A Blind Sorting Method and System for Unmanned Aerial Vehicle (UAV) Remote Control Signals Based on Time-Slotted Grid Phase-Locked Loop
By using a YOLO detection model and time-slot grid phase-locked loop method, combined with dynamic histogram and phase residual calibration, the problem of high-precision sorting of UAV remote control signals in complex electromagnetic environments was solved. This enabled blind sorting of signals and deep mining of discontinuous temporal patterns under multi-UAV conditions, improving sorting accuracy and countermeasure capabilities.
Patent Information
- Application Number
- CN202611261801.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-19
- Publication Date
- 2026-11-17
AI Technical Summary
Existing UAV remote control signal sorting and detection methods are difficult to achieve high-precision alignment and phase-locking of time slot grids in complex electromagnetic environments, and cannot effectively mine frequency hopping timing dependence features, resulting in sorting decision failure. In particular, it is difficult to achieve high-precision blind sorting under the conditions of multiple UAVs with clock drift and frequency point loss.
The signal features are reconstructed using a YOLO-based target detection model. Combined with a time-slot grid reorganization mechanism based on dynamic adaptive histogram and phase residual calibration, the PrefixSpan algorithm is used to mine discontinuous time-series patterns. The Lempel-Ziv complex quantification and Levenshtein edit distance are used for fault-tolerant comparison to achieve high-confidence signal sorting.
It improves the blind detection capability of signals in environments with strong noise and multipath interference, solves the problem of time slot grid reassembly under clock jitter and transmission delay, realizes accurate deoverlapping and sorting of multiple batches of overlapping signals in complex environments, and enhances the multi-target countermeasure capability.
Smart Images

Figure CN123122576A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of UAV signal processing and non-cooperative electronic reconnaissance technology, specifically relating to a blind sorting method and system for UAV remote control signals based on time-slot grid phase-locked loop. Background Technology
[0002] With the opening of low-altitude airspace and the widespread adoption of drone technology, the security threats posed by non-cooperative drones are becoming increasingly prominent. Frequency hopping (FH) communication, due to its strong anti-jamming and anti-interception capabilities, has become the most important communication standard for drones. Therefore, the ability to detect and accurately separate the frequency hopping signals of target drones in complex electromagnetic environments, especially in harsh scenarios where multiple batches of unknown drones coexist and their time-frequency signals overlap and mix, is crucial for low-altitude defense and electronic countermeasures. Blind sorting of frequency hopping signals aims to classify and separate the frequency hopping signal sequences emitted by multiple unknown drones according to their sources (i.e., different physical entities) with high precision, without any prior communication protocols.
[0003] Currently, scholars have conducted a series of studies on the detection and parameter sorting of frequency hopping signals.
[0004] In signal detection and localization, patent CN114492540B proposes a signal time-frequency map detection and classification method based on the YOLO target detection model. This method transforms a one-dimensional signal into a two-dimensional time-frequency map and uses a convolutional target detection network to regress the pixel bounding boxes of the signal. However, such methods often only obtain pixel-level location boxes. Under strong electromagnetic interference and low signal-to-noise ratio conditions, they lack a high-precision reconstruction mapping mechanism for the physical world's start time and physical bandwidth, easily outputting a large number of noisy time-frequency fragments.
[0005] In the areas of parameter estimation and frequency-hopping station sorting, patent CN107612587A proposes a parameter estimation method for frequency-hopping signals in non-cooperative frequency-hopping communication, calculating the frequency-hopping time by first-order differential analysis of the time-frequency spectrum; patent CN115085761B proposes an asynchronous frequency-hopping station sorting method based on frequency-hopping descriptors, utilizing the frequency-hopping period and physical bandwidth characteristics to complete station partitioning. However, these traditional methods typically assume that the received signals do not overlap or collide and that the transmit and receive clocks are perfectly synchronized. In real-world adversarial scenarios, transmission delays, multi-source collisions, and receiver clock drift can cause the extracted Time of Arrival (ToA) to contain severe deviations and random jitter. Traditional methods based on ToA differential histograms or parameter clustering experience a precipitous drop in sorting performance when faced with clock drift and high bit error rates, and cannot reconstruct an absolute time-slot grid with strict phase alignment. In addition, existing intelligent sorting methods focus on macroscopic frequency hopping point classification, lacking the exploration of deep discontinuous timing patterns behind frequency hopping patterns. Once faced with local frequency point loss (missed detection) caused by channel multipath fading, it is easy to cause the state transition path to be interrupted, making blind sorting and frequency hopping sequence distribution decisions completely ineffective.
[0006] In summary, while existing UAV remote control signal sorting and detection methods have achieved a certain degree of signal extraction and sorting through target detection or parameter clustering, the complexity of the electromagnetic environment still presents the following problems that urgently need to be addressed: First, the lack of an effective time slot grid absolute starting point calibration and phase-locked loop mechanism makes it difficult to achieve high-precision alignment and collision detection of the time slot grid under multi-UAV conditions; second, the failure to effectively mine the frequency hopping time-series dependent features makes it highly susceptible to sorting decision failure under the dual interference of random frequency hopping interference and local frequency point loss; and third, the lack of fault-tolerant frequency hopping sequence frequency hopping pattern matching and chaotic quantization mechanisms makes it impossible to generate high-confidence frequency hopping patterns under non-cooperative detection conditions.
[0007] Therefore, how to achieve blind sorting of UAV frequency hopping signals and frequency hopping pattern mining under the condition of overlapping remote control signals of multiple batches of UAVs is a key technical problem that urgently needs to be solved in the field of non-cooperative low-altitude defense. Summary of the Invention
[0008] The purpose of this invention is to provide a blind sorting method and system for UAV remote control signals based on time slot grid phase-locked loop. This invention solves the problems of difficulty in accurately reconstructing and judging the grid due to the lack of effective time slot alignment and phase-locked loop mechanism in blind sorting of UAV remote control signals, as well as the difficulty in extracting deep discontinuous timing patterns under frequency point loss and interference. This invention effectively improves the accuracy of blind sorting of UAV remote control signals and has good non-cooperative electronic countermeasure value.
[0009] To achieve the above objectives, the technical solution adopted by the present invention is as follows:
[0010] Firstly, a blind sorting method for UAV remote control signals based on time-slot grid phase-locked loop is provided, including:
[0011] Based on the YOLO target detection model, the collected UAV radio frequency time domain signals are used to construct an initial frequency hopping signal set;
[0012] The initial frequency hopping signal is purified and denoised using the target morphological feature benchmark to obtain the UAV frequency hopping signal pool;
[0013] The start times of all frequency hopping signals in the UAV frequency hopping signal pool are extracted and constructed into a time series. The first-order time difference sequence of all adjacent frequency hopping signals is calculated, and the global frequency hopping period is estimated by finding the peak through dynamic adaptive histogram.
[0014] The absolute starting point of the grid synchronization to be calibrated is determined based on the time series and global frequency hopping period. The frequency hopping signal in the UAV frequency hopping signal pool is mapped to the grid slot. The frequency points of non-empty slots in the entire panoramic grid are extracted in ascending order of time slot index to generate the basis vector of the UAV frequency hopping center frequency that is logically continuous in time.
[0015] Traverse the basis vectors, replace each observed frequency point with the standard absolute frequency point with the smallest distance in the dictionary, and output a set of standardized frequency hopping sequences;
[0016] The standardized frequency hopping sequence set is sliced according to a fixed observation time window length to construct a projection sequence database. The projection sequence database is processed based on the sequence pattern mining algorithm to obtain a set of frequency hopping patterns and perform complex quantification. Finally, the core frequency hopping pattern sequence is output through multi-segment joint extraction and large number voting purification.
[0017] Several alternative methods are provided below, but they are not intended as additional limitations on the overall solution above. They are merely further additions or optimizations. Provided there are no technical or logical contradictions, each alternative method can be combined individually with respect to the overall solution above, or multiple alternative methods can be combined with each other.
[0018] Preferably, the YOLO-based target detection model constructs an initial frequency-hopping signal set from the collected UAV radio frequency time-domain signals, including:
[0019] Convert the UAV radio frequency time domain signal into a two-dimensional energy power spectrum;
[0020] The two-dimensional energy power spectrum is generated into a two-dimensional time spectrum image by means of amplitude logarithmic compression and normalization operations, as well as resolution resampling and channel expansion operations.
[0021] The two-dimensional time-spectrum image is input into a pre-trained YOLO object detection model to predict the class probability, confidence level, and pixel-level bounding box coordinates of the frequency-hopping signal target box.
[0022] The feature parameters of each frequency hopping signal are reconstructed based on the predicted pixel-level bounding box coordinates to obtain the initial set of frequency hopping signals.
[0023] Preferably, the step of reconstructing the feature parameters of each frequency-hopping signal based on the predicted pixel-level bounding box coordinates to obtain an initial set of frequency-hopping signals includes:
[0024] Take the first pixel-level bounding box coordinates ,in For the first The minimum value of the bounding box coordinates at the pixel level on the x-axis. For the first The maximum value of the bounding box coordinates on the x-axis at the pixel level. For the first The minimum value of the bounding box coordinates at the pixel level on the y-axis. For the first The maximum value of the bounding box coordinates at the pixel level on the y-axis;
[0025] Reconstructing the start time of the frequency hopping signal With end time ,as follows:
[0026]
[0027]
[0028] in, The total pixel width of the two-dimensional time-spectral image. This represents the absolute time span corresponding to the two-dimensional time spectrum image. The starting time corresponding to the left boundary of the two-dimensional time-spectrum image;
[0029] Reconstructing the center frequency With physical bandwidth ,as follows:
[0030]
[0031]
[0032] in, The total pixel height of the two-dimensional time-spectral image. The total analysis bandwidth of the two-dimensional time-spectral image. The starting physical frequency at the bottom of the spectrum;
[0033] Each reconstructed frequency-hopping signal feature parameter is expressed as a four-dimensional vector. By combining the characteristic parameters of all frequency hopping signals, the initial set of frequency hopping signals is obtained. , This represents the total number of frequency-hopping signals.
[0034] Preferably, the step of refining and denoising the initial frequency-hopping signal using target morphological feature benchmarks to obtain the UAV frequency-hopping signal pool includes:
[0035] Based on the termination time and start time in the characteristic parameters of the frequency hopping signal, the actual dwell time span of each frequency hopping signal can be obtained;
[0036] If the actual dwell time span of the frequency hopping signal is less than or equal to 0, or the physical bandwidth of the frequency hopping signal is less than or equal to 0, or the physical bandwidth of the frequency hopping signal is greater than the upper limit threshold of the environmental bandwidth, then the corresponding frequency hopping signal is removed to complete the coarse screening of the frequency hopping signal.
[0037] For the frequency-hopping signals remaining after coarse screening, the statistical median of the physical bandwidth and the actual dwell time span is calculated and used as the bandwidth benchmark value. and the baseline value of stay ;
[0038] After traversing the remaining frequency-hopping signals after the initial screening, only those satisfying the joint logic condition are retained. and The frequency hopping signal is used to obtain the UAV frequency hopping signal pool, in which For the first The physical bandwidth of a frequency hopping signal To provide a tolerance threshold, For the first The actual dwell time span of a frequency-hopping signal This is the dwell time tolerance threshold.
[0039] Preferably, the step of estimating the global frequency hopping period by dynamically adaptive histogram peak finding includes:
[0040] Remove the difference values smaller than the limit protection interval from the first-order time difference sequence to form an effective jump difference set;
[0041] The histogram group interval is dynamically set based on the maximum and minimum values of the effective jump difference set, and the maximum value indexing algorithm is executed to find the first statistical peak interval containing the most difference samples.
[0042] Extract all difference values that fall within the first statistical peak interval from the effective hopping difference set, and calculate the arithmetic mean of the extracted difference values as the global frequency hopping period of the UAV transmitter.
[0043] Preferably, determining the absolute starting point of the calibrated grid synchronization based on the time series and the global frequency hopping period includes:
[0044] Select the first valid point in the time series As the initial anchor point;
[0045] Calculate the frequency hopping signal relative to the global frequency hopping cycle. fractional phase residual sequence , Indicates the first The start time of a frequency hopping signal This indicates that a modulo-1 operation will be performed;
[0046] For fractional phase residual sequences Frequency hopping signals implement folding correction mechanism ;
[0047] Calculate the median of all fractional phase residual sequences after folding correction. Obtain the absolute starting point of the calibrated grid synchronization. .
[0048] Preferably, mapping the frequency hopping signal in the UAV frequency hopping signal pool to the grid slot includes:
[0049] Based on the calibrated absolute starting point of grid synchronization, the theoretical time slot index into which the frequency-hopping signal falls in the UAV frequency-hopping signal pool is calculated. And calculate the relative phase error magnitude. ;
[0050] When the relative phase error amplitude When the theoretical time slot index is less than the maximum error tolerance, If the corresponding slot is empty, then directly record the current frequency hopping signal; if the theoretical time slot index... If the corresponding slot is already occupied, a collision check is initiated. The collision check includes comparing the current relative phase error magnitude. The relative phase error magnitude with respect to the occupant's history Only when When the current frequency hopping signal is overwritten, the occupying frequency hopping signal is removed; otherwise, the current frequency hopping signal is discarded.
[0051] When the relative phase error amplitude If the error is greater than or equal to the maximum error tolerance, discard the current frequency hopping signal.
[0052] Preferably, the frequency hopping pattern set is subjected to complex quantification, including:
[0053] Perform non-overlapping substring parsing on the standardized frequency hopping sequence set, and count the total number of historical independent substrings parsed from the standardized frequency hopping sequence set;
[0054] For each frequency hopping pattern in the set of frequency hopping patterns, calculate the Levenshtein edit distance between any independent substring and the frequency hopping pattern. If there is a Levenshtein edit distance less than or equal to the single frequency point replacement tolerance, retain the current frequency hopping pattern; otherwise, discard the current frequency hopping pattern.
[0055] Secondly, a blind sorting system for UAV remote control signals based on time-slot grid phase-locked loop is provided, including a processor and a memory storing a number of computer instructions, wherein the computer instructions are executed by the processor to implement the steps of the blind sorting method for UAV remote control signals based on time-slot grid phase-locked loop.
[0056] Compared with the prior art, the beneficial effects of the present invention are as follows:
[0057] 1. Improved signal blind detection capability under strong noise and multipath interference environment: This invention uses YOLO network to locate targets in time spectrum map and designs a high-precision reconstruction mapping mechanism from pixel space to real physical feature space. Combined with adaptive two-dimensional median morphological quantization for coarse screening and fine filtering, it can accurately extract pure candidate frequency hopping signal sequences from backgrounds with extremely low signal-to-noise ratio and strong electromagnetic interference. It effectively solves the problem that the threshold of traditional energy detection algorithm is difficult to adapt and has a high false alarm rate in multipath noise background.
[0058] 2. Overcoming the challenge of time slot grid reassembly under clock jitter and transmission delay: This invention introduces a time slot grid reassembly mechanism based on phase residual calibration. By calculating the fractional residual phase, folding correction, and phase origin folding compensation, the starting point of the time-frequency grid of the frequency hopping signal is successfully locked. The problem of time slot collision of multiple physical sources is solved by judging with minimal error. Order-preserving reconstruction of the time slot grid is achieved under the condition of no prior protocol.
[0059] 3. Deep discontinuous causal pattern extraction under fading channels and spoofed hopping points is realized: This invention constructs a sequence projection database and uses the PrefixSpan algorithm to mine discontinuous hidden frequency hopping patterns, which can effectively avoid the defects of traditional continuous frequency hopping pattern mining (such as N-Gram) which is easily blocked by spoofed frequency hopping or multipath missed detection; at the same time, by combining Lempel-Ziv complex quantization and Levenshtein fault-tolerant comparison, high confidence mining of frequency hopping sequences in complex environments is realized.
[0060] 4. Possesses extremely strong blind deoverlapping and sorting capabilities for multi-batch overlapping and mixed signals: This invention introduces a phase judgment with minimal error in the time domain and incorporates the PrefixSpan algorithm, which supports discontinuous modes, into time-series mining, breaking the stringent limitation of traditional algorithms that require the signal flow to be continuous and interference-free. Even in complex, non-cooperative scenarios involving simultaneous intrusion by multiple drones and severe overlap and collision of frequency-hopping patterns in the time and frequency domains, it can still achieve accurate deoverlapping and independent sequence reconstruction of multi-source mixed signals, greatly enhancing multi-target countermeasure capabilities. Attached Figure Description
[0061] Figure 1 This is a flowchart of a blind sorting method for UAV remote control signals based on time-slot grid phase-locked loop according to the present invention;
[0062] Figure 2 This is a schematic diagram of the time-spectrum detection and pixel physical feature mapping of UAV remote control signals based on the YOLOX-S network according to the present invention;
[0063] Figure 3 This is a schematic diagram of the first-order arrival time difference histogram and the global absolute frequency hopping cycle lock point of the present invention.
[0064] Figure 4 This is a schematic diagram of the time slot reassembly and minimum error conflict judgment based on phase residual calibration of the present invention;
[0065] Figure 5 This is a schematic diagram of one-dimensional adjacent channel clustering and measurement frequency point normalization mapping according to the present invention;
[0066] Figure 6 This is a schematic diagram of the non-continuous hidden causal pattern extraction process based on PrefixSpan in this invention. Detailed Implementation
[0067] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0068] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to limit the invention.
[0069] In order to overcome the problems existing in the current technology, such as Figure 1As shown, this embodiment provides a blind sorting method for UAV remote control signals based on time-slot grid phase-locked loop, specifically including the following steps:
[0070] Step 1: Blind detection of UAV frequency hopping pulses based on YOLO target detection network: Based on the YOLO target detection model (this embodiment uses the YOLOX-S target detection model), the collected UAV radio frequency time domain signals (high-speed discrete I / Q signal sequences) are constructed into an initial set of frequency hopping signals; the aim is to use the target detection network to locate and detect the frequency hopping signals in the time-frequency map and reconstruct their physical characteristics.
[0071] Step 1.1, Complex Spectrum Matrix Construction: The receiver collects the high-speed discrete I / Q signal sequence from the RF antenna. Framing and windowing are performed, and a Short-Time Fourier Transform (STFT) is executed to obtain the time-frequency two-dimensional complex spectrum matrix. When performing a short-time Fourier transform, the analyzed resolution parameters satisfy the time-frequency resolution product constraint. The overlapping sliding mechanism of the windowing function is used to extract the transient features of signals with high time-frequency overlap. Subsequently, the modulus square is calculated to generate a two-dimensional energy power spectrum.
[0072]
[0073] in, For time frame index, corresponding to physical time ; This is a discrete frequency index, corresponding to the physical frequency. Specifically, in this embodiment, the physical sampling rate of the receiving antenna... When using sliding frames, the window function used for windowing is a Hamming window with a length of 1024 points, and the overlap between adjacent time frames is set to 75%.
[0074] Step 1.2, Time-Spectrum Image Normalization Transformation: Transform the two-dimensional energy power spectrum... Convert to normalized image (1) Amplitude logarithmic compression and normalization operation: Mapping the power spectrum energy level to Interval:
[0075]
[0076] in, For normalized images, and These are the maximum and minimum logarithmic values of the power spectrum within the current observation window. To prevent extremely small positive numbers from overflowing zero. (2) Resolution resampling and channel expansion operations: The image resolution is adjusted to the standard input size of the YOLOX-S object detection model, and a standard two-dimensional time-spectral image is generated through pseudo-color mapping or channel duplication. In this embodiment, the temporal spectrum image is resampled at a higher resolution to adapt to the input feature map dimension required by the depth object detection network. This includes extremely small positive numbers to prevent zero overflow. The settings are dynamically adjusted adaptively based on the system noise floor level. Specifically, the noise floor power is set to 0.01 to 0.05 times, and upper and lower limit clamping constraints are applied to avoid division by zero and logarithmic zero overflow. In this embodiment, a very small positive number to prevent zero overflow is adaptively set based on the current system noise floor level. for ; generated standardized two-dimensional time-spectral image The dimensions are width Pixels, High A three-channel color image of pixels.
[0077] Step 1.3, Target Detection and Localization of UAV Remote Control Signals Based on YOLOX-S Target Detection Model: (e.g.) Figure 2 As shown, the standardized two-dimensional time-spectrum image Input the data into the pre-trained YOLOX-S object detection model and perform the following operations:
[0078] (1) Feature extraction and multi-scale fusion: Image Deep time-frequency spatial feature extraction is performed through the backbone network of YOLOX-S, and then multi-scale feature map fusion is achieved through the neck network to cope with frequency hopping signals of different durations and bandwidths.
[0079] (2) Target detection head regression prediction: The detection head predicts the class probability and confidence level of the candidate frequency hopping signal target box. And pixel-level bounding box coordinates:
[0080]
[0081] (3) Non-maximum suppression (NMS): Set a confidence threshold Intersection over Union (IoU) threshold Remove redundant bounding boxes and output a set of high-confidence signal pixel-level bounding boxes:
[0082]
[0083] In the formula, For the first pixel-level bounding box coordinates , This represents the total number of pixel-level bounding box coordinates. For the first The minimum value of the bounding box coordinates at the pixel level on the x-axis. For the first The maximum value of the bounding box coordinates on the x-axis at the pixel level. For the first The minimum value of the bounding box coordinates at the pixel level on the y-axis. For the first The maximum value of the pixel-level bounding box coordinates on the y-axis. When outputting detection boxes, the YOLOX-S network's object detection head performs non-maximum suppression processing using a dual threshold filtering mechanism of intersection-overlap ratio (IoU) and confidence score to filter out redundant candidate detection boxes that overlap in the time-frequency space. In this embodiment, the IoU threshold for non-maximum suppression... The confidence threshold is set to 0.45. Set to 0.50.
[0084] Step 1.4, Feature Reconstruction Mapping from Primitive Space to Physical Space: For each output pixel bounding box The true physical characteristic parameters of the corresponding frequency hopping signal are reconstructed through mapping:
[0085] (1) Start time With end time reconstruction:
[0086]
[0087]
[0088] in, The total pixel width of the two-dimensional time-spectral image. This represents the absolute time span corresponding to the two-dimensional time spectrum image. The starting time corresponds to the left boundary of the two-dimensional time-spectrum image.
[0089] (2) Center frequency With physical (channel) bandwidth reconstruction:
[0090]
[0091]
[0092] in, The total pixel height of the two-dimensional time-spectral image. The total analysis bandwidth of the two-dimensional time-spectral image. This is the physical frequency starting at the bottom of the spectrum. Specifically, in this embodiment, the total pixel width of the image... Pixels, total pixel height Pixel; the absolute time span represented by the spectrogram at that time. The starting time corresponding to the left boundary Total physical bandwidth for time-frequency spectrum analysis The starting physical frequency at the bottom of the spectrum .
[0093] Step 1.5: Construct and output the noisy signal dataset: Express the characteristic parameters of each reconstructed frequency-hopping signal as a four-dimensional vector. All detected signals are collected to construct an initial frequency hopping signal set. Specifically, in this embodiment, all signals after target detection regression and physical feature reconstruction are directly summarized to output an initial frequency hopping signal set. .
[0094] Step 2: Adaptive two-dimensional morphological purification based on statistical median: The initial frequency hopping signal is purified and denoised using the target morphological feature benchmark to obtain a high-purity UAV frequency hopping signal pool.
[0095] Step 2.1, Feature Extraction and Initialization: Obtain the initial frequency hopping signal set of the noisy signal output from Step 1. The coordinate features of each signal are extracted, and the actual dwell time span of each signal in the time domain is calculated based on the start and end times of each signal. .
[0096] Step 2.2, Physical Boundary Constraint Coarse Screening: Set physical failure boundaries, and screen by judging whether the residence time and bandwidth meet the physical reality requirements, eliminating unreasonable time-frequency fragments, specifically eliminating those that meet the constraint conditions. or Invalid measurement signals. Simultaneously set an upper limit threshold for environmental bandwidth. Forced stripping of physical bandwidth Greater than the set threshold The frequency-hopping signal is coarsely screened. The upper limit threshold of the environmental bandwidth is set proportionally based on the current global bandwidth received by the system to eliminate broadband continuous wave interference. In this embodiment, an upper limit threshold of the environmental bandwidth is set. That is, 0.1 times the total analysis bandwidth received by the system.
[0097] Step 2.3, Core Benchmark Median Estimation: For the frequency-hopping signals remaining after coarse screening, calculate the statistical median of the physical bandwidth and the actual dwell time span to adaptively establish the UAV target morphological benchmark under the current electromagnetic environment: bandwidth benchmark value. Baseline value of stay , The median is calculated statistically. In this embodiment, the current adaptive bandwidth baseline value is statistically estimated. Adaptive pulse width median baseline value .
[0098] Step 2.4, Bidirectional Tolerance Logic and Filtering: Set the stringent bidirectional tolerance threshold for the bandpass filter. and Traversing each frequency hopping signal remaining after the initial coarse screening If and only if the joint logic condition is satisfied. and These signals will be retained to form a frequency-hopping signal pool for UAVs. The bidirectional tolerance threshold of the bandpass filter is proportionally set based on the established morphological bandwidth and residence time reference values to construct a bandpass filter capable of adapting to signal parameter jitter. In this embodiment, the stringent bidirectional tolerance threshold of the bandpass filter satisfies... ,and .
[0099] Step 3: Blind global period estimation based on the first-order arrival time difference histogram: (e.g.) Figure 3 As shown, the start times of all frequency hopping signals in the UAV frequency hopping signal pool are extracted and constructed into a time series. The first-order time difference sequence of all adjacent frequency hopping signals is calculated, and the global frequency hopping period is estimated by finding the peak through dynamic adaptive histogram.
[0100] Step 3.1, Time Series Extraction and Differentiation: Extracting the UAV Frequency Hopping Signal Pool Start time of all frequency hopping signals Based on the start time of extraction Constructing time series Arrange all start times in the time series along the time axis, and obtain the first-order time difference series by subtracting the first-order adjacent times. , Represents the first in a time series A start time, Represents the first in a time series A start time, Indicates the first Each difference value.
[0101] Step 3.2, Guard Interval Removal: Set the limit guard interval for transceiver RF hardware switching. From first-order time difference sequences Remove all values less than The invalid difference values are used to form the effective jump difference set. The extreme protection interval for transmitter RF hardware switching is set according to the minimum protection boundary of the target transmitter RF local oscillator phase-locked loop switching, filtering out pseudo-differential points introduced by transient switching in the time domain.
[0102] Step 3.3, Dynamic Adaptive Histogram Peak Finding: Dynamically set the histogram group interval based on the maximum and minimum values of the effective difference set, and execute the maximum value index (Argmax) algorithm to find the first statistical main peak interval containing the most difference samples. , This is the lower bound of the first statistical peak's binning level. This is the upper bound of the binning for the first statistical peak. The histogram group interval for dynamic adaptive histogram peak finding is finely divided based on the resolution accuracy extracted at the arrival time to ensure the temporal clustering of the peak interval. In this embodiment, the dynamic adaptive group interval of the histogram is set to... The first statistical main peak interval obtained by histogram peak location is: .
[0103] Step 3.4, Frequency Hopping Period Locking: Extract all difference samples that strictly fall within the first statistical peak interval and calculate their arithmetic mean. Establish this arithmetic mean as the global frequency hopping period of the UAV transmitter's underlying layer. This is used as the theoretical heartbeat step size. In this embodiment, the arithmetic mean of the samples within this interval is calculated to lock the obtained global absolute frequency hopping period. .
[0104] Step 4, Global Time-Slotted Grid Reassembly Mechanism Based on Residual Phase: such as Figure 4 As shown, the absolute starting point of the calibrated grid synchronization is determined based on the time series and the global frequency hopping period. The frequency hopping signal in the UAV frequency hopping signal pool is mapped to the grid slot. The frequency points of non-empty slots in the entire panoramic grid are extracted in ascending order of the time slot index to generate the basis vector of the UAV frequency hopping center frequency that is logically continuous in time. The aim is to reconstruct the phase-aligned absolute time slot grid and handle time slot collisions.
[0105] Step 4.1, Phase Residual Origin Folding Compensation Calculation: Select the time series constructed in Step 3.1 The first valid point As the initial anchor point, the first valid point refers to the first starting time after arranging all the starting times in the time series along the time axis. The frequency hopping signal is calculated relative to the theoretical heartbeat step size. fractional phase residual sequence , For the first The start time of a frequency hopping signal For the first A fractional phase residual sequence of a frequency-hopping signal. For The sample performs a folding correction mechanism Calculate the median of the residuals after folding correction. Obtain the absolute starting point of the calibrated grid synchronization. In this embodiment, the first valid arrival time is used. For reference, calculate the estimated absolute offset of grid synchronization. Therefore, the formula for calculating the absolute starting point of the calibrated grid synchronization is: In another embodiment, after calculating the fractional phase residual sequence, a weighted window function can be introduced to correct the calculated median of the residuals in order to suppress random phase jitter caused by noise and lock in a smoother grid synchronization starting point.
[0106] Step 4.2, Absolute Time Slot Theoretical Mesh Mapping: Calculate the UAV frequency hopping signal pool based on the calibrated grid synchronization absolute starting point. Frequency hopping signal in Theoretical time slot index to be included And calculate the relative phase error magnitude. In this embodiment, a locked frequency hopping period is used. In conjunction with the synchronization starting point, the theoretical slots where the signals are located are calculated one by one. With phase error .
[0107] Step 4.3, Minimal Error Judgment Mechanism: If If it is less than the maximum error tolerance, then when the theoretical slot... If the theoretical slot is empty, then the frequency point of the frequency hopping will be recorded directly in the theoretical slot; if the theoretical slot is empty... The space is occupied. Initiate collision detection and compare the current phase error. Phase error with occupier history Only when If the current frequency hopping signal is overwritten and the occupant is removed, otherwise the current frequency hopping signal is discarded. If the current phase error... If the error is greater than or equal to the maximum error tolerance, the signal is directly identified as a "grid mismatch signal" and discarded, and is not recorded in the current UAV's absolute time slot grid. The minimum error judgment mechanism sets an upper limit for relative phase deviation tolerance that is linked to the frequency hopping period. Specifically, the maximum error tolerance is set to 0.1 to 0.25 times the global absolute frequency hopping period. Conflict judgment is performed by comparing the relative errors of different signals in the slot, and the signal with the highest phase match is retained.
[0108] Step 4.4, Strictly Order-Preserving Trajectory Generation: Based on Theoretical Time Slot Indexing In ascending order, extract the frequency hopping points of non-empty slots in the entire panoramic mesh, and generate a time-logically continuous UAV frequency hopping center frequency basis vector. In this embodiment, the nominal frequency within each valid time slot is extracted in ascending order of the time slot index, and the frequency characteristic base direction is output. .
[0109] Step 5: One-dimensional adjacent channel clustering and measurement frequency standardization: such as Figure 5 As shown, the basis vectors are traversed, and each observed frequency point is replaced with the standard absolute frequency point with the smallest distance in the dictionary, outputting a set of standardized frequency hopping sequences; the aim is to construct an absolute physical channel dictionary and standardize the measurement frequency points.
[0110] Step 5.1, One-dimensional agglomerative clustering based on threshold distance: This is implemented using a one-dimensional neighbor distance threshold clustering algorithm. Specifically, if the UAV frequency hopping center frequency basis vector... The distance between adjacent frequency points arranged in time slot order is less than the judgment threshold. If they are similar, they are grouped into the same frequency cluster. The average frequency within the cluster is calculated, and an absolute physical standard channel dictionary vector is constructed. The threshold for one-dimensional agglomerative clustering is set at 0.4 to 0.6 times the ratio of the minimum nominal interval between adjacent physical channels, so that drift measurement frequency points within the same physical channel are merged and converged into the same physical channel cluster.
[0111] Step 5.2, Frequency Standardization: Traversing Basis Vectors Forcibly replace each observation frequency point with the nearest dictionary value. The standard absolute frequency point with the smallest distance from it , is a nominal frequency in the basis vectors. For the first in the dictionary A standard absolute frequency point, Indicates retrieving from the dictionary The standard absolute frequency point with the smallest mid-range distance outputs an absolutely pure set of standardized frequency hopping sequences. In this embodiment, the basis vectors are... The measured frequency values are rigidly projected onto the dictionary nominal values, and the output is an absolutely normalized denoised frequency hopping sequence. .
[0112] Step 6: Multi-dimensional continuous data mining and extraction of discontinuous causal patterns: such as Figure 6 As shown, the standardized frequency hopping sequence set is sliced according to a fixed observation time window length to construct a projection sequence database. The projection sequence database is processed based on the sequence pattern mining algorithm to obtain a set of frequency hopping patterns and perform complex quantification. Finally, through multi-segment joint extraction and large number voting purification, the core frequency hopping pattern sequence is output. The aim is to mine deep discontinuous time series patterns from the standardized frequency hopping sequence and obtain the distributed frequency hopping patterns.
[0113] Step 6.1: Construct a time window-based sequence projection database: Extract the long-span normalized frequency hopping sequence set. Based on a fixed observation time window length Slices are constructed into a projection sequence database. Each segment (subsequence) Corresponding to an independent frequency hopping sequence action window, For the first A segment, This represents the total number of segments. In this embodiment, the observation time window length of the sliding slice is... Set as This ensures that each subsequence segment contains at least 20 complete cycles in time slots. When constructing the projected sequence database, the observation time window length used for slicing is determined based on the frequency hopping cycle and the minimum cycle step size contained in the frequency hopping sequence. Specifically, the window length is taken as 1.2 to 2 times the minimum cycle step size, while ensuring that the window contains at least 3 to 6 frequency hopping cycles. The maximum value of the two is taken as the final slicing window, ensuring that each subsequence covers a complete regular state cycle in the time domain.
[0114] Step 6.2: Extraction of non-continuous hidden causal patterns based on the PrefixSpan (sequence pattern mining) algorithm: Addressing the weakness of traditional continuous frequency hopping pattern mining, which is easily truncated by random frequency hopping interference, this embodiment utilizes the PrefixSpan algorithm to find hidden frequency hopping patterns with cross-window temporal correlation. A minimum support threshold is set. The algorithm first scans the projection sequence database. Obtain frequent single-frequency points (i.e., initial prefixes of length 1). Next, for each prefix... Construct its corresponding projection database By recursively calling, the prefix is continuously... Add new associated frequency points Generate new prefix And calculate its support:
[0115]
[0116] in, For the new prefix Support For subsequence inclusion relationships, This represents the size of the projection sequence database. The algorithm, through a prefix space projection mechanism, effectively extracts sequences of the form "frequency point A" within polynomial time. (Any intermediate jump point) Discontinuous causal chain at frequency point B The final output is a set of frequency hopping patterns.
[0117] The minimum support threshold in the PrefixSpan sequence pattern mining algorithm is set based on the signal-to-noise ratio (SNR) and probability distribution in non-cooperative reception scenarios. Specifically, the lower the SNR, the higher the probability of missed signal detection, and the lower the proportion of samples that can be observed for the true pattern, requiring a reduction in the minimum support threshold. The higher the signal-to-noise ratio, the lower the false negative rate, which requires improvement. To suppress noise pseudo-patterns, low-frequency irregular frequency hopping states are filtered out, and frequent sequence patterns are extracted. In this embodiment, the minimum support threshold for PrefixSpan sequence pattern mining is... Set to 0.15.
[0118] Step 6.3, Sequence Complexity Measurement Evaluation Based on Lempel-Ziv: Evaluation of Normalized Frequency Hopping Sequence Sets Perform non-overlapping substring parsing. Let the total length of the sequence be... The size of the frequency dictionary used is The total number of independent substrings parsed from the statistical sequence throughout history. Calculate its normalized LZ complexity score. :
[0119]
[0120] This indicator is used as a macroscopic qualitative criterion to determine the level of chaotic randomness in a communication protocol. In this embodiment, the total sequence analysis length is set. The size of the channel dictionary used The estimated normalized LZ complexity score is calculated. .
[0121] Step 6.4, Frequency hopping pattern mutation denoising and frequency hopping sequence distribution based on Levenshtein fault tolerance: For the frequency hopping pattern detected by the PrefixSpan algorithm Considering the local frequency errors caused by channel fading, a biological sequence alignment concept is introduced. This involves calculating arbitrary candidate observation substrings. With frequency hopping pattern Levenshtein edit distance between .when hour, For permissible single-frequency replacement tolerance, if the target is identified as the same, the current frequency hopping pattern is retained; otherwise, the current frequency hopping pattern is discarded. The single-frequency replacement tolerance threshold used for Levenshtein editing distance fault tolerance determination is adaptively determined based on the fading characteristics of the transmission channel and the probability of occasional missed detections. Specifically: the single-symbol replacement error probability is obtained based on the received signal-to-noise ratio and channel fading characteristics. Combined with the comparison sequence length, the single-frequency replacement tolerance is calculated using a binomial distribution statistical model, and upper and lower limits are clamped. Simultaneously, the deletion tolerance is calculated using the missed detection probability to compare and capture sudden changes in the same target frequency hopping pattern caused by channel mutations. In this embodiment, the single-frequency replacement tolerance for Levenshtein editing distance fault tolerance determination... Set to 2.
[0122] Finally, a high-confidence core frequency hopping pattern sequence is obtained through multi-segment joint extraction and large-number voting purification. Specifically, steps 1-6 are performed on the collected UAV radio frequency time domain signals to obtain multiple frequency hopping patterns. All frequency hopping patterns are clustered, and then a large-number voting mechanism is used for hierarchical purification: first, the proportion of candidate frequency hopping patterns appearing across multiple observation segments is statistically analyzed at the cluster level to eliminate false pattern clusters generated by noise; then, within each cluster, the symbols of each time-series frequency hopping pattern are voted on to filter out noise and erroneous symbols, compensate for missed detections, and obtain a high-confidence complete core frequency hopping pattern.
[0123] In another embodiment, the present invention also provides a blind sorting system for UAV remote control signals based on time-slot grid phase-locked loop, namely a computer device including a processor and a memory storing a plurality of computer instructions, wherein the computer instructions are executed by the processor to implement the steps of the blind sorting method for UAV remote control signals based on time-slot grid phase-locked loop.
[0124] The memory and processor are electrically connected directly or indirectly to enable data transmission or interaction. For example, these components can be electrically connected to each other via one or more communication buses or signal lines. The memory stores a computer program that can run on the processor, and the processor implements the method of the present invention by running the computer program stored in the memory.
[0125] The memory can be, but is not limited to, Random Access Memory (RAM), Read Only Memory (ROM), Programmable Read-Only Memory (PROM), Erasable Programmable Read-Only Memory (EPROM), and Electrically Erasable Programmable Read-Only Memory (EEPROM). The memory stores the program, which the processor executes upon receiving execution instructions.
[0126] A processor can be an integrated circuit chip with data processing capabilities. The processor can be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), etc. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this invention.
[0127] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this invention.
[0128] The above embodiments merely illustrate several implementation methods of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of the present invention should be determined by the appended claims.
Claims
1. A blind sorting method for UAV remote control signals based on time-slotted grid phase-locked loop, characterized in that, include: Based on the YOLO target detection model, the collected UAV radio frequency time domain signals are used to construct an initial frequency hopping signal set; The initial frequency hopping signal is purified and denoised using the target morphological feature benchmark to obtain the UAV frequency hopping signal pool; The start times of all frequency hopping signals in the UAV frequency hopping signal pool are extracted and constructed into a time series. The first-order time difference sequence of all adjacent frequency hopping signals is calculated, and the global frequency hopping period is estimated by finding the peak through dynamic adaptive histogram. The absolute starting point of the grid synchronization to be calibrated is determined based on the time series and global frequency hopping period. The frequency hopping signal in the UAV frequency hopping signal pool is mapped to the grid slot. The frequency points of non-empty slots in the entire panoramic grid are extracted in ascending order of time slot index to generate the basis vector of the UAV frequency hopping center frequency that is logically continuous in time. Traverse the basis vectors, replace each observed frequency point with the standard absolute frequency point with the smallest distance in the dictionary, and output a set of standardized frequency hopping sequences; The standardized frequency hopping sequence set is sliced according to a fixed observation time window length to construct a projection sequence database. The projection sequence database is processed based on the sequence pattern mining algorithm to obtain a set of frequency hopping patterns and perform complex quantification. Finally, the core frequency hopping pattern sequence is output through multi-segment joint extraction and large number voting purification.
2. The blind sorting method for UAV remote control signals based on time-slotted grid phase-locked loop as described in claim 1, characterized in that, The YOLO-based target detection model constructs an initial frequency-hopping signal set from the collected UAV radio frequency time-domain signals, including: Convert the UAV radio frequency time domain signal into a two-dimensional energy power spectrum; The two-dimensional energy power spectrum is generated into a two-dimensional time spectrum image by means of amplitude logarithmic compression and normalization operations, as well as resolution resampling and channel expansion operations. The two-dimensional time-spectrum image is input into a pre-trained YOLO object detection model to predict the class probability, confidence level, and pixel-level bounding box coordinates of the frequency-hopping signal target box. The feature parameters of each frequency hopping signal are reconstructed based on the predicted pixel-level bounding box coordinates to obtain the initial set of frequency hopping signals.
3. The blind sorting method for UAV remote control signals based on time-slot grid phase-locked loop according to claim 2, characterized in that, The process involves reconstructing the feature parameters of each frequency-hopping signal based on the predicted pixel-level bounding box coordinates to obtain an initial set of frequency-hopping signals, including: Take the first pixel-level bounding box coordinates ,in For the first The minimum value of the bounding box coordinates at the pixel level on the x-axis. For the first The maximum value of the bounding box coordinates on the x-axis at the pixel level. For the first The minimum value of the bounding box coordinates at the pixel level on the y-axis. For the first The maximum value of the bounding box coordinates at the pixel level on the y-axis; Reconstructing the start time of the frequency hopping signal With end time ,as follows: in, The total pixel width of the two-dimensional time-spectral image. This represents the absolute time span corresponding to the two-dimensional time spectrum image. The starting time corresponding to the left boundary of the two-dimensional time-spectrum image; Reconstructing the center frequency With physical bandwidth ,as follows: in, The total pixel height of the two-dimensional time-spectral image. The total analysis bandwidth of the two-dimensional time-spectral image. The starting physical frequency at the bottom of the spectrum; Each reconstructed frequency-hopping signal feature parameter is expressed as a four-dimensional vector. By combining the characteristic parameters of all frequency hopping signals, the initial set of frequency hopping signals is obtained. , This represents the total number of frequency-hopping signals.
4. The blind sorting method for UAV remote control signals based on time-slotted grid phase-locked loop as described in claim 1, characterized in that, The process of refining and denoising the initial frequency-hopping signal using target morphological features to obtain the UAV frequency-hopping signal pool includes: Based on the termination time and start time in the characteristic parameters of the frequency hopping signal, the actual dwell time span of each frequency hopping signal can be obtained; If the actual dwell time span of the frequency hopping signal is less than or equal to 0, or the physical bandwidth of the frequency hopping signal is less than or equal to 0, or the physical bandwidth of the frequency hopping signal is greater than the upper limit threshold of the environmental bandwidth, then the corresponding frequency hopping signal is removed to complete the coarse screening of the frequency hopping signal. For the frequency-hopping signals remaining after coarse screening, the statistical median of the physical bandwidth and the actual dwell time span is calculated and used as the bandwidth benchmark value. and the baseline value of stay ; After traversing the remaining frequency-hopping signals after the initial screening, only those satisfying the joint logic condition are retained. and The frequency hopping signal is used to obtain the UAV frequency hopping signal pool, in which For the first The physical bandwidth of a frequency hopping signal To provide a tolerance threshold, For the first The actual dwell time span of a frequency-hopping signal This is the dwell time tolerance threshold.
5. The blind sorting method for UAV remote control signals based on time-slotted grid phase-locked loop as described in claim 1, characterized in that, The method of estimating the global frequency hopping period by finding peaks using a dynamic adaptive histogram includes: Remove the difference values smaller than the limit protection interval from the first-order time difference sequence to form an effective jump difference set; The histogram group interval is dynamically set based on the maximum and minimum values of the effective jump difference set, and the maximum value indexing algorithm is executed to find the first statistical peak interval containing the most difference samples. Extract all difference values that fall within the first statistical peak interval from the effective hopping difference set, and calculate the arithmetic mean of the extracted difference values as the global frequency hopping period of the UAV transmitter.
6. The blind sorting method for UAV remote control signals based on time-slotted grid phase-locked loop according to claim 1, characterized in that, The process of determining the absolute starting point of the calibrated grid synchronization based on time series and global frequency hopping period includes: Select the first valid point in the time series As the initial anchor point; Calculate the frequency hopping signal relative to the global frequency hopping cycle. fractional phase residual sequence , Indicates the first The start time of a frequency hopping signal This indicates that a modulo-1 operation will be performed; For fractional phase residual sequences Frequency hopping signals implement folding correction mechanism ; Calculate the median of all fractional phase residual sequences after folding correction. Obtain the absolute starting point of the calibrated grid synchronization. .
7. The blind sorting method for UAV remote control signals based on time-slotted grid phase-locked loop according to claim 1, characterized in that, The step of mapping the frequency hopping signals in the UAV frequency hopping signal pool to the grid slots includes: Based on the calibrated absolute starting point of grid synchronization, the theoretical time slot index into which the frequency-hopping signal falls in the UAV frequency-hopping signal pool is calculated. And calculate the relative phase error magnitude. ; When the relative phase error amplitude When the theoretical time slot index is less than the maximum error tolerance, If the corresponding slot is empty, then directly record the current frequency hopping signal; if the theoretical time slot index... If the corresponding slot is already occupied, a collision check is initiated. The collision check includes comparing the current relative phase error magnitude. The relative phase error magnitude with respect to the occupant's history Only when When the current frequency hopping signal is overwritten, the occupying frequency hopping signal is removed; otherwise, the current frequency hopping signal is discarded. When the relative phase error amplitude If the error is greater than or equal to the maximum error tolerance, discard the current frequency hopping signal.
8. The blind sorting method for UAV remote control signals based on time-slot grid phase-locked loop according to claim 1, characterized in that, Complex quantization of the frequency hopping pattern set includes: Perform non-overlapping substring parsing on the standardized frequency hopping sequence set, and count the total number of historical independent substrings parsed from the standardized frequency hopping sequence set; For each frequency hopping pattern in the set of frequency hopping patterns, calculate the Levenshtein edit distance between any independent substring and the frequency hopping pattern. If there is a Levenshtein edit distance less than or equal to the single frequency point replacement tolerance, retain the current frequency hopping pattern; otherwise, discard the current frequency hopping pattern.
9. A blind sorting system for UAV remote control signals based on time-slot grid phase-locked loop, comprising a processor and a memory storing a plurality of computer instructions, characterized in that, When the computer instructions are executed by the processor, they implement the steps of the blind sorting method for UAV remote control signals based on time slot grid phase-locked loop as described in any one of claims 1 to 8.
Citation Information
Patent Citations
Parameter estimation method for frequency hopping signals in frequency hopping non-cooperative communications
CN107612587A
