Drone state-aware protection method with beamforming feedback frame injection

Abstract

The application discloses a kind of unmanned aerial vehicle state awareness protection methods of beamforming feedback frame injection, first obtain legal beamforming feedback information BFI sequence, construct protection reference sample library;According to the state and condition of unmanned aerial vehicle, determine flight safety protection time window, and be divided into multiple candidate protection time slots;Joint optimization protection BFI frame injection time and frame content, so that the BFI sequence observed by unauthorized awareness party deviates from the discrimination area corresponding to real flight state or real track, or be induced to preset safe flight category;Finally, the protective BFI matrix obtained by optimization is encoded into protective feedback frame, and is sent in the corresponding time slot.The application can exert antagonistic influence on unauthorized unmanned aerial vehicle radio frequency awareness and trajectory inference result by a small amount of protocol-compatible protective feedback frame, improve the safety protection capability of unmanned aerial vehicle flight trajectory, task state and key area activity.

Description

Technical Field

[0001] This invention belongs to the technical fields of UAV communication security, wireless sensing security and air-to-ground link security protection. It relates to an active protection method for UAVs when beamforming feedback information is used by a third-party radio frequency sensing system in the air-to-ground wireless link. It mainly relates to a UAV state perception protection method that injects beamforming feedback frames. Background Technology

[0002] With the development of applications such as low-altitude economy, drone inspection, logistics delivery, emergency communication, park security, and smart city monitoring, drones need to maintain stable wireless communication with ground stations, access points, relay nodes, or edge computing nodes in complex low-altitude environments. In order to improve the capacity, coverage, and multipath resistance of air-to-ground links, wireless links that support orthogonal frequency division multiplexing, multiple-input multiple-output, and beamforming are increasingly being used for drone mission data, telemetry data, and video backhaul scenarios.

[0003] During beamforming, the transmitter typically sends training fields or probe frames, and the receiver estimates the air-to-ground channel based on the received training fields and feeds back the compressed beamforming feedback information to the transmitter. This type of beamforming feedback information was originally used to assist in beam selection, link adaptation, and spatial flow scheduling, but its time-series variations are also closely related to the UAV's position changes, speed, heading, attitude, takeoff and landing, hovering, turning, and approach to the target area.

[0004] Because some beamforming feedback frames can be passively eavesdropped on by third-party devices within the same wireless coverage area in a real system, unauthorized radio frequency sensing parties can infer the UAV's takeoff and landing points, inspection routes, sensitive target approach behavior, mission phases, and activities in key areas simply by training trajectory inference models, mission phase identification models, or flight state classification models based on the BFI sequence without decrypting the UAV's payload data. This type of side-channel trajectory sensing poses security risks such as tracking, interception, interference, deception, and mission disclosure.

[0005] Existing UAV communication security methods primarily focus on business data encryption, identity authentication, anti-interference transmission, frequency hopping spread spectrum, or physical layer confidentiality. While these methods can protect data content or improve link anti-interference capabilities, they still lack specificity in addressing trajectory and mission intent leakage caused by the timing characteristics of beamforming feedback information. Directly disabling beamforming feedback, reducing the feedback frequency, or increasing random noise can affect the adaptive capability and reliability of legitimate air-to-ground links; simple random interference lacks design for perception model discrimination boundaries, often requires high injection strength, and is easily filtered out due to abnormal traffic characteristics.

[0006] Therefore, there is an urgent need for an active counter-reconnaissance protection method for UAV air-to-ground communication links. This method would enable legitimate UAVs, ground stations, or authorized protection equipment to alter the timing characteristics of the BFI observed by unauthorized sensing parties by accurately generating and sending a small number of protective beamforming feedback frames, while maintaining protocol compatibility, low communication overhead, and low link impact. This would prevent unauthorized sensing parties from correctly inferring the true flight trajectory or only obtaining a preset safe flight category. Summary of the Invention

[0007] This invention addresses the problem that BFI sequences in existing UAV air-to-ground wireless links can be eavesdropped on by third parties and used for unauthorized flight trajectory inference, mission phase identification, and sensitive area activity identification. It also addresses the issues of unstable random disturbance protection, high deployment costs for general encryption or protocol modifications, and the impact of reducing feedback frequency on legitimate flight control communication performance. The invention provides a UAV state awareness protection method using beamforming feedback frame injection. First, legitimate beamforming feedback information (BFI) sequences are acquired, and a protection reference sample library is constructed. Flight safety protection time windows are determined based on UAV state and conditions and divided into multiple candidate protection time slots. The injection time and frame content of protective BFI frames are jointly optimized to cause the BFI sequences observed by unauthorized sensing parties to deviate from the discrimination area corresponding to the actual flight state or actual trajectory, or to be induced to a preset safe flight category. Finally, the optimized protective BFI matrix is ​​encoded into protective feedback frames and transmitted in the corresponding time slots. This invention can exert adversarial influence on the radio frequency sensing and trajectory inference results of unauthorized UAVs through a small number of protocol-compatible protective feedback frames, improving the security protection capabilities for UAV flight trajectories, mission states, and activities in critical areas.

[0008] To achieve the above objectives, the technical solution adopted by the present invention is: a UAV state awareness protection method with beamforming feedback frame injection, comprising the following steps:

[0009] S1. Acquisition of UAV air-to-ground link BFI sequence and construction of flight status reference sample library: Acquire beamforming feedback information (BFI) sequence between legitimate UAVs and ground stations, access points or relay nodes, and decode, reconstruct, time-align and normalize the BFI sequence to obtain a protection reference sample library corresponding to the UAV flight status, track segment or mission stage.

[0010] S2. Flight safety protection window and candidate protection time slot division: The flight safety protection window is determined based on the initial change of flight status in the real-time BFI stream or the preset flight safety protection trigger conditions, and the flight safety protection window is divided into multiple candidate protection time slots;

[0011] S3. Establishment of the protective injection optimization problem: Set injection selection variables and protective frame content parameters for each candidate protection time slot, and establish the generation mapping from protective frame content parameters to the protective BFI matrix, flight safety protection loss function, injection sparsity constraints, local statistical bias constraints, as well as delay robustness enhancement and multi-sample robustness enhancement.

[0012] S4. Joint optimization solution of protective injection strategy: Based on the unauthorized perception model or its alternative model, solve the protective injection optimization problem established in step S3, jointly determine the transmission time and frame content of the protective BFI frame, so that the unauthorized perception model's identification result of the real UAV flight status is non-directionally confused or induced to the preset security category.

[0013] S5. Protective Feedback Frame Encoding and Legitimate Injection: The optimized protective BFI matrix is ​​encoded into a protective feedback frame conforming to the air-to-ground wireless link beamforming feedback format. The protective feedback frame is then transmitted by a legitimate UAV, ground station, or authorized protection device in the candidate protection time slot, so that the unauthorized sensing party receives the protected BFI sequence composed of the real BFI frame and the protective BFI frame.

[0014] As an improvement to the present invention, step S1 includes at least air interface monitoring and BFI flight state sequence reconstruction:

[0015] Authorized protection equipment listens to the air interface beamforming feedback frame and parses it to obtain the first... The feedback frame in the first Quantization feedback angle set on each subcarrier The set of quantitative feedback angles By drones from legitimate sites Channel matrix of subcarriers After performing singular value decomposition, phase adjustment, and Givens rotation quantization, the following is obtained:

[0016]

[0017] in, For the first Channel matrix of subcarriers, and These are the left singular matrix and the right singular matrix, respectively. It is a singular value matrix. for The conjugate transpose of;

[0018] Authorized protective equipment Perform inverse quantization and inverse Givens rotation to reconstruct the corresponding BFI matrix:

[0019]

[0020] in, This represents the reconstruction mapping from the feedback angle to the phase-adjusted beamforming matrix; it reconstructs the beamforming matrix of each subcarrier within the same feedback frame. A single-frame BFI representation is obtained by concatenating the indices. and will be multiple Arranged chronologically, the BFI UAV flight status observation sequence is obtained. .

[0021] As an improvement of the present invention, in step S1, the arrival timestamp of the original feedback frame is recorded as... In the time interval Internal structure Uniform sampling time:

[0022]

[0023] in, For the r-th uniform sampling time, The target frame number after resampling; if Located in adjacent timestamps and Between, that is Then, linear interpolation is used to obtain the corresponding uniformly sampled BFI frame:

[0024]

[0025] in, For the first BFI information corresponding to each uniform sampling time point. This yields the time-uniformed BFI action sequence. .

[0026] As an improvement of the present invention, in step S1, each BFI element is scaled according to the average amplitude of the entire BFI sequence so that the BFI sequence under different acquisition conditions has a consistent amplitude range.

[0027] As an improvement of the present invention, in step S2, the flight safety protection window Classified as There are 1 candidate protection time slots, and the center time of each candidate protection time slot is 1. ; Regarding the first One candidate protection time slot, set encoding variables ,in To inject selection variables, These are parameters for the protective frame content.

[0028] As another improvement of the present invention, in step S3, a generation mapping from the protective frame content parameters to the protective BFI matrix is ​​established; the generation mapping specifically includes:

[0029]

[0030] in, For the first The protective BFI matrix corresponding to each candidate protection slot, The reference BFI matrix is ​​selected from the protection reference sample library. For the protective frame content parameters to be optimized, and These are the Fast Fourier Transform and Inverse Fast Fourier Transform along the subcarrier dimension, respectively. For the number of subcarriers, For column-wise normalization operation; the normalization makes The following energy constraints must be satisfied:

[0031]

[0032] in, Number of transmitting antennas For the protective BFI matrix in the 1st The first transmitting antenna and the first Elements at each subcarrier.

[0033] As another improvement of the present invention, in step S3, the joint optimization objective of the protective injection optimization problem is:

[0034]

[0035] in, For protective injection strategy, For inputs after protection observed by unauthorized sensing parties, To protect flight safety from losses, For the number of protective feedback frames, This refers to the local statistical deviation between protective BFI frames and normal BFI sequences. , and These are non-negative weighting coefficients; specifically, the local statistical bias is:

[0036]

[0037] in, For the first A protective BFI frame, and These are the left and right real BFI frames that are temporally adjacent to the protective BFI frame. It is the Frobenius norm.

[0038] As another improvement of the present invention, step S3 further includes a delayed robust protection constraint, which injects a protective sequence. Delay The process of one unit is denoted as The unauthorized sensing model outputs under multiple possible delay units are weighted and summed:

[0039]

[0040] in, For the first Class-based delay-enhanced output scores, The maximum number of delay units. For the first The weight corresponding to each delay unit and , For injection operation, This is a true BFI sequence. For unauthorized perception models or alternative models, the first The class outputs the score;

[0041] Step S3 also includes a multi-sample adaptation mechanism, which records multiple BFI samples collected by different drones or in different environments as... Based on the aforementioned delay robust protection constraints, multiple BFI samples and their unauthorized sensing model outputs under different delay conditions are jointly weighted to obtain a multi-sample delay enhancement output score:

[0042]

[0043] in, For the first Multi-sample delay augmentation of output scores for this class The number of BFI samples participating in multi-sample fitting. For the first One BFI drone status observation sample, For the first The weights of each BFI sample and .

[0044] Based on enhanced output score Define flight safety protection loss, flight safety protection loss Based on the protection mode, the following is selected:

[0045]

[0046] in, It is used in conjunction with the injection sparsity constraint and local statistical bias constraint in the protective injection optimization problem to jointly determine the injection time and frame content of the protective feedback frame.

[0047] As another improvement of the present invention, in step S4, the enhanced particle swarm optimization algorithm EPSO is used to jointly solve the protective injection optimization problem. Specifically, multiple candidate protective injection strategies are randomly generated as initial particles, and each particle includes an injection selection variable. And protective BFI frame content parameters; calculate the fitness value of each particle according to the optimization objective, and iteratively update the particle position based on the particle's historical best strategy and global best strategy; for the injection selection variables To address the discrete nature of the problem, crossover mutation operations involving probability flipping and inter-particle exchange are introduced to improve the solution capability in the hybrid discrete-continuous search space. The optimal protective injection strategy is output when the maximum number of iterations, fitness convergence, or flight safety protection effect meets preset requirements. .

[0048] As a further improvement of the present invention, in step S5, according to the optimal protective injection strategy... Generate and send protective feedback frames; according to Determine the candidate protection time slots to be injected and the corresponding protective BFI matrix. For the protective BFI matrix Column normalization and Givens rotation quantization are performed to obtain the beamforming feedback angle field; the feedback angle field is encapsulated into a protective feedback frame conforming to the beamforming feedback format of the air-to-ground wireless link, and sent by the legitimate UAV end, ground station end or air-to-ground link receiver in the corresponding candidate protection time slot, so that the unauthorized sensing party receives the protected BFI sequence.

[0049] Compared with the prior art, the present invention has at least the following beneficial effects:

[0050] (1) This invention extends the protection target from traditional communication data content to the UAV flight trajectory, flight status and mission stage leaked by BFI timing characteristics, thereby improving the protection capability against side-channel radio frequency sensing risks;

[0051] (2) This invention is not simple random noise or strong interference, but rather a joint optimization of the injection time and frame content of protective BFI frames based on unauthorized sensing models or alternative models, which can achieve stronger trajectory confusion or security category induction effects under conditions of a small amount of injection.

[0052] (3) The present invention introduces local statistical continuity constraints and injection frame quantity constraints, so that the protective feedback frame is closer to the normal flight link feedback in terms of timing statistics;

[0053] (4) The present invention improves the protection stability under UAV movement, environmental multipath changes, aircraft type differences and mission phase changes by delay robust enhancement and multi-flight sample robust enhancement. Attached Figure Description

[0054] Figure 1 This is a flowchart of the steps of the method of the present invention. Detailed Implementation

[0055] The present invention will be further illustrated below with reference to the accompanying drawings and specific embodiments. It should be understood that the following specific embodiments are for illustrative purposes only and are not intended to limit the scope of the invention.

[0056] Example 1

[0057] A method for protecting UAVs from state perception using beamforming feedback frame injection, such as Figure 1 As shown, the specific steps include the following:

[0058] Step S1: Acquisition of UAV Air-to-Ground Link BFI Sequence and Construction of Flight State Reference Sample Library: Acquire the beamforming feedback information (BFI) sequence within at least one UAV flight state time window between a legitimate UAV and a ground station, access point, or relay node. Perform time-uniform sampling and amplitude normalization processing on the BFI sequence to obtain the protection reference sample library. Step S1 includes the following sub-steps:

[0059] S11. Air Interface Monitoring and Feedback Analysis: In UAV air-to-ground wireless links supporting multiple-input multiple-output (MIMO) and beamforming, the ground station, access point, or UAV sends training frames, probe frames, or empty data packets. The link peer estimates the air-to-ground channel based on the received training fields and feeds back beamforming feedback frames. Authorized protection equipment monitors the air interface beamforming feedback frames of the protected wireless link and parses the beamforming data from the feedback frames to obtain the first beamforming data. The feedback frame in the first Quantization feedback angle set on each subcarrier .

[0060] S12, BFI matrix reconstruction: The set of quantized feedback angles obtained from step S11. By drones from legitimate sites Channel matrix of subcarriers The result is obtained after performing singular value decomposition, phase adjustment, and Givens rotation quantization, where:

[0061]

[0062] in, For the first Channel matrix of subcarriers, and These are the left singular matrix and the right singular matrix, respectively. It is a singular value matrix. for The conjugate transpose of . Authorized protective equipment for the set of quantized feedback angles. Perform inverse quantization and inverse Givens rotation to reconstruct the corresponding BFI matrix:

[0063]

[0064] in, This represents the reconstruction mapping from the feedback angle to the phase-adjusted beamforming matrix.

[0065] S13, Flight Status BFI Sequence Formation: This involves generating the sequence of subcarriers within the same feedback frame. A single-frame BFI representation is obtained by concatenating subcarrier indices. :

[0066]

[0067] Arranging multiple consecutive single-frame BFI representations in chronological order yields the UAV flight state BFI sequence:

[0068]

[0069] in, For the number of subcarriers, This represents the number of feedback frames.

[0070] S14. Time-uniform sampling and amplitude normalization:

[0071] The temporal uniform sampling of the BFI sequence is specifically performed as follows: the original BFI sequence is resampled by linear interpolation according to the target number of sampling frames, and the corresponding uniformly sampled BFI frames are obtained through linear interpolation.

[0072]

[0073] in, For the first BFI information corresponding to each uniform sampling time. and Adjacent timestamps For the r-th uniform sampling time, the time-uniformed BFI action sequence is obtained. ;

[0074] Let the arrival timestamp of the original feedback frame be... In the time interval Internal structure Uniform sampling time:

[0075]

[0076] in, For the r-th uniform sampling time, The target frame number after resampling; if Located in adjacent timestamps and Between, that is Then, linear interpolation is used to obtain the corresponding uniformly sampled BFI frame:

[0077]

[0078] in, For the first The BFI information corresponding to each uniform sampling time point is used to obtain the time-uniformed BFI action sequence. .

[0079] Next, amplitude normalization is performed, scaling each BFI element according to the average amplitude of the entire BFI sequence to ensure that the BFI sequences under different acquisition conditions have a consistent amplitude range.

[0080] S15. Construction of the protection reference sample library: For the flight state BFI sequence... Time alignment and resampling are performed to form a protection reference sample library. The protection reference sample library may include one or more of the following: ordinary cruise samples, hovering inspection samples, takeoff and landing samples, flight path turning samples, samples approaching sensitive areas, samples away from sensitive areas, decoy flight path samples, and preset safe flight category samples.

[0081] Step S2, Flight safety protection window and candidate protection time slot division: The flight safety protection window is determined based on conditions such as the UAV entering sensitive airspace, performing critical mission actions, changing flight paths, taking off and landing, hovering and inspecting, the operator, the mission control terminal or flight control system manually triggering protection or detecting unauthorized radio frequency sensing risks, etc. The flight safety protection window is divided into multiple candidate protection time slots.

[0082] During the online protection phase, legitimate drones, ground stations, or authorized protection equipment continuously monitor real-time BFI flow, telemetry status, mission status, and airspace safety status. A flight safety protection window is determined when a drone is detected entering sensitive airspace, approaching critical infrastructure, performing hovering inspections or image / mapping, experiencing a sudden turn in its flight path, a sudden change in speed or altitude, switching mission modes, being manually triggered by operators, the mission control terminal, or the flight control system, or when a third party perceives an increased risk. For situations where the starting point of the flight status is difficult to detect precisely, a buffer period can be set before and after the trigger time to cover early trajectory characteristics.

[0083] Flight safety protection window Divided into There are 1 candidate protection time slots, and the center time of each candidate protection time slot is denoted as _____. Each candidate protection slot serves as a potential transmission location for a protective feedback frame, used in subsequent joint optimization to determine whether to inject a protective feedback frame. Each candidate protection slot corresponds to a set of coding variables. ,in To inject selection variables, used to characterize whether in the first... Each candidate protection time slot sends a protective BFI frame. These are protective frame content parameters, used to characterize the transform domain coefficients or structured perturbation parameters of the protective BFI matrix to be generated.

[0084] Step S3: Establishment of the Protective Injection Optimization Problem: Based on the multiple candidate protection time slots obtained in Step S2, a protective injection optimization problem is established to jointly determine the transmission time and content of the protective BFI frame. The protective injection optimization problem aims to reduce the flight state recognition capability of the unauthorized sensing model, reduce the number of protective feedback frame injections, and maintain the local statistical continuity of the BFI sequence. Its joint optimization objective is:

[0085]

[0086] in, For protective injection strategy, For inputs after protection observed by unauthorized sensing parties, To protect flight safety from losses, To constrain the number of frames injected for protection, To constrain the local statistical deviation between the protective BFI frame and the real BFI sequence, , and These are non-negative weighting coefficients. The solution to this optimization problem is used in step S4 to determine whether each candidate protection time slot should send a protective feedback frame and the corresponding protective BFI frame content.

[0087] Protective injection strategy variable settings: Based on the time slot division and corresponding coding variables in S2, the protective injection strategy is composed of the coding variables of all candidate protection time slots. .

[0088] Input modeling: Let the actual flight state BFI sequence be... Unauthorized sensing party The input after preprocessing is ,in, This indicates the resampling normalization or feature extraction preprocessing that the unauthorized sensing model might employ. After the protective feedback frame is injected, the protected input observed by the unauthorized sensing party is:

[0089]

[0090] in, This indicates a protective injection operation. This indicates a protective injection strategy.

[0091] The unauthorized perception model refers to a model that an unauthorized perception party may use, which takes the BFI sequence as input and outputs the UAV flight state category judgment result; when this model cannot be obtained, an alternative model pre-established by the protection reference sample library is used to simulate the unauthorized perception party's judgment result on the BFI sequence.

[0092] First objective: Flight safety protection loss. To measure the protective effect of the protective injection strategy on the unauthorized perception model, flight safety protection loss is defined. .make This indicates that the unauthorized perception model or its alternative model affects the protected input. The output of the first Enhanced scores.

[0093] The flight safety protection in this invention includes two modes: one is a non-directional confusion mode, which makes it difficult for unauthorized sensing parties to correctly judge the actual flight status of the UAV through protective BFI frames; the other is a directional safety category induction mode, which makes unauthorized sensing parties more inclined to judge the current BFI sequence as a preset safe flight category through protective BFI frames.

[0094] In the non-directional flight state confusion mode, the loss of flight safety protection is:

[0095]

[0096] in, Loss due to confusion caused by non-directional flight state Classified as either a true flight status category or a sensitive flight status category. This is the classification margin parameter. It is minimized... Reduce the accuracy of unauthorized perception models in classifying real flight states. The discriminative advantage. At this point, the output category of the unauthorized perception model is non-directionally changed to a category other than the true category.

[0097] Under the directional safety category guidance mode, the flight safety protection loss is:

[0098]

[0099] in, To induce losses for targeted safety categories, To preset safe flight status categories. By minimizing Improve the unauthorized perception model for preset security categories The discriminative advantage. At this point, the output category of the unauthorized perception model is redirected to the preset state category.

[0100] The flight safety protection loss Based on the protection mode, the following is selected:

[0101]

[0102] The second objective is to constrain the number of injected frames. To reduce the overhead of injecting protective feedback frames and avoid abnormal traffic characteristics caused by too many protective feedback frames, a constraint on the number of injected frames is introduced:

[0103]

[0104] in, This indicates the number of protective feedback frames that need to be injected within the flight safety protection time window. It is used to limit the number of protective feedback frames sent within the flight safety protection time window and is determined by summing the injection selection variables corresponding to each candidate protection time slot. By minimizing the penalty for the number of injected frames in the objective function, the number of protective feedback frames is reduced as much as possible while still satisfying the protection effect, thus making the protective injection strategy as sparse as possible.

[0105] The third objective: Local statistical bias constraint. To reduce the abrupt changes in the protective BFI frames within the BFI time series, a local statistical bias constraint is introduced:

[0106]

[0107] in, For the number of protective feedback frames, For the first A protective BFI frame, and These are the left and right real BFI frames that are temporally adjacent to the protective BFI frame. This is the Frobenius norm. By minimizing this term, the local variation trends of the protective BFI frame are made consistent with those of the real BFI sequence.

[0108] Protective BFI Matrix Generation Mapping: To enable the protective frame content parameters to be converted into a protective BFI matrix, a reference BFI matrix is ​​selected from the protective reference sample library. And establish a protective frame content parameter To the protective BFI matrix Generate mapping:

[0109]

[0110] in, For the first The protective BFI matrix corresponding to each candidate protection slot, For reference BFI matrix, and Let these represent the Fast Fourier Transform and the Inverse Fast Fourier Transform along the subcarrier dimension, respectively. For the number of subcarriers, This indicates a column-wise normalization operation. The normalization makes... The following energy constraints must be satisfied:

[0111]

[0112] in, The number of transmitting antennas, For the protective BFI matrix in the 1st The first transmitting antenna and the first Elements at each subcarrier.

[0113] Delay-based robustness enhancement: To enable flight safety protection losses to adapt to media access backoff, queuing, hardware processing, and propagation delays, a protective injection sequence is implemented. Delay The process of one unit is denoted as And perform a weighted summation of the unauthorized perception model outputs under multiple possible delay units:

[0114]

[0115] in, For the first Class-based delay-enhanced output scores, The maximum number of delay units. For the first The weight corresponding to each delay unit and , For unauthorized perception models or alternative models, the first The class outputs a score.

[0116] Multi-sample robust enhancement: To ensure that flight safety protection losses can adapt to differences in BFI caused by different UAV equipment, different flight samples, or different environments, a protection reference sample library is selected. BFI samples Through collaborative optimization, multi-sample delay-enhanced output scores are obtained:

[0117]

[0118] in, For the first Multi-sample delay augmentation of output scores for this class For the first One BFI sample, For the first The weights corresponding to each BFI sample and When using delayed robust enhancement and multi-sample robust enhancement, in S33... Pick This incorporates delay disturbances and sample differences into the calculation of flight safety protection losses.

[0119] The aforementioned flight safety protection losses, injection frame quantity constraints, local statistical bias constraints, and delay robustness enhancement and multi-sample robustness enhancement are combined to form a protective injection optimization problem. This protective injection optimization problem is used in subsequent step S4 to jointly solve whether each candidate protection time slot should send a protective feedback frame and the corresponding protective BFI frame content.

[0120] Step S4: Joint Optimization Solution of Protective Injection Strategy: Based on the unauthorized perception model or its alternative model, solve the protective injection optimization problem established in Step S3, jointly determine the transmission time and frame content parameters of the protective BFI frame, and obtain the protective BFI matrix using the content parameters of the protective BFI frame and the generation mapping described in S3. The defender can, depending on the selected protection mode, cause the unauthorized perception model's identification results of the real flight state to be non-directionally confused or induced to a preset security category.

[0121] S41. Particle Swarm Initialization: This embodiment uses the Enhanced Particle Swarm Optimization (EPSO) algorithm for solving the problem. Randomly generated... Each candidate protective injection strategy is used as an initial particle, and each particle includes an injection selection variable corresponding to the candidate protective time slot. And the protective BFI frame content parameters corresponding to each candidate protection time slot.

[0122] S42. Fitness Calculation: Decode each candidate protective injection strategy into a corresponding protective injection sequence. And calculate its fitness value based on the protective injection optimization objective established in step S3. :

[0123]

[0124] S43, Particle Update and Crossover Mutation: Based on the historical optimal protective injection strategy for each particle and the global optimal protective injection strategy of the particle swarm, the velocity and position of the particles are iteratively updated, causing the candidate protective injection strategies to gradually search towards directions with better fitness. Specifically, in the... In the nth iteration, let the th The current position of each particle is The speed is Its historically optimal protective injection strategy is The globally optimal protective injection strategy for particle swarm optimization is: The particle velocity is then updated as follows:

[0125]

[0126] And update the particle positions:

[0127]

[0128] in, Inertial weights are used to balance global and local search capabilities. and These are individual learning factors and group learning factors, respectively. and A random number that follows a uniform distribution in the range [0,1].

[0129] S44, Crossover Mutation Enhancement: Targeting Injection Selection Variables To address the discrete nature of the problem, cross-mutation operations involving probability flipping and inter-particle exchange are introduced during the iteration process to improve the solution capability of the hybrid discrete-continuous search space.

[0130] S45. Optimize the output: Repeat steps S42 (fitness calculation), S43 (particle update), and S44 (crossover mutation) until the maximum number of iterations is reached, fitness converges, or the flight safety protection effect meets the preset requirements. Output the optimal protective injection strategy. The optimal protective injection strategy This includes whether protective feedback frames are injected into each candidate protection time slot and the content parameters of the corresponding protective BFI frames. The protective BFI matrix is ​​obtained using the content parameters of the protective BFI frames and the generation mapping described in S36.

[0131] Step S5, Protective Feedback Frame Encoding and Legitimate Injection: The protective BFI matrix mapped in S45 is encoded into a protective feedback frame conforming to the air-to-ground radio link beamforming feedback format. This protective feedback frame is then transmitted by a legitimate UAV, ground station, or authorized protection equipment in the candidate protection time slot confirmed in Step S4, ensuring that the unauthorized sensing party receives a protected BFI sequence composed of both the real BFI frame and the protective BFI frame. This step specifically includes the following sub-steps:

[0132] S51. Determination of the protective BFI matrix: based on the optimal protective injection strategy. Determine the candidate protection time slots to be injected and the corresponding protective BFI matrix. .

[0133] S52, Feedback Angle Field Generation: For the Protective BFI Matrix Column normalization is performed to ensure it meets the energy constraint of the beamforming feedback matrix; subsequently, the normalized protective BFI matrix is... Perform Givens rotation quantization to obtain the corresponding beamforming feedback angle field.

[0134] S53. Protective Feedback Frame Encapsulation and Transmission: The feedback angle field is written into the corresponding field of the air-to-ground wireless link beamforming feedback frame to generate a protocol-compatible protective feedback frame. Authorized protective devices then follow the optimal protective injection strategy. The protective feedback frame is sent in the corresponding candidate protection time slot.

[0135] S54. Formation of the protected BFI sequence: Since the unauthorized sensing party cannot distinguish between the real BFI frame and the protective feedback frame, it will use both as sensing inputs to obtain the protected BFI sequence, which will cause non-directional confusion or induce the identification results of the real flight status to the preset safety category.

[0136] In summary, the method of this invention is applicable to UAV air-to-ground wireless link scenarios supporting beamforming feedback. Addressing the issues of unauthorized flight trajectory inference and mission status leakage caused by plaintext propagation of BFIs (Broadcast Information Framework Frames) or their potential for eavesdropping by third parties, it proposes a protective feedback frame injection mechanism actively executed by a legitimate UAV, ground station, or authorized protection device. Unlike schemes that rely on random noise, simply reducing the feedback frequency, or entirely on protocol encryption, this invention first acquires and reconstructs the UAV flight state BFI sequence, then divides the flight safety protection time window and candidate protection time slots. Subsequently, it establishes a protective injection optimization problem that includes injection time, frame content, flight safety protection loss, constraints on the number of injected frames, constraints on local statistical bias, delay robustness enhancement, and multi-flight sample robustness enhancement. This problem is then jointly optimized and solved, resulting in a small number of protective BFI frames with low statistical bias. This achieves confusion and inducement against unauthorized third-party radio frequency sensing and trajectory inference, demonstrating good engineering deployability and UAV security protection effectiveness.

[0137] It should be noted that the above content merely illustrates the technical concept of the present invention and should not be construed as limiting the scope of protection of the present invention. For those skilled in the art, various improvements and modifications can be made without departing from the principle of the present invention, and all such improvements and modifications fall within the scope of protection of the claims of the present invention.

Claims

1. A method for protecting UAVs from state perception using beamforming feedback frame injection, characterized in that, Includes the following steps: S1. Acquisition of UAV air-to-ground link BFI sequence and construction of flight status reference sample library: Acquire beamforming feedback information (BFI) sequence of the legitimate wireless link of the UAV. The BFI sequence is obtained by splicing the subcarriers in the same feedback frame according to the index to obtain a single frame BFI representation, and then arranging multiple consecutive single frame BFI representations in chronological order; After performing time uniform sampling and amplitude normalization processing on the BFI sequence, a protection reference sample library is obtained. S2. Flight safety protection window and candidate protection time slot division: Determine the flight safety protection window based on the UAV flight status or preset flight safety protection trigger conditions, and divide the flight safety protection window into multiple candidate protection time slots; S3. Establishment of the protective injection optimization problem: Based on the flight safety protection time window determined in step S2 and the multiple candidate protection time slots obtained by its division, injection selection variables and protective frame content parameters are set for each candidate protection time slot to construct a variable representation of the protective injection strategy; the flight safety protection loss function is defined based on the output results of the unauthorized perception model or its alternative model; the flight safety protection loss function is combined with the injection frame number constraint, local statistical bias constraint, delay robustness enhancement term and multi-sample robustness enhancement term to form the protective injection optimization problem; S4. Joint optimization solution of protective injection strategy: Based on the unauthorized perception model or its alternative model, solve the protective injection optimization problem established in step S3, jointly determine the transmission time and frame content parameters of the protective BFI frame, and then use the generation mapping in step S3 to obtain the protective BFI matrix. Insert the protective BFI matrix into the real BFI sequence to obtain the protected BFI sequence. This causes the unauthorized perception model to be non-directionally confused or induced to the preset safety category when using the protected BFI sequence to identify the real UAV flight status, thereby realizing non-directional flight status confusion and directional safety category induction. S5. Encoding of Protective Feedback Frames and Legitimate Injection: The protective BFI matrix optimized in step S4 is encoded into a protective feedback frame. The protective feedback frame is sent at the transmission time determined in step S4, so that the unauthorized sensing party receives the protected BFI sequence composed of real BFI frames and protective BFI frames.

2. The UAV state awareness protection method with beamforming feedback frame injection as described in claim 1, characterized in that: In step S1, the legitimate site drone targets the first... Channel matrix of subcarriers After performing singular value decomposition, phase adjustment, and Givens rotation quantization, the first... The feedback frame in the first Quantization feedback angle set on each subcarrier Authorized protective equipment for quantitative feedback angle set Perform inverse quantization and inverse Givens rotation to reconstruct the corresponding BFI matrix: ； in, This represents the reconstruction mapping from the feedback angle to the phase-adjusted beamforming matrix; This represents the BFI matrix reconstructed from the perspective of the original feedback. Subcarriers within the same feedback frame A single-frame BFI representation is obtained by concatenating the indices. and will be multiple Arranged chronologically, the BFI UAV flight status observation sequence is obtained. : ； ； in, K represents the number of feedback frames in the UAV state sequence, and K represents the number of subcarriers.

3. The UAV state awareness protection method with beamforming feedback frame injection as described in claim 2, characterized in that: In step S1, the temporal uniform sampling of the BFI sequence specifically involves: linearly interpolating and resampling the original BFI sequence according to the target number of sampling frames, and obtaining the corresponding uniformly sampled BFI frames through linear interpolation. ； in, For the first BFI information corresponding to each uniform sampling time. and Adjacent timestamps For the r-th uniform sampling time, the time-uniformed BFI action sequence is obtained. ; The amplitude normalization specifically involves performing uniform scaling on the BFI sequences after time-uniform sampling to ensure that BFI sequences under different acquisition conditions have a consistent amplitude range.

4. The UAV state awareness protection method with beamforming feedback frame injection as described in claim 1, characterized in that: In step S2, the flight safety protection window is set. Evenly divided into There are 1 candidate protection time slots, and the center time of each candidate protection time slot is denoted as _____. Each candidate protection time slot corresponds to a set of coded variables. ,in To inject selection variables, used to characterize whether in the first... Each candidate protection time slot sends a protective BFI frame. The protective frame content parameters are used to characterize the transform domain coefficients or structured perturbation parameters of the protective BFI matrix to be generated; each candidate protective time slot serves as a potential transmission location for the protective feedback frame.

5. The UAV state awareness protection method with beamforming feedback frame injection as described in claim 1, characterized in that: In step S3, the objective function of the protective injection optimization problem includes flight safety protection loss, injection frame number penalty, and local statistical bias penalty, specifically: ； in, For protective injection strategy, For input received by the unauthorized sensing party after protection. To protect flight safety from losses, To increase the number of protective injection frames, This is due to local statistical bias. , and These are the weighting coefficients.

6. The UAV state awareness protection method with beamforming feedback frame injection as described in claim 5, characterized in that: In the delayed robustness enhancement of step S3, the protective injection sequence is... Delay The process of one unit is denoted as The corresponding enhanced output score is: ； in, For the first Class-based delay-enhanced output scores, The maximum number of delay units. For the first Weight of delay units and , For injection operation, For unauthorized perception models or alternative models, the first The class outputs the score; In multi-sample robust enhancement, samples are selected from the protection reference sample library. BFI samples The multi-sample delay augmentation output score is obtained as follows: ； in, For the first Multi-sample delay augmentation of output scores for this class The number of BFI samples participating in multi-sample fitting. For the first One BFI sample, For the first The weights of each BFI sample and ; By based on Constructing flight safety protection losses: Under the confusion mode of non-directional flight state, the flight safety protection losses are: ； in, Loss due to confusion caused by non-directional flight state Classified as either a true flight status category or a sensitive flight status category. This is the classification interval parameter; Pick ; Under the directional safety category guidance mode, the flight safety protection loss is: ； in, To induce losses for targeted safety categories, This is a preset safe flight status category.

7. The UAV state awareness protection method with beamforming feedback frame injection as described in claim 5, characterized in that: The local statistical bias penalty is determined based on the distance between the protective BFI frame and its adjacent real BFI frames, specifically: ； in, For the first A protective BFI frame, and These are the left and right real BFI frames that are temporally adjacent to the protective BFI frame. It is the Frobenius norm.

8. The UAV state awareness protection method with beamforming feedback frame injection as described in claim 5, characterized in that... In step S3, a generation mapping from protective frame content parameters to a protective BFI matrix is ​​established, the generation mapping being based on a protective reference BFI matrix. ,right Perform a Fast Fourier Transform along the subcarrier dimension, replacing some of its low-frequency coefficients with protective frame content parameters. Then, perform an inverse fast Fourier transform and column normalization to ensure that the protective BFI matrix satisfies the energy constraint of the beamforming feedback matrix, specifically: ； in, For the first The protective BFI matrix corresponding to each candidate protection slot, This indicates column-based normalization. and Let these represent the Fast Fourier Transform and the Inverse Fast Fourier Transform along the subcarrier dimension, respectively. This represents the number of subcarriers.

9. The UAV state awareness protection method with beamforming feedback frame injection as described in claim 5, characterized in that: In step S4, the protective injection optimization problem is jointly solved using the Enhanced Particle Swarm Optimization (EPSO) algorithm, specifically including the following steps: S41. Initialize the particle swarm: Randomly generate Each candidate protective injection strategy is used as an initial particle, and each particle includes an injection selection variable corresponding to the candidate protective time slot. And the protective BFI frame content parameters corresponding to each candidate protection time slot; S42. Fitness Calculation: Decode each candidate protective injection strategy into a corresponding protective injection sequence. Calculate its fitness value; S43. Particle Update: Based on the historical best protective injection strategy for each particle and the global best protective injection strategy for the particle swarm, the velocity and position of the particles are iteratively updated to make the candidate protective injection strategy search in the direction of better fitness. S44, Cross-mutation Enhancement: A cross-mutation operation is introduced during the particle update process in step S43; the cross-mutation operation includes at least probabilistic flipping of the binary elements in the injection selection variable and exchanging the injection selection variable at the corresponding positions between different particles; S45. Termination and Output: Repeat steps S42-S44 until the maximum number of iterations is reached, the fitness value converges, or the flight safety protection effect meets the preset requirements. Output the globally optimal protective injection strategy. The globally optimal protective injection strategy This includes whether protective feedback frames are injected into each candidate protection time slot and the content parameters of the corresponding protective BFI frames, generating a protective BFI matrix sequence.

10. The UAV state awareness protection method with beamforming feedback frame injection as described in claim 9, characterized in that: In step S5, the protective BFI matrix is ​​quantized into a beamforming feedback angle field using Givens rotation, and then encapsulated according to the beamforming feedback frame format to obtain a protective feedback frame; the protective feedback pin is injected into the selection variable. At the corresponding time, a protective feedback frame is injected over the air interface.

Images