameter is The right-tailed probability function of the non-central chi-square variable;

[0239] The accuracy of tracking attacker Eve's location is quantified by the posterior mean squared error (PMSE), which is defined as the posterior error covariance matrix. traces The posterior mean square error (PMSE) depends on the AN beamforming strategy. Specifically, the posterior error covariance matrix... Depends on Kalman gain And the posterior error covariance matrix Covariance of measurement noise Related. Intravariance and perceived SNR Inversely proportional to the perceived SNR Ultimately, it is produced by the AN beamformer. Sure;

[0240] Step 3.3, Two-Layer Security Performance Modeling: To prevent information interception and exposure of sensitive locations, a two-layer security model was established. This model is quantified by two performance indicators: Eve's information leakage rate and Eve's detection probability. Specifically, it includes the following steps:

[0241] Step 3.3.1, Eve's Information Leakage Rate: To design a robust and secure communication scheme, a powerful Eve is defined as capable of perfectly eliminating interference between users, thus considering the worst-case scenario. Under this definition, the... Eve's worst-case information leakage rate per user is:

[0242]

[0243] in, In time slot The conjugate matrix of the channel vector from BS to Eve receiver. This represents the noise power received by the attacker.

[0244] The purpose of this safety operation is to ensure that any decoding attempt by Eve results in negligible throughput or an unacceptably high bit error rate.

[0245] Step 3.3.2, Eve's Detection Probability: This aims to conceal the existence of the transmission itself, thereby protecting the transmitter's location privacy. Therefore, Eve's detection probability performance is used to evaluate communication concealment. Similarly, The formula for the binary hypothesis test is:

[0246]

[0247] in, For the friendly passive sensing receiver SR at time Received signal, null hypothesis Instruct the base station (BS) to remain silent, and exist The following distribution is in Then, conversely, the alternative hypothesis This indicates that there is communication activity between the base station (BS) and the legitimate user, therefore ;

[0248] For the sake of simplicity of symbols, define Suppose Eve uses the Neyman-Pearson criterion to design a function with a preset threshold. The optimal LRT detector:

[0249] in, ,

[0250] Utilizing the established optimal LRT detector To model the detection probability of Eve and quantify the stealth of communication, the following steps are taken:

[0251] Given a preset threshold Eve detection probability The assessment is as follows:

[0252] in, .

[0253] Step 4: Reconstruct the Eve eavesdropping channel based on the predicted state and geometric relationship of Eve calculated in Step 2. Model the actual eavesdropping channel as the sum of the norms of the predicted eavesdropping channel and the prediction error, providing prior knowledge for subsequent secure transmission design. This specifically includes the following steps:

[0254] Step 4.1, Prediction Phase: Eavesdropping Channel Reconstruction, first using the previous Eve prediction state. To construct the first A time-slot eavesdropping channel based on Eve's predicted state and BS coordinates. Predicted distance and launch angle :

[0255]

[0256]

[0257] Then the predicted eavesdropping channel Reconstructed as:

[0258]

[0259] Due to unavoidable noise interference, the actual eavesdropping channel will be... Modeling as a predictive eavesdropping channel and the norm of prediction error The sum:

[0260]

[0261] Wherein, the norm of the prediction error is From the known constant Definition, that is , This is a known upper bound for the channel prediction error energy.

[0262] Step 5: Jointly design the data beamformer and the AN beamformer to suppress eavesdropping while ensuring legitimate communication, thereby obtaining the optimal beamformer and the optimal AN beamformer. This specifically includes the following steps:

[0263] Step 5.1: To maximize the accuracy of collaborative perception To simultaneously meet the Quality of Service (QoS) requirements of legitimate users, eavesdropper tracking accuracy, and dual-layer security under channel prediction uncertainty, all performance metrics are omitted for the sake of symbol simplicity. The explicit dependency means that the optimization problem is formulated as follows:

[0264] ;

[0265] in, Minimum Quality of Service (QoS) for each legitimate user. This represents the probability of Eve being detected. This serves as a unified upper bound for the probability of Eve's detection. Eve's information leakage rate; This serves as a unified upper bound for Eve's information leakage rate; This represents the total transmission power.

[0266] Step 5.2, Solve the problem This problem presents significant challenges due to its non-convex nature, stemming from three aspects. First, the non-convexity arises from the fractional structure of the rate and SINR expressions within the constraints. Second, the objective function and constraints involve complex nonlinear dependencies on beamforming variables. Third, uncertainties in the prediction channels lead to infinite inequality constraints within the set of limitations. Due to these characteristics, existing convex optimization techniques are difficult to apply to solving this problem. This necessitates the development of an efficient algorithm to find high-quality solutions. To address these challenges, a three-step optimization algorithm was developed to solve the problem. This includes: problem restatement, beamforming design and randomization based on semi-definite relaxation (SDR): where,

[0267] Problem restatement: To improve the tractability of optimization problems, we first need to improve the accuracy of collaborative perception. Performing an equivalent transformation, for any fixed threshold and degrees of freedom (DoF), the function... Is it a non-centrality parameter? It is a monotonically increasing function; therefore, maximizing the accuracy of collaborative sensing... Equivalent to maximizing the noncentrality parameter The noncentrality parameter is then simplified to maximizing the quadratic term. ;

[0268] Step 5.2, Solve non-convex constraints: Let , The Quality of Service (QoS) constraints in the code are rewritten as follows:

[0269]

[0270] Step 5.3, the tricky hidden constraint is And rewrite the concealment constraint as follows:

[0271] ;

[0272] The expression follows the general form. And for any constant a ≥ 1, the function The function is monotonically increasing with respect to x in two feasible regions x∈(0,1) and x∈(1,∞).

[0273] The original implicit constraint is equivalently transformed into:

[0274] ,in It is to satisfy The value;

[0275] Step 5.4: For the sake of simplicity, the index has been omitted. The original hidden constraint equation is restated as follows:

[0276]

[0277] Step 5.5, similarly, combining Bounded prediction error constraints Rephrased as:

[0278]

[0279] Step 5.6: The two security constraints mentioned above must hold for all possible channel realizations within the bounded uncertainty set. This constitutes a semi-infinite number of constraints, making it difficult to handle in its current form. Therefore, this infinite number of constraints is reconstructed into a tractable linear matrix inequality (LMI). Thus, the original implicit constraints can be... Convert to a linear matrix inequality (LMI):

[0280]

[0281] in, ;

[0282] Similarly, the constraint on bounded prediction error is expressed as:

[0283]

[0284] in, ,and ;

[0285] analyze The tracking accuracy constraint is addressed. To make it easier to handle, a set of non-negative auxiliary variables is first introduced. And use them to constrain each diagonal element of the posterior covariance matrix. This enables the equivalent use of linear constraints. and diagonal elements The four individual constraints on the trace are replaced with the single-trace inequalities. Although the constraints are now linear, the diagonal element constraints remain non-convex. Therefore, each of these non-convex inequalities is transformed into a computable LMI using the Schur complement. Thus, the original tracking accuracy constraints are... It is transformed into the following set of linear matrix inequalities (LMIs):

[0286] .

[0287] in, It is a selection vector where the l-th position is 1 and the other positions are 0;

[0288] Based on the above transformation, the problem The equivalent restatement is as follows:

[0289]

[0290] Step 5.7, Beamforming Design Based on Semi-Deterministic Relaxation (SDR): Although the problem It has a more manageable structure, but it remains non-convex due to its secondary dependence on the beamformer. Therefore, SDR technology is used to address the problem. The relaxation is a semi-deterministic relaxation (SDR) beam, specifically:

[0291] Define the following matrix:

[0292] and

[0293] constraint and So, the question is... Equivalent to

[0294]

[0295] in ,

[0296] because The non-convex rank-1 constraint still exists, which is quite tricky. Then, this tricky problem is alleviated by abandoning the first-order constraint, and... Converting to a convex semidefinite program (SDP) allows the problem to be solved efficiently using standard optimization solvers such as the CVX Toolbox.

[0297] Step 5.8, Randomization: The solution obtained from the convex semidefinite procedure (SDP). It cannot be guaranteed to be rank one if the obtained matrix If the rank is 1, then the optimal beamformer can be directly recovered through eigenvalue decomposition. and the best AN beamformer Otherwise, a standard Gaussian randomization process is used to generate a high-quality rank-1 solution. The worst-case complexity of the proposed optimization algorithm is O(n log n). ,in Precision of the solution

[0298] Step 6: Receive new echo signals from the direction of the unknown attacker Eve, update the parameters of the tracking model constructed in Step 2, and proceed to the next iteration. This includes the following steps:

[0299] Step 6.1: Obtain the optimal beamformer and the best AN beamformer Then, the base station BS transmits the actively designed ISAC signal at the beginning of time slot n;

[0300] Step 6.2: Capture the echo signal reflected by Eve, the unknown attacker. The measurement vector for time slot n is generated through MF processing. ;

[0301] Step 6.3: Utilize this measurement vector Obtained according to EKF update rules and and will and As a subsequent time slot The key inputs in the prediction phase complete the proactive safety cycle.

[0302] To demonstrate the practical feasibility of the theoretical analysis, we will verify it using real channel data. The specific steps include:

[0303] Step 6.1: Set system parameters based on the public ISAC DeepSense 6G dataset, which provides real-world channel measurements in a V2ImmWave scenario. The test bench for this scenario includes a static BS transmitter with a 16-element mmWave phased array and a mobile vehicle receiver (Eve) with high-precision GPS-RTK for ground-based real-world positioning. Additionally, three static points along the vehicle's trajectory are selected as communication users. Then, using the provided real-world channel measurements, the real and dynamic channels of the BS-Eve and BS-Users downlink can be instantiated. This scenario is then mapped to... Figure 3The two-dimensional simulation topology is as follows: BS is located at the origin [0,0] m, while Eve moves from the starting position [13,22] m based on GPS data from the dataset. The positions of the three users K are [13,18] m, [14,7] m, and [15,-1] m, respectively. To complete the specific problem setup of this application, two SRs and a stationary target are positioned at [-5,15] m, [-5,5] m, and [5,10] m, respectively. The EKF process is initialized with prior state estimates [10,20] m and a velocity [5,0] m / s, with an initial error covariance of . = diag{80,80,5,5}.

[0304] The key system parameters are set as follows:

[0305] Step 6.2, Baselines. Three baselines are introduced for comparison: Static ISAC (S-ISAC): S-ISAC uses pre-assumed channel knowledge of the static Eve to design secure beamforming strategies without using sensing and tracking capabilities. Reactive ISAC (R-ISAC): R-ISAC uses beam scanning to estimate the Eve's angle and then designs secure beamforming based on this in real-time measurement. Monostatic ISAC (M-ISAC): M-ISAC uses a Long Short-Term Memory (LSTM) network to track the Eve to achieve secure beamforming. Its design is limited to monostatic target detection and does not consider communication stealth requirements.

[0306] Step 6.3: Track performance analysis. Figure 4 The real-world time-tracking performance of the novel ISActive active physical layer safety framework based on the prediction-then-action paradigm was validated using 12.5 seconds of vehicle trajectory data from the dataset and accompanying ground-based data. Results show that the estimated distances and angles are almost perfectly consistent with their true values. The estimated velocity, despite minor deviations, accurately captures the overall trend of the ground-based data.

[0307] Step 6.4: Evaluate the average tracking performance, and further use the normalized MSE (NMSE) metric to measure the tracking performance under different power budgets P. max The average tracking performance was evaluated numerically, and the results were summarized in Figure 5 Given P max At >30 dBm, it can be seen that the NMSE can be achieved below -10 dB for all three conditions. These findings confirm the effectiveness of the designed active beamforming design in providing reliable Eve state prediction.

[0308] Step 6.5: The trade-off between perception and two-layer security. Figure 6This illustrates the collaborative perception performance P D and tolerable leakage rate R E The trade-offs between them, the P of all options D Improvement occurs as confidentiality constraints are relaxed, i.e., R E The increase is due to less stringent confidentiality requirements, allowing more transmit power to be allocated to enhance sensing SNR. This can be seen in ISActive throughout the entire R... E It consistently outperforms all baselines within the range. For example, the proposed method in R... E Under the constraint of 0.5 bps / Hz, it exhibits 83% P D Compared to the S-ISAC, R-ISAC, and M-ISAC baselines, it produces significant sensing performance gains of 36.62%, 16.45%, and 7.16%, respectively. This is attributed to two key advantages of ISActive: (1) the active safety design based on EKF prediction facilitates a more resource-efficient beamforming strategy for primary sensing tasks; and (2) the cooperative sensing architecture provides enhanced detection gains by leveraging spatial diversity.

[0309] Step 6.6 Figure 7 Assess collaborative perceived performance P D And Eve P D,E The trade-off between tolerable detection probabilities at each location. To facilitate fair comparison, the same concealment constraint was used to enhance the baseline scheme. Results show that, with the concealment constraint relaxed, ISActive's P... D Furthermore, it can be observed that for any fixed P, the value can exceed 80. D ISActive achieves significantly lower P D,E That is, better concealment than the baseline indicates its superior ability to hide its communication activities from the opponent.

[0310] from Figure 6 and Figure 7 It can be seen that ISActive demonstrates superior communication security by jointly considering data confidentiality and communication concealment, while also achieving better perception accuracy.

[0311] Step 6.7, Impact of Key System Parameters. Figure 8 The graph shows the QoS requirement Rmin for P. D The impact. As expected, the P values ​​for all scenarios... D It decreases as Rmin increases. This tradeoff occurs because meeting stricter QoS requirements necessitates allocating more power to the data beamformer, reducing the power available for sensing. However, ISActive exhibits greater resilience to this tradeoff. Even with R... minEven with the higher QoS requirement of 3.4 bps / Hz, ISActive maintains approximately 80% of its performance. D The baseline, however, exhibited poor sensing performance (below 70%).

[0312] Step 6.8 Figure 9 This explains the transmit power budget P. max For P D The impact of P. For all scenarios, a larger P... max Producing higher P D This is because it directly increases the energy of the transmitted ISAC signal and enhances the echo SNR. Furthermore, it is clear that ISActive effectively utilizes the increased power budget to achieve optimal sensing accuracy. Specifically, in P... max At 34 dBm, ISActive achieved 92.67% P. D This resulted in significant performance gains of approximately 25%, 7%, and 4% on the S-ISAC, R-ISAC, and M-ISAC baselines, respectively. Figure 8 and Figure 9 It was observed that R min = 2.6 bps / Hz or P max At a voltage of 34 dBm, performance can be improved to approximately 92% of the high P. D This proves its practical feasibility.

[0313] This application not only improves the sensing accuracy and two-layer communication security of the ISAC system, but also effectively counters eavesdropping and transmission detection attacks, which has important practical significance.

[0314] The above description is merely an embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of the present invention should be included within the scope of the claims of the present invention.

Claims

1. A prediction-action based ISAC proactive secure transmission method, implemented through the ISActive framework, characterized in that: The ISAC active secure transmission method specifically includes the following steps: Step 1: ISAC signal design. Establish a multi-user ISAC system, which includes a base station (BS), users (User), friendly passive sensing receivers (SR), and a decision fusion center (DFC). The base station (BS) uses the ISAC waveform to track and interfere with potential unknown attackers (Eve). The base station (BS) receives echo signals from unknown attackers (Eve) to achieve secure beamforming. Step 2: Establish a tracking model, which includes a state transition model and a measurement model. To establish the state transition model, use the Extended Kalman Filter (EKF) to calculate the predicted state of Eve based on the posterior state estimate of the previous time step. To establish the measurement model, obtain the current measurement value through matched filtering, substitute the current measurement value into the EKF update rule, and calculate the posterior state estimate and posterior error covariance matrix at the current time step. Step 3: Develop key performance indicators to evaluate the multi-user ISAC system built in Step 1 from the perspectives of communication performance, perception performance, and two-layer security performance. Step 4: Reconstruct the Eve eavesdropping channel based on the predicted state and geometric relationship of Eve calculated in Step 2. Model the real eavesdropping channel as the sum of the norm of the predicted eavesdropping channel and the prediction error, providing prior knowledge for subsequent secure transmission design. Step 5: Jointly design the data beamformer and the AN beamformer to suppress eavesdropping performance while ensuring legitimate communication, and obtain the best beamformer and the best AN beamformer. Step 6: Receive new echo signals from the direction of the unknown attacker Eve, update the parameters of the tracking model constructed in Step 2, and proceed to the next iteration.

2. The ISAC proactive secure transmission method based on prediction-action as described in claim 1, characterized in that: In step 1, the multi-user ISAC system includes a system equipped with The base station BS of the root antenna is indexed as follows Single-antenna legitimate user User, index is of Antenna-friendly and passive sensing receiver (SR) and a decision fusion center (DFC), base station (BS) simultaneously send... Individual users transmit confidential information, i.e., composite signals. A confidential data stream is transmitted, and the Region of Interest (RoI) is actively sensed. Simultaneously, echo signals from targets within the RoI are captured by a friendly passive sensing receiver (SR). All signals from the base station (BS) and... The echo information collected by the distributed friendly passive sensing receiver (SR) is aggregated at the decision fusion center (DFC). The DFC performs data fusion to make a final detection decision. The base station (BS) uses the ISAC waveform to track and interfere with potential unknown attackers (Eve) to achieve secure beamforming. Specifically, this includes the following steps: Step 1.1: Establish a communication signal model in a multi-user ISAC system, for each time slot. Depend on It consists of symbols, at each symbol time. The composite signal transmitted by the base station (BS) consists of a linear superposition of K confidential data streams and an AN waveform. for: in, User The confidential information symbols are the symbols of the K confidential data streams in the composite signal. It is the symbol for dual-purpose artificial noise AN. It is a data beamformer. It is an AN beamformer; Transmitted composite signal covariance matrix Represented as: in, For the conjugate transpose of the data beamformer, This is the conjugate transpose of the AN beamformer. It is the conjugate transpose; Step 1.2: Establish a channel model, from base station BS to the nth time slot. The channel vectors of the friendly passive sensing receivers (SRs) are: in, This indicates Eve, a potential unknown attacker. , Indicates the large-scale path loss gain. It is distance Reference channel gain at that location, It is the base station BS and the first The distance between the three friendly passive sensing receivers (SRs) is α, where α is the path loss exponent. Indicates the emission angle AoD Corresponding transmit array steering vector: ,in, This indicates the phase difference between adjacent antennas; Step 1.3: Establish the received signal model. A friendly passive sensing receiver SR at any time The received signal is: in, It is complex Gaussian noise. Channel vector The conjugate transpose of; Step 1.4: Establish a sensing signal model. To detect targets in the Region of Interest (RoI), the multi-user ISAC system utilizes echoes from both monostable and M-bistable links for collaborative sensing. For any receiving node... Where m=0 represents BS, SR represents the echo signal vector. The model is as follows: in It is the complex reflection coefficient and , This represents complex Gaussian sensing noise; Indicates the emission angle AoD The conjugate transpose of the corresponding transmit array steering vector, Indicates the angle of arrival AoA The corresponding transmit array steering vector, where yes =1 indicates the corresponding AoA at SR; Step 1.5: Define the echo signal collected at time t, let... By superimposing the echo signals from all M+1 friendly passive sensing receivers (SRs), The echo signal collected at time t is: ， in, It is a block diagonal matrix, vector This represents the collected complex gain vector. This represents the collected noise vector; Decision Fusion Center (DFC) in time slots In Collect echo signals at each time point, and let The entire time slot The collected echo signals are compressed as follows: in, Represents the aggregated noise vector; Step 1.6: Define the round-trip channel matrix between base station BS and unknown attacker Eve: in, For complex gain, The reflection coefficient, Radar cross-section, For dynamic distance; Indicates AoA The corresponding transmission array steering vector, Indicates the angle of AoD or AoA. Indicates AoD The conjugate transpose of the corresponding transmit array steering vector; Step 1.7: The echo signal received by BS from the unknown attacker Eve is: in, In order to combine the delay of propagation, For Doppler frequency shift, The phase rotation caused by the target motion This is complex Gaussian sensing noise.

3. The ISAC proactive secure transmission method based on prediction-action according to claim 1, characterized in that: Step 2 specifically includes the following steps: Step 2.1: Establish the state model: Define the unknown attacker Eve in the state... The actual state in each time slot is: in, For location, For speed, The representative moment is The x-coordinate of the attacker Eve is currently unknown. The representative moment is The attacker Eve's ordinate position is currently unknown. The representative moment is The attacker Eve's horizontal velocity is currently unknown. The representative moment is The attacker Eve's vertical velocity is currently unknown. From time slot Time slot The state transition model is represented as: in It is the state transition matrix. Indicates the duration of a time slot. It is state evolution noise. Let be the covariance matrix of the state evolution noise; Step 2.2, Eve's predicted state and its corresponding prediction error covariance matrix They are respectively: ， in, Indicates time slot The posterior error covariance matrix in This is the state transition matrix; Step 2.3: Establish a measurement model; the base station starts from the time slot. echo signal To obtain new measurements, a matched filter MF is applied to utilize the known AN waveform. Get an estimate time slot The output signal-to-noise ratio (SNR) of the matched filter MF is written as: in, Indicates MF gain. It is distance Reference channel gain at that location, Indicates the emission angle AoD The corresponding transmit array steering vector conjugate transpose, This represents the receiver's total noise power. Represents the fourth power of the dynamic distance; Step 2.4, Real State and measurement vector The relationship between them is modeled as follows: in, Represents the speed of light. Indicates the carrier frequency. Indicates the target angle parameter sine value Measurement error, Indicates Doppler frequency shift Measurement error, Indicates combined propagation delay Measurement error, , and Modeled as a zero-mean Gaussian random variable with variances of , and , For location, For base station (BS) coordinates, The horizontal coordinate represents the fixed position of the base station (BS). This represents the fixed vertical coordinate position of the base station (BS). Step 2.5: Define a nonlinear function To encapsulate , and The new measurement model is written as: in, It is a zero-mean Gaussian measurement noise vector. Let covariance matrix be the variance matrix. , and Modeled as , , , All are constants, using a new measurement model. Predicted state for Eve Make corrections, due to the measurement function It is non-linear, a non-linear function. First, by predicting the state around Eve. Linearizing the first-order Taylor expansion, we get: in, yes relative to Eve's predicted state Surrounding The Jacobian matrix, using a linearized model, is obtained through Kalman gain. Posterior state estimation and posterior error covariance Three consecutive calculations to update the state in, This represents a 4×4 identity matrix.

4. The ISAC proactive secure transmission method based on prediction-action according to claim 1, characterized in that: Step 3 specifically includes the following steps: Step 3.1, Communication Performance Modeling: Communication performance is evaluated through the quality of service for each legitimate user: The signal-to-noise ratio (SINR) of the kth legitimate user is expressed as: in, In time slot The conjugate matrix of the channel vector from BS to user k receiver. For users Data beamformer, For AN beamformer, For users Data beamformer; Step 3.2, Perception Performance Modeling: The Decision Fusion Center (DFC) processes the aggregated data across the entire time slot. Collected echo signals To make a testing decision, this is represented as a binary hypothesis test: in, This indicates that there is no target in the RoI. This indicates that the target exists; We employ the generalized likelihood ratio test (GLRT) framework to establish a detector with hypothesis testing capabilities: Let... The decision threshold is represented by the GLRT detector. : in, and They are and Next moment Collected echo signals The probability density function (PDF); Analysis of GLRT detectors The detection performance, characterizing the test statistic. The distribution exist and The distribution converges to different forms of non-central chi-square distributions: in, This indicates parameters with degrees of freedom (DoF) and non-centrality. Non-central chi-square distribution, non-centrality parameter Represented as: in, Represents the trace of a matrix. , The average power of the target reflected signal; given a threshold Collaborative sensing solution The detection probability assessment is as follows: ,in It has 2 degrees of freedom and the non-central parameter is The right-tailed probability function of the non-central chi-square variable; Step 3.3, Two-Layer Security Performance Modeling: To prevent information interception and exposure of sensitive locations, a two-layer security model was established. This model is quantified by two performance indicators: Eve's information leakage rate and Eve's detection probability. Specifically, it includes the following steps: Step 3.3.1, Eve's information leakage rate: Eve's worst-case information leakage rate per user is: in, In time slot The conjugate matrix of the channel vector from BS to Eve receiver. This represents the noise power received by the attacker. Step 3.3.2, Eve's detection probability: The communication stealth is evaluated by using Eve's detection probability performance. Similarly, The formula for the binary hypothesis test is: in, For the friendly passive sensing receiver SR at time Received signal, null hypothesis Instruct the base station (BS) to remain silent, and exist The following distribution is in Then, conversely, the alternative hypothesis This indicates that there is communication activity between the base station (BS) and the legitimate user, therefore ; definition Suppose Eve uses the Neyman-Pearson criterion to design a function with a preset threshold. The optimal LRT detector: in, , Utilizing the established optimal LRT detector To model the detection probability of Eve and quantify the stealth of communication, the following steps are taken: Given a preset threshold Eve detection probability The assessment is as follows: in, .

5. The ISAC proactive secure transmission method based on prediction-action according to claim 1, characterized in that: Step 4 specifically includes the following steps: Step 4.1, Prediction Phase: Eavesdropping Channel Reconstruction, first using the previous Eve prediction state. To construct the first A time-slot eavesdropping channel based on Eve's predicted state and BS coordinates. Predicted distance and launch angle : Then the predicted eavesdropping channel Reconstructed as: Due to unavoidable noise interference, the actual eavesdropping channel will be... Modeling as a predictive eavesdropping channel and the norm of prediction error The sum: Wherein, the norm of the prediction error is From the known constant Definition, that is , This is a known upper bound for the channel prediction error energy.

6. The ISAC proactive secure transmission method based on prediction-action according to claim 1, characterized in that: Step 5 specifically includes the following steps: Step 5.1: To maximize the accuracy of collaborative perception To simultaneously meet the Quality of Service (QoS) requirements of legitimate users, eavesdropper tracking accuracy, and dual-layer security under channel prediction uncertainty, all performance metrics are omitted for the sake of symbol simplicity. The explicit dependency means that the optimization problem is formulated as follows: in, Minimum Quality of Service (QoS) for each legitimate user. This represents the probability of Eve being detected. This serves as a unified upper bound for the probability of Eve's detection. Eve's information leakage rate; This serves as a unified upper bound for Eve's information leakage rate; This represents the total transmission power. Step 5.2: A three-step optimization algorithm was developed to solve the problem. This includes: problem restatement, beamforming design and randomization based on semi-definite relaxation (SDR): where, Problem rephrased: Accuracy of collaborative perception Performing an equivalent transformation, for any fixed threshold and degrees of freedom (DoF), the function... Is it a non-centrality parameter? It is a monotonically increasing function; therefore, maximizing the accuracy of collaborative sensing... Equivalent to maximizing the noncentrality parameter The noncentrality parameter is then simplified to maximizing the quadratic term. ; Step 5.2, Solve non-convex constraints: Let , The Quality of Service (QoS) constraints in the code are rewritten as follows: Step 5.3, the concealment constraint is And rewrite the concealment constraint as follows: ； The original implicit constraint is equivalently transformed into: ,in It is to satisfy The value; Step 5.4: For the sake of simplicity, the index has been omitted. The original hidden constraint equation is restated as follows: Step 5.5, Combining Bounded prediction error constraints Rephrased as: Step 5.6: Apply the original concealment constraints Convert to a linear matrix inequality (LMI): in, ; The constraint on bounded prediction error is expressed as: in, ,and ; Original tracking accuracy constraints It is transformed into the following set of linear matrix inequalities (LMIs): ； in, It is a selection vector, the first One position is 1, and the other positions are 0; Based on the above transformation, the problem The equivalent restatement is as follows: Step 5.7, Beamforming Design Based on Semi-Deterministic Relaxation (SDR): Using SDR technology to address the problem... The relaxation is a semi-deterministic relaxation (SDR) beam, specifically: Define the following matrix: and constraint and So, the question is... Equivalent to in , ,Will Convert to a convex semidefinite program (SDP); Step 5.8, Randomization: The solution obtained from the convex semidefinite procedure (SDP). It cannot be guaranteed to be rank one if the obtained matrix If the rank is 1, then the optimal beamformer can be directly recovered through eigenvalue decomposition. and the best AN beamformer Otherwise, a standard Gaussian randomization process is used to generate a high-quality rank-one solution.

7. The ISAC proactive secure transmission method based on prediction-action according to claim 1, characterized in that: Step 6 specifically includes the following steps: Step 6.1: Obtain the optimal beamformer and the best AN beamformer Then, the base station BS transmits the actively designed ISAC signal at the beginning of time slot n; Step 6.2: Capture the echo signal reflected by Eve, the unknown attacker. The measurement vector for time slot n is generated through MF processing. ; Step 6.3: Utilize this measurement vector Obtained according to EKF update rules and and will and As a subsequent time slot The key inputs in the prediction phase complete the proactive safety cycle.

8. The ISAC proactive secure transmission method based on prediction-action according to claim 1, characterized in that: The ISActive framework includes an input phase, a prediction phase, an action phase, and a perception and update phase, wherein... During the input phase, the predicted state of Eve in time slot n-1. and posterior error covariance matrix As input; In the prediction phase, the predicted state of Eve is used as input. The current time slot is predicted using the extended Kalman filter (EKF) based on the posterior error covariance matrix. Predicted distance and launch angle Based on the predicted target distance and angle, reconstruct the eavesdropping channel between BS and Eve; During the operation phase, active beamforming is generated based on the reconstructed eavesdropping channel, and the optimal data beamformer and the optimal AN beamformer are generated to achieve secure ISAC transmission. During the sensing and updating phase, the BS receives the echo signal from Eve, performs sensing measurements through a matched filter to obtain the measurement value of the current time slot, and substitutes the measurement value into the Extended Kalman Filter (EKF) update rule to calculate the current time slot. Eve's predicted state That is, the posterior state estimate and the posterior error covariance matrix. This will be used as the input for the subsequent step 1.

Images