High frequency radar networking detection method, system and device based on non-cooperative emitter signal blind extraction

Abstract

The application discloses a high-frequency radar networking detection method, system and device based on non-cooperative emission source signal blind extraction, and belongs to the technical field of high-frequency radar detection. The method collects multi-radiation source aliasing signals in the adjacent area of the non-cooperative emission source, screens a target frequency band through cross-frame frequency band association and pulse interval structure, and obtains a center frequency, a bandwidth and a pulse arrival time sequence; determines an actual pulse boundary based on a pulse timing model, generates a reference IQ sequence through cross-correlation screening, time and phase alignment and coherent synthesis; updates the reference IQ version according to emission parameter changes, and determines the current effective reference IQ sequence according to the direct path matching result by a receiving station; and performs matching filtering and non-uniform slow-time Doppler processing by using the effective reference IQ sequence, and obtains a target detection result. The application can reduce the reference signal mismatch caused by multi-source aliasing, non-constant pulse timing and dynamic parameter changes, and improve the stability of networking detection.

Description

Technical Field

[0001] This invention relates to the field of network detection technology, specifically to a high-frequency radar network detection method, system, and device based on blind extraction of signals from non-cooperative transmitters. Background Technology

[0002] High-frequency (HF, 3-30MHz) radar utilizes skywaves formed by ground waves or ionospheric reflections for long-range detection, featuring long detection range and wide coverage, making it important for applications in marine surveillance and long-range early warning. By using existing non-cooperative high-power transmitters as illumination sources and receiving target-scattered echoes through off-site receiving stations, bistatic or multistatic passive detection systems can be constructed, reducing the need for high-power transmitter equipment. Existing passive radars typically acquire the direct wave from the non-cooperative transmitter through a reference channel and use it as a reference signal for target echo matching. In high-frequency long-distance propagation environments, the direct wave is susceptible to propagation attenuation, ionospheric multipath propagation, and environmental interference, leading to a decrease in the signal-to-noise ratio and waveform stability of the reference signal. Placing the reference signal acquisition equipment near the non-cooperative transmitter can improve the received signal strength; however, the electromagnetic environment in the high-frequency band is dense, and the acquired data may often contain signals from other radars, shortwave communications, and adjacent-channel radiation sources.

[0003] For this type of multi-source aliasing signal, when using single-frame spectrum analysis and fixed spectrum thresholds for signal parameter estimation, strong adjacent-channel signals and spectral leakage caused by finite sampling can easily alter the spectral boundaries of the target signal, causing shifts in the center frequency and bandwidth estimations, and even mixing other radiation source components into the reference signal. This reduces the matching degree between the reference signal and the target echo. Non-cooperative radar transmission signals may employ different pulse repetition intervals, such as fixed, jittery, or staggered intervals. For jittery or staggered pulse sequences, directly estimating the single pulse repetition period based on the continuous envelope autocorrelation peaks is easily affected by harmonic peaks, spurious peaks, and pulse interval variations, causing pulse boundary judgment errors and resulting in cross-pulse or incomplete waveforms in the intercepted reference signal. Furthermore, parameters such as the center frequency, bandwidth, and pulse interval of non-cooperative transmitters may change with operating conditions. The reference signal generated based on historical data may no longer match the current actual transmitted waveform, thereby reducing the matched filter gain of the remote receiving station. When there are jittery or staggered changes in the pulse repetition interval, the actual positions of each pulse in the slow time dimension are not strictly equal. If conventional slow-time Doppler processing is still performed according to a fixed pulse repetition period, it is easy to cause Doppler response broadening, shift, or coherence accumulation loss. Summary of the Invention

[0004] The purpose of this invention is to provide a high-frequency radar network detection method, system, and device based on blind extraction of signals from non-cooperative transmitters, thereby solving the problems existing in the prior art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a high-frequency radar network detection method based on blind extraction of signals from non-cooperative transmitters, comprising: S1. Collect a first sampling sequence containing signals from multiple radiation sources in the vicinity of a non-cooperative emission source; S2. Perform frequency domain transformation on the first sampling sequence in frames, and form candidate frequency band trajectories based on the overlap relationship of the frequency bands occupied by adjacent frames; perform envelope pulse detection on the sub-band sequences corresponding to each candidate frequency band, and filter the target frequency band trajectory based on the consistency of the pulse interval structure to obtain the center frequency, bandwidth and pulse arrival time sequence. S3. Establish a pulse timing model based on the pulse arrival time sequence and determine candidate arrival time windows. Detect the actual pulse boundary within the time window. Extract candidate reference sequences according to the actual pulse boundary. After cross-correlation screening, time and phase alignment, and coherent synthesis, perform down-conversion and decimation filtering based on the center frequency and bandwidth to obtain the reference IQ sequence. S4. Determine the position of parameter state change based on the changes in center frequency, bandwidth and pulse interval characteristics within the continuous time window, and update the reference IQ sequence accordingly to form a set of reference IQ sequences with version identifiers and corresponding signal parameters. S5. Send the reference IQ sequence set to the receiving station, perform baseband conversion on the second sampling sequence of the receiving station according to the corresponding signal parameters of each version, and perform matched filtering using the reference IQ sequences of each version. Determine the current valid reference IQ sequence and the corresponding receiving baseband IQ sequence based on the amplitude of the matching peak of the direct path and the consistency of the cross-pulse position. S6. Perform matched filtering on the received baseband IQ sequence using the current valid reference IQ sequence, perform non-uniform slow-time Doppler transformation based on the pulse arrival time sequence, and obtain the target detection result based on the delay difference of the target response relative to the matching peak of the direct path and the Doppler position.

[0006] Specifically, S1 includes: The 3-30MHz high-frequency signal received in the vicinity of the non-cooperative transmitter is divided into multiple sub-band signals by a filter bank; Each sub-band signal is synchronously sampled based on a unified sampling clock and divided into frames according to time to obtain multiple sub-band sampling sequences; Calculate the frame power of each sub-band sampling sequence, construct the channel power feature vector, determine the acquisition threshold based on the background power statistics, and filter the effective sub-bands; Based on the pre-obtained channel delay calibration parameters for each effective sub-band, the sampling position of each sub-band sampling sequence is compensated, and the corresponding channel identifier is associated; The multiple sub-band sampling sequences after compensation are established according to the sampling time and stored continuously to form a first sampling sequence that contains signals from multiple radiation sources and retains the sub-band correspondence.

[0007] Specifically, S2 includes: The first sampled sequence is divided into frames and windowed frequency domain transform is performed to obtain the time-frequency power matrix; The spectral threshold is determined based on the spectral power and background statistics of each frame, and continuous over-threshold frequency units are extracted to form candidate occupied frequency bands. The candidate frequency band trajectory is formed by associating the frequency overlap rate and center frequency deviation of candidate occupied frequency bands in adjacent frames. Band-limited extraction is performed on the data corresponding to each candidate frequency band trajectory to obtain sub-band sequences, and pulse arrival times are detected based on the adaptive threshold of their envelope sequences. A pulse interval sequence is formed based on the arrival times of adjacent pulses. A time sequence consistency feature value is constructed based on the clustering characteristics of the pulse interval values ​​and the repetition of the interval arrangement pattern. Target frequency band trajectories that meet the consistency threshold are selected, and the center frequency, bandwidth, and pulse arrival time sequence are determined accordingly.

[0008] Specifically, S3 includes: The pulse arrival time series is subtracted to obtain the pulse interval series, and a pulse time series model is established based on the distribution characteristics and arrangement relationship of the pulse intervals; The arrival time window of candidate pulses is predicted using the pulse timing model, and the actual boundary of each candidate pulse is determined based on the envelope adaptive threshold. Candidate reference sequences are truncated according to the actual boundary, the normalized cross-correlation between candidate reference sequences is calculated, reference sequences are filtered according to the correlation threshold, and time-domain alignment and phase compensation are performed according to the cross-correlation peak position and cross-correlation peak complex phase, respectively. Coherent synthesis is performed on the time- and phase-aligned reference sequences to obtain the reference signal sequence; The reference signal sequence is orthogonally downconverted based on the center frequency, and the decimation parameters are determined based on the bandwidth. The reference IQ sequence is then obtained through decimation filtering.

[0009] Specifically, S4 includes: Extract the distribution and arrangement features of center frequency, bandwidth and pulse interval within a continuous sliding time window, and construct a parameter state feature vector; Calculate the degree of state change between the parameter state feature vectors corresponding to adjacent time windows, and determine the position of parameter state change based on the preset change threshold; The parameter stability intervals are divided according to the positions of adjacent parameter state changes, and a set of state parameters including center frequency, bandwidth, and pulse interval distribution and arrangement parameters are determined for each parameter stability interval; The reference signal is re-triggered and processed based on the state parameter set corresponding to the stable interval of each parameter, an updated reference IQ sequence is generated, and the corresponding baseband sampling rate is recorded. The updated reference IQ sequence is associated with the corresponding state parameter set and parameter stability interval, and a version identifier is assigned to form a set of reference IQ sequences arranged in the order of parameter state changes.

[0010] Specifically, S5 includes: A set of reference IQ sequences with version identifiers and associated center frequency, bandwidth and baseband sampling rate is sent to the receiving station, and a second sampling sequence containing direct components and target echoes is acquired. Based on the center frequency, bandwidth and baseband sampling rate of each version, the second sampling sequence is subjected to quadrature downconversion and decimation filtering to obtain the second baseband IQ sequence corresponding to the reference IQ sequence of each version. Matched filter coefficients were constructed using the conjugate inversion sequences of the reference IQ sequences for each version, and matched filtering was performed on the corresponding second baseband IQ sequences to obtain the matched response sequences. Extract the direct path matching peaks of continuous pulses from each matching response sequence, and generate version consistency evaluation values ​​based on the peak amplitude to background response ratio and cross-pulse position dispersion. The reference IQ sequence corresponding to the version with the highest version consistency evaluation value that meets the preset threshold is determined as the current valid reference IQ sequence, and the corresponding second baseband IQ sequence is determined as the current received baseband IQ sequence.

[0011] Specifically, S6 includes: Based on the current valid reference IQ sequence, the matched filter coefficients are constructed, and the continuous pulses in the current received baseband IQ sequence are matched and filtered to obtain multiple matched response sequences. The actual slow time position corresponding to each matched response sequence is determined based on the pulse arrival time sequence, and the delay-non-uniform slow time response data is organized according to the propagation delay dimension and the actual slow time position. For each propagation delay unit, a non-uniform Doppler transformation is performed on the delay-non-uniform slow-time response data based on the corresponding actual slow-time position to obtain the range-Doppler response matrix; The target detection threshold is determined based on the background response statistics of the distance-Doppler response matrix, and target response units that exceed the target detection threshold are filtered out. Based on the delay difference of the target response unit relative to the matching peak of the direct path and its Doppler position, the target propagation distance difference information and target Doppler information are determined, and the target detection results are obtained.

[0012] The present invention also provides a system applicable to the high-frequency radar network detection method based on blind extraction of signals from non-cooperative transmitters as described above, comprising: The front-end acquisition device is deployed in the vicinity of the non-cooperative emission source to acquire a first sampling sequence containing signals from multiple radiation sources. The first sampling sequence is processed in a frame-by-frame frequency domain. Candidate frequency band trajectories are formed based on the overlap relationship of the frequency bands occupied by adjacent frames. The target frequency band trajectory is selected based on the pulse interval structure of the sub-band sequence corresponding to each candidate frequency band to obtain the center frequency, bandwidth and pulse arrival time sequence. The front-end acquisition device is also used to establish a pulse timing model based on the pulse arrival time sequence, extract candidate reference sequences according to the detected actual pulse boundaries, perform cross-correlation screening, time and phase alignment and coherent synthesis on the candidate reference sequences, and perform down-conversion and decimation filtering to obtain a reference IQ sequence; update the reference IQ sequence according to the changes in signal parameters within a continuous time window to form a set of reference IQ sequences with version identifiers and corresponding signal parameters; A communication link for transmitting the reference IQ sequence set to the receiving station; The receiving station is used to acquire a second sampling sequence containing the direct component and the target echo. It performs baseband conversion on the second sampling sequence according to the corresponding signal parameters of each version, and performs matched filtering using the reference IQ sequence of each version. It determines the current valid reference IQ sequence and the corresponding received baseband IQ sequence based on the amplitude of the matching peak of the direct path and the consistency of the cross-pulse position. The receiving station is also used to perform matched filtering on the received baseband IQ sequence using the currently valid reference IQ sequence, perform non-uniform slow-time Doppler transformation according to the pulse arrival time sequence, and obtain the target detection result based on the delay difference of the target response relative to the matching peak of the direct path and the Doppler position.

[0013] Furthermore, the present invention also provides an apparatus suitable for the high-frequency radar network detection method based on blind extraction of signals from non-cooperative transmitters as described above, comprising: The sampling module is used to acquire the first sampling sequence of signals from multiple radiation sources in the vicinity of a non-cooperative emission source; The frequency band filtering module is used to perform frequency domain transformation on the first sampling sequence in frames, form candidate frequency band trajectories according to the overlap relationship of the frequency bands occupied by adjacent frames, and filter the target frequency band trajectory according to the pulse interval structure of the sub-band sequence corresponding to the candidate frequency band to obtain the center frequency, bandwidth and pulse arrival time sequence. The reference generation module is used to establish a pulse timing model based on the pulse arrival time sequence and detect the actual pulse boundary. It then extracts a candidate reference sequence according to the actual pulse boundary and obtains a reference IQ sequence after cross-correlation screening, time and phase alignment, coherent synthesis, down-conversion and decimation filtering. The version update module is used to determine the position of parameter state change based on the changes in center frequency, bandwidth and pulse interval characteristics within a continuous time window, and update the reference IQ sequence to obtain a set of reference IQ sequences with version identifiers and corresponding signal parameters. The version confirmation module is used to perform baseband conversion on the second sampling sequence collected by the receiving station according to the signal parameters corresponding to each version, perform matched filtering using the reference IQ sequence of each version, and determine the current valid reference IQ sequence and the corresponding receiving baseband IQ sequence based on the amplitude of the matching peak of the direct path and the consistency of the cross-pulse position. The target detection module is used to perform matched filtering on the received baseband IQ sequence using the currently valid reference IQ sequence, perform non-uniform slow-time Doppler transformation according to the pulse arrival time sequence, and obtain the target detection result based on the delay difference of the target response relative to the matching peak of the direct path and the Doppler position.

[0014] As can be seen from the above technical solution, the present invention has the following beneficial effects: The method of this invention improves the stability of center frequency, bandwidth, and pulse timing parameter extraction in complex electromagnetic environments by performing cross-frame frequency band correlation on near-field multi-radiation source aliasing sampling data and screening target frequency bands in conjunction with pulse spacing structure. It establishes a pulse timing model based on pulse arrival time and extracts, filters, and coherently synthesizes reference signals according to actual pulse boundaries, reducing the impact of cross-pulse interception and waveform mismatch under jitter or uneven pulse conditions. Furthermore, it updates the reference IQ sequence according to changes in transmission parameter status, and the receiving station determines the current valid reference IQ sequence based on the direct path matching result, reducing reference version mismatch caused by dynamic parameter changes and transmission delays. Simultaneously, it performs non-uniform slow-time Doppler processing based on the actual pulse arrival time, reducing Doppler broadening and coherent accumulation loss caused by non-equal interval pulse sampling, thereby improving the reliability of reference signal matching and the consistency of target range and Doppler detection results in non-cooperative high-frequency radar networks. Attached Figure Description

[0015] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a timing diagram of the present invention; Figure 3 This is a time-frequency diagram of the experimental signal scene of this invention; Figure 4 This is a comparison chart of parameter extraction errors in this invention; Figure 5 This is a graph showing the coherent accumulation output gain of the present invention as a function of pulse number. Figure 6 This is a Doppler spectrum comparison diagram of the present invention. Detailed Implementation

[0016] 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.

[0017] Example 1: like Figure 1 and Figure 2 As shown, this invention provides a technical solution: a high-frequency radar network detection method based on blind extraction of signals from non-cooperative transmitters, comprising: S1. Collect a first sampling sequence containing signals from multiple radiation sources in the vicinity of a non-cooperative emission source; S2. Perform frequency domain transformation on the first sampling sequence in frames, and form candidate frequency band trajectories based on the overlap relationship of the frequency bands occupied by adjacent frames; perform envelope pulse detection on the sub-band sequences corresponding to each candidate frequency band, and filter the target frequency band trajectory based on the consistency of the pulse interval structure to obtain the center frequency, bandwidth and pulse arrival time sequence. S3. Establish a pulse timing model based on the pulse arrival time sequence and determine candidate arrival time windows. Detect the actual pulse boundary within the time window. Extract candidate reference sequences according to the actual pulse boundary. After cross-correlation screening, time and phase alignment, and coherent synthesis, perform down-conversion and decimation filtering based on the center frequency and bandwidth to obtain the reference IQ sequence. S4. Determine the position of parameter state change based on the changes in center frequency, bandwidth and pulse interval characteristics within the continuous time window, and update the reference IQ sequence accordingly to form a set of reference IQ sequences with version identifiers and corresponding signal parameters. S5. Send the reference IQ sequence set to the receiving station, perform baseband conversion on the second sampling sequence of the receiving station according to the corresponding signal parameters of each version, and perform matched filtering using the reference IQ sequences of each version. Determine the current valid reference IQ sequence and the corresponding receiving baseband IQ sequence based on the amplitude of the matching peak of the direct path and the consistency of the cross-pulse position. S6. Perform matched filtering on the received baseband IQ sequence using the current valid reference IQ sequence, perform non-uniform slow-time Doppler transformation based on the pulse arrival time sequence, and obtain the target detection result based on the delay difference of the target response relative to the matching peak of the direct path and the Doppler position.

[0018] In the above scheme, a reference signal acquisition terminal is set up 3 kilometers away from the non-cooperative emission source. The reference signal acquisition terminal includes a receiving antenna, a radio frequency front-end, an analog-to-digital converter (ADC), and a digital signal processing module. The monitoring frequency range of the radio frequency front-end is set to 9 MHz to 11 MHz. The ADC continuously acquires data for 60 seconds at a sampling rate of 4 million complex samples per second, forming the first sampling sequence. The first sampling sequence contains the target non-cooperative emission source signal, the same-channel radiated signal, and the adjacent-channel radiated signal. The digital signal processing module sequentially performs frequency band trajectory construction, pulse detection, pulse timing identification, reference waveform extraction, reference sequence update, and version management.

[0019] In step one, the receiving antenna receives spatial electromagnetic signals in the frequency range of 9 MHz to 11 MHz. The radio frequency front-end performs out-of-band filtering and fixed-gain amplification on the received signal, and sends the processed signal to the analog-to-digital converter (ADC). The ADC continuously samples at 4 million complex sampling points per second for 60 seconds. During the sampling process, the amplification gain of the radio frequency front-end remains constant, and the digital signal processing module saves all complex sampling points in the order of sampling time, thus forming the first sampling sequence.

[0020] In step two, the first sampling sequence is divided into frames of 4096 sampling points each, with 2048 identical sampling points retained between adjacent frames. Therefore, each subsequent frame is shifted 2048 sampling points relative to the previous frame. Each frame is first multiplied by a Hanning window of length 4096 points, and then a 4096-point Fast Fourier Transform is performed. Since the sampling rate is 4 million sampling points per second, the frequency interval between two adjacent frequency units is 976.5625 Hz.

[0021] The power of all frequency units in each frame is calculated, using the full-scale power of the analog-to-digital converter as a reference. When the power of a frequency unit reaches -70 dB relative to its full-scale power, that frequency unit is marked as occupied. When three consecutive frequency units reach or exceed -70 dB, these three, and subsequent consecutively satisfying frequency units, are merged into one occupied frequency band. The left boundary of the occupied frequency band is taken as the center frequency of the first occupied frequency unit, and the right boundary is taken as the center frequency of the last occupied frequency unit.

[0022] After identifying the occupied frequency bands in a single frame, the occupied frequency bands of two adjacent frames are correlated. Specifically, the frequency range jointly covered by the occupied frequency bands of the previous and subsequent frames is first obtained, and then the width of this common frequency range is calculated. Subsequently, the widths of the two occupied frequency bands are calculated separately, and the smaller value is selected from the two bandwidths. The width of the common frequency range is then divided by this smaller bandwidth to obtain the frequency band overlap coefficient.

[0023] When the frequency band overlap coefficient reaches 0.60, and the difference in center frequency between two adjacent occupied frequency bands does not exceed 1953.125 Hz, these two occupied frequency bands are connected into the same frequency band trajectory. When the above connection conditions are met for 8 consecutive frames, the corresponding trajectory is registered as a candidate frequency band trajectory. If the above connection conditions are not met in any frame, the current trajectory is terminated before that frame, and cross-frame interpolation is not performed.

[0024] For each candidate frequency band trajectory, the center frequency of all valid frames of that trajectory is calculated. Specifically, the left and right boundary frequencies of each valid frame are first added together and then divided by 2 to obtain the center frequency of that frame. Then, the center frequencies of all valid frames are added together and divided by the total number of valid frames to obtain the center frequency of the candidate frequency band trajectory. The bandwidth of the candidate frequency band trajectory is processed in the same way: the difference between the right and left boundary frequencies of each valid frame is calculated, and then the bandwidths of all valid frames are added together and divided by the total number of valid frames.

[0025] Using the center frequency of the candidate frequency band trajectory as the down-conversion frequency, the corresponding frequency band is moved to the zero-frequency position. A 129th-order finite impulse response filter is configured for the sub-band signal. The total passband width of the filter is the value obtained by multiplying the candidate frequency band trajectory bandwidth by 1.20, and the initial total stopband width is the value obtained by multiplying the candidate frequency band trajectory bandwidth by 1.40.

[0026] The envelope power is calculated point-by-point for each sample of the sub-band sequence. Each complex sample point includes an in-phase component and a quadrature component. The squares of the in-phase component and the quadrature component are calculated separately, and then the two are added together. The result is taken as the envelope power of that sample point.

[0027] When the envelope power reaches or exceeds -55 dB relative to the full-scale power for eight consecutive sampling points, the first of these eight sampling points is designated as the candidate pulse start point. After the pulse starts, when the envelope power is below -55 dB for eight consecutive sampling points, the first of these eight sampling points is designated as the candidate pulse end point. The pulse duration is calculated based on the candidate pulse start and end points. Pulses with a duration less than 100 microseconds or greater than 20 milliseconds are directly deleted.

[0028] For a candidate frequency band trajectory, the start times of 33 valid pulses are obtained consecutively in chronological order. The first pulse interval is obtained by subtracting the start time of the first pulse from the start time of the second pulse; this process is repeated until 32 pulse intervals are obtained.

[0029] Add up all 32 pulse intervals and divide by 32 to get the average pulse interval. Then calculate the difference between each actual pulse interval and the average pulse interval, take the absolute value of the difference, and divide the absolute value by the average pulse interval to get the relative deviation of the corresponding pulse interval.

[0030] When the relative deviation of at least 30 pulse intervals out of 32 pulse intervals does not exceed 2%, the candidate frequency band trajectory is determined as the target frequency band trajectory; when the number of pulse intervals that meet this condition is less than 30, the candidate frequency band trajectory is deleted.

[0031] After the target frequency band trajectory is determined, 32 consecutive frames of spectrum containing the target frequency band are acquired. The power of the same frequency unit in each of the 32 frames is added together to form a cumulative spectrum. The consecutive frequency units in the cumulative spectrum that meet the specified detection conditions constitute the target effective frequency band. The leftmost frequency of the target effective frequency band is taken as the left boundary, and the rightmost frequency as the right boundary. The center frequency of the target signal is obtained by adding the left boundary frequency and dividing by 2; the bandwidth of the target signal is obtained by subtracting the left boundary frequency from the right boundary frequency. Simultaneously, the confirmed pulse start times are saved in chronological order to form a pulse arrival time sequence.

[0032] In step three, 33 pulse arrival times are continuously obtained from the confirmed pulse arrival time sequence. Based on the time difference between adjacent pulses, 32 pulse intervals are obtained. The 32 pulse intervals are added together and divided by 32 to obtain the average value of the current pulse interval.

[0033] The actual arrival time of the first pulse is used as the starting point of the time series model. When predicting the arrival time of the second pulse, an average pulse interval is added to the arrival time of the first pulse; when predicting the arrival time of the third pulse, two average pulse intervals are added to the arrival time of the first pulse; thereafter, the corresponding number of average pulse intervals are added according to the pulse sequence number, thus obtaining the predicted arrival time of each pulse.

[0034] A candidate arrival time window is set for each predicted arrival time. The starting position of the candidate arrival time window is set before the predicted arrival time, at a position equivalent to 5% of the average pulse interval, and the ending position is set after the predicted arrival time, at a position equivalent to 5% of the average pulse interval. Therefore, the total width of a candidate arrival time window is equal to 10% of the average pulse interval.

[0035] The envelope power is recalculated within the candidate arrival time window. When the envelope power reaches or exceeds -55 dB relative to the full-scale power for 8 consecutive sampling points, the first sampling point that meets the condition is determined as the actual pulse start boundary. When the envelope power of 8 consecutive sampling points after the pulse starts is lower than -55 dB, the first sampling point that meets the condition is determined as the actual pulse termination boundary.

[0036] Thirty-two candidate reference sequences are continuously extracted based on the actual pulse start and end boundaries. For each candidate reference sequence, the square of the amplitude of the complex sample value is calculated for each sampling point. The squares of the amplitudes of all sampling points in the sequence are summed to obtain the total energy of the candidate reference sequence. The total energies of the 32 candidate reference sequences are compared, and the candidate reference sequence with the largest total energy value is determined as the initial template.

[0037] Subsequently, the normalized cross-correlation between the remaining candidate reference sequences and the initial template was calculated. During the calculation, the candidate reference sequences to be compared were moved point by point according to the sampling points. For each moving sampling point, the complex sample values ​​at the corresponding positions within the overlap interval between the candidate reference sequence and the initial template were multiplied point by point. The complex sample values ​​of the initial template were first conjugated. The magnitude of the result was obtained by summing all the products.

[0038] Simultaneously, the total energy of the two overlapping sequences within the current overlap interval is calculated. The two total energies are multiplied, and the square root is taken. Then, the amplitude of the aforementioned correlation result is divided by this square root to obtain the normalized cross-correlation value corresponding to the current time shift position. After traversing all allowed time shift positions, the largest normalized cross-correlation value is selected as the cross-correlation peak of the candidate reference sequence.

[0039] If the peak value of the normalized cross-correlation of a candidate reference sequence reaches or exceeds 0.80, the candidate reference sequence is retained; if the peak value of the normalized cross-correlation is less than 0.80, the candidate reference sequence is deleted.

[0040] For the retained candidate reference sequences, time alignment is performed based on the sampling point positions where the normalized cross-correlation peaks occur. When the peak is exactly at an integer sampling point position, the integer sampling point is directly shifted according to the peak position. When the actual position of the peak is between two adjacent sampling points, the correlation values ​​of the peak sampling point and one sampling point before and after it are obtained. Quadratic interpolation is used to determine the fractional sampling position of the peak, and then a 21st-order fractional-delay finite-impulse response filter is used to compensate for the fractional sampling time difference.

[0041] After time alignment, phase alignment is performed. Specifically, the time-aligned candidate reference sequence is multiplied point-by-point by the corresponding complex sampling points of the initial template, with the initial template undergoing conjugation processing. All products are then summed. The phase difference between the candidate reference sequence and the initial template is determined based on the resulting complex result. Then, all complex sampling points of the candidate reference sequence are rotated in the opposite direction of this phase difference to ensure a unified phase reference between the candidate reference sequence and the initial template.

[0042] From the candidate reference sequences that meet the condition of a normalized cross-correlation peak value of 0.80, the first 16 sequences are obtained in chronological order for coherent synthesis. For the same sampling position, the complex sample values ​​corresponding to the 16 candidate reference sequences are added together and then divided by 16. The resulting complex value is used as the synthesis result for that position. The same process is performed on all sampling positions to obtain the coherently synthesized reference sequence.

[0043] If the number of candidate reference sequences that meet the normalized cross-correlation condition is less than 16, no reference sequence is generated. Subsequent pulses are received and the same correlation screening is performed until a total of 16 candidate reference sequences that meet the condition are obtained.

[0044] After coherent synthesis is completed, digital downconversion is performed according to the center frequency determined in step two. Specifically, a digital complex local oscillator with the same magnitude but opposite direction to the target center frequency is generated for each complex sampling point, and the reference sequence is multiplied by the digital local oscillator point by point to shift the center frequency of the target signal to zero.

[0045] After down-conversion, baseband filtering is performed. The filter passband extends from zero frequency towards both the positive and negative frequency directions to 55% of the target bandwidth, and the stopband starts at 70% of the target bandwidth. After filtering, decimation is performed.

[0046] The decimation factor is determined as follows: starting from an integer of 1, the decimation factor is increased incrementally. Each increase is performed by dividing the original sampling rate by the current decimation factor to obtain the decimated sampling rate. The decimated sampling rate must not be less than 2.5 times the target signal bandwidth. The largest integer decimation factor that meets this condition is selected as the final decimation factor. After decimation, a reference orthogonal baseband sequence is obtained.

[0047] In step four, 32 consecutive valid pulses are used as a parameter for statistical time window. After completing the first statistical time window, the time window is shifted forward by 16 pulses to form the second statistical time window. Subsequent statistical time windows are formed by shifting forward by 16 pulses each time.

[0048] For each parameter's statistical time window, the center frequency, bandwidth, and average pulse interval are recalculated. When comparing the center frequencies of two adjacent parameter statistical time windows, the center frequency of the later statistical time window is subtracted from the center frequency of the earlier statistical time window, and the absolute value of the difference is obtained. When this absolute value reaches or exceeds 5 kHz, the center frequency change condition is met.

[0049] When comparing the bandwidth of two adjacent statistical time windows, first calculate the difference between the bandwidth of the later statistical time window and the bandwidth of the earlier statistical time window, and take the absolute value. Then divide this absolute value by the bandwidth of the earlier statistical time window. When the result reaches or exceeds 10%, the bandwidth change condition is met.

[0050] When comparing the pulse interval characteristics of two adjacent statistical time windows, first calculate the difference between the average pulse interval of the later statistical time window and the average pulse interval of the earlier statistical time window, and take the absolute value. Then divide this absolute value by the average pulse interval of the earlier statistical time window. When the result reaches or exceeds 5%, the pulse interval change condition is met.

[0051] When any one of the following conditions—center frequency change, bandwidth change, or pulse interval change—is met for three consecutive statistical time windows, the starting pulse position of the first statistical time window among these three statistical time windows is determined as the parameter state change position.

[0052] Once the position of parameter state change is determined, candidate reference sequences are reacquired from that position onwards, and the actual pulse boundary detection, normalized cross-correlation screening, time alignment, phase alignment, coherent synthesis of 16 valid candidate reference sequences, digital downconversion, baseband filtering, and decimation are re-executed to generate a new reference orthogonal baseband sequence.

[0053] The first reference sequence is designated as version 001. Subsequently, the version number is incremented by 1 for each confirmed parameter status change, resulting in versions 002, 003, and subsequent versions. Each version stores the version number, center frequency, bandwidth, average pulse interval, reference sequence sampling rate, number of valid reference sequence sampling points, version start pulse number, and version generation time.

[0054] In step five, this embodiment sets up three receiving stations. The reference signal acquisition terminal sends the generated reference sequence version and the corresponding signal parameters for each version to the three receiving stations. Each receiving station obtains the second sampling sequence at a rate of 4 million complex sampling points per second.

[0055] For each reference sequence version, the receiving station first performs digital downconversion on the second sampled sequence according to the center frequency recorded in that version, and then performs baseband filtering and decimation according to the bandwidth recorded in that version, so that the sampling rate of the processed received baseband sequence is consistent with the sampling rate of the corresponding reference sequence.

[0056] The matched filter is then constructed. During construction, all complex sample points of the reference sequence are first conjugated, and then the temporal order of the sample points is completely reversed. The reversed sequence is used as the matched filter. The received baseband sequence is then sequentially multiplied and added with the matched filter using sliding multiplication to obtain the matched filter output corresponding to the reference version.

[0057] For each reference version, 16 pulses are processed consecutively. For each pulse, a direct path search interval is set within 20 milliseconds starting from the pulse arrival time. Within this search interval, the amplitudes of all matched filter sampling points are compared, and the sampling point with the largest amplitude is determined as the direct path matching peak of that pulse, and the location of the sampling point where the matching peak is located is recorded.

[0058] After processing 16 pulses, the maximum and minimum sampling point numbers are obtained from the matching peak positions of the 16 direct paths. The minimum sampling point number is subtracted from the maximum sampling point number. If the difference is no more than 2 sampling points, the reference version is deemed to meet the cross-pulse position consistency condition.

[0059] Simultaneously, for each pulse, 1024 matched filter sampling points are continuously obtained after the end position of the direct path search interval as the background detection interval. For these 1024 sampling points, the amplitude square is calculated point by point. The sum of the 1024 amplitude squares is divided by 1024, and the square root of the result is taken to obtain the background root mean square amplitude of the pulse.

[0060] Add the peak amplitude values ​​of the 16 direct paths and divide by 16 to obtain the average peak amplitude of the direct paths for this reference version; add the root mean square amplitude values ​​of the 16 background paths and divide by 16 to obtain the average background amplitude for this reference version. Then divide the average peak amplitude of the direct paths by the average background amplitude.

[0061] When the ratio reaches or exceeds 4, and the difference between the maximum and minimum positions among the 16 direct path matching peaks does not exceed 2 sampling points, the reference version is determined as a valid candidate version.

[0062] When only one version meets both of the above conditions, that version is directly designated as the current valid reference version. When two or more versions simultaneously meet the above conditions, the average peak amplitude of the direct path of each valid candidate version is compared, and the version with the largest value is designated as the current valid reference version. Simultaneously, the received baseband sequence formed according to the center frequency, bandwidth, and sampling rate of that version is determined as the current received baseband sequence.

[0063] In step six, the current received baseband sequence is matched and filtered using the reference sequence corresponding to the current valid reference version, and 64 valid pulses are selected continuously for slow-time processing.

[0064] For each distance-delay sampling position, the complex matched filtering results of 64 pulses at that position are obtained, and Doppler processing is strictly performed using the actual arrival time of each pulse, without rearranging the 64 pulses into an equally spaced time sequence.

[0065] The Doppler search frequency starts at -20 Hz and increases incrementally in 0.02 Hz increments up to +20 Hz. For each search frequency and each delay position, 64 pulses are processed. When processing a pulse, the current Doppler search frequency is first multiplied by the pulse's actual arrival time to obtain the cumulative phase cycles of the pulse relative to the current search frequency. This phase cycle number is then multiplied by 360 degrees, and the opposite direction is used as the phase compensation angle. The complex matched filter result of the pulse at the current delay position is then rotated according to this compensation angle. After phase compensation for 64 pulses, the 64 complex results are summed, and the resulting complex amplitude is used as the response intensity corresponding to the current delay position and the current Doppler search frequency. The same processing is performed for all delay positions and all search frequencies to form a two-dimensional delay and Doppler response result.

[0066] Two-dimensional object detection employs an average background power decision method with fixed training units. For each target object, eight training units are selected consecutively before the delay direction at the same Doppler position, skipping two guard units; similarly, eight training units are selected consecutively after the delay direction, skipping two guard units. Simultaneously, four training units are selected consecutively after the Doppler direction below the same delay position, and four training units are selected consecutively after the Doppler direction above the same delay position. Thus, each target object uses 24 training units.

[0067] Calculate the power of each of the 24 training units, sum the 24 power values, and divide by 24 to obtain the average background power. Multiply the average background power by 12 to obtain the detection threshold for the current unit to be detected. When the power of the unit to be detected reaches or exceeds the detection threshold, it is registered as a target candidate unit; when it is below the detection threshold, it is not registered as a target candidate unit.

[0068] Two-dimensional adjacency merging is performed on all target candidate elements. Two target candidate elements are grouped into the same connected region if they are directly adjacent in the delay direction, Doppler direction, or diagonal direction. In each connected region, only the target candidate element with the largest power value is retained and determined as a target response.

[0069] For a given target response, record the sampling point location of its matched filter peak, and simultaneously record the sampling point location of the direct path matching peak within the same processing cycle. Subtract the direct path matching peak sampling point number from the target response sampling point number, and then multiply by the single sampling point time interval of the current received baseband sequence to obtain the delay difference between the target response and the direct path matching peak.

[0070] Multiplying this delay difference by 299,792,458 meters per second yields the increase in propagation distance of the target scattering propagation path relative to the direct propagation path. The Doppler search frequency at which the target response occurs is directly taken as the Doppler position of the target.

[0071] Through the above processing, step one completes the acquisition of the first sampling sequence, step two completes the extraction of the target frequency band trajectory and pulse arrival time, step three completes the precise interception, filtering, time alignment, phase alignment and coherent synthesis of the reference sequence, step four completes the update of the reference sequence version according to the determined parameter change conditions, step five selects the current valid reference version according to the direct path matching result of the receiving station, and step six performs non-uniform slow time Doppler processing according to the actual pulse arrival time, and outputs the delay difference of the target relative to the direct path and the Doppler position, thereby realizing high-frequency radar network detection based on blind extraction of signals from non-cooperative transmitters.

[0072] Experimental platform and overall parameter settings: Experimental platform: MATLAB R2024b signal level simulation. Digital signal processing modules are implemented item by item according to numerical settings: frame-by-frame frequency domain transformation and frequency band trajectory construction, envelope pulse detection, pulse timing model and candidate arrival time window, candidate reference sequence truncation, normalized cross-correlation screening (threshold 0.80), time and phase alignment, 16-pulse coherent synthesis, digital downconversion and decimation filtering, version management, matched filtering, direct path matching peak consistency determination, non-uniform slow-time Doppler transformation, and two-dimensional CFAR detection (24 training units, threshold factor 12).

[0073] Deployment and acquisition parameters: The reference signal acquisition terminal is located 3km away from the non-cooperative transmitter, the RF front-end monitors the frequency range of 9-11MHz, and the gain is fixed; the analog-to-digital conversion module continuously acquires data at 4Msps (4 million complex sampling points per second) for 60s to form the first sampling sequence; the three receiving stations acquire the second sampling sequence containing the direct component and the target echo at the same sampling rate.

[0074] Key processing parameters: frame length 4096 points, frame shift 2048 points, Hanning window, frequency unit interval 976.5625Hz; occupied frequency band determination threshold −70dBFS (3 consecutive frequency units); frequency band association conditions are overlap coefficient ≥0.60 and center frequency difference ≤1953.125Hz, 8 consecutive frames registered as candidate trajectories; pulse envelope determination threshold −55dBFS (8 consecutive points), pulses with duration <100μs or >20ms are removed; the number of pulse interval relative deviations ≤2% ≥30 / 32 are determined as target frequency band trajectories; statistical time window 32 pulses, slip 16 pulses; parameter change thresholds are frequency difference ≥5kHz, bandwidth relative change ≥10%, interval relative change ≥5%, 3 consecutive windows confirmed; direct path version determination conditions are peak / background amplitude ratio ≥4 and the position range of 16 matching peaks ≤2 sampling points.

[0075] Statistical method: Each quantitative indicator is the mean of 20 Monte Carlo experiments (achieved with different noise levels).

[0076] Experimental signal scenario construction: The first sampling sequence includes a non-cooperative target emission source, co-channel radiation interference, adjacent-channel radiation interference, and background noise; the second sampling sequence at the receiving station superimposes the target echo (target distance 300km, Doppler +12.5Hz, propagated via single-hop skywave, echo signal-to-noise ratio set at 8-12dB at the receiving station) and multipath components. Radiation source parameter settings are shown in Table 1, and the scenario time-frequency structure is shown in... Figure 1 .

[0077] Table 1 Radiation source signal parameter settings

[0078] like Figure 3 As shown, Figure 3 The experimental signal scene time-frequency diagram shows that the target signal is a pulse train with a period of 10ms at 10.0MHz; the adjacent channel interference is a continuous narrow band at 9.3MHz; and the co-channel interference is another pulse train with a period of 7ms.

[0079] Experiment 1: Stability Verification of Parameter Extraction under Multi-Source Aliasing Experimental setup: target pulse period 10ms, pulse width 1ms; co-channel interference (7ms period) partially overlaps with the target frequency band, and adjacent channel interference is continuous.

[0080] Comparison method: "Direct Energy Detection Method" - only extracts the occupied frequency band and directly estimates the parameters based on the single frame −70dBFS threshold, without performing cross-frame frequency band trajectory correlation or pulse interval structure consistency screening.

[0081] Experimental results: see Table 2 and Figure 4 .

[0082] Table 2 Comparison of parameter extraction and pulse detection results (mean of 20 Monte Carlo simulations)

[0083] Figure 4 Comparing parameter extraction errors with a dumbbell plot, the direct energy detection method (DEM) suffers from several drawbacks when co-channel interference overlaps with the target frequency band. The occupied frequency band boundary is widened by the interference, the center frequency is skewed, and interference pulses with a 7ms period are mixed into the target pulse sequence, causing disordered interval sequences (only 71% of the intervals meet the 2% deviation requirement), resulting in a pulse detection error rate of 25.7%. The proposed method uses the overlap coefficient of adjacent frame frequency bands to form candidate trajectories, and then utilizes the timing consistency condition of "at least 30 of the 32 intervals with a relative deviation ≤ 2%" to eliminate co-channel interference pulses with different periods. Ultimately, the center frequency error is reduced to 0.09kHz (less than 1 / 10 of a single frequency unit of 976.5625Hz, thanks to the improved signal-to-noise ratio from the accumulated spectrum of 32 frames), the bandwidth error is 1.5kHz (relative error 0.75%), and the pulse detection error rate is 1.3%. The results demonstrate that cross-frame frequency band correlation and pulse interval structure screening significantly improve the stability of parameter extraction under complex electromagnetic environments.

[0084] Experiment 2: Verification of the Coherent Synthesis Quality of Reference Signal under Pulse Jitter Conditions Experimental setup: target pulse width 5ms, pulse arrival time randomly jittered within ±3% of the cycle (i.e., jittered evenly within ±0.3ms); signal-to-noise ratio at the acquisition end 25dB.

[0085] Comparison method: Method A (Equal-interval resampling method): Without pulse timing modeling and boundary detection, the reference segment is truncated at equal intervals according to the nominal 10ms period, and then resampled at equal intervals before direct coherent accumulation; Method B (the method of this invention): Predict candidate arrival time windows (±5% of the cycle) according to the pulse timing model, detect the actual pulse boundary within the window, extract candidate reference sequences, and after normalized cross-correlation screening (≥0.80) and time / phase alignment, take 16 sequences and coherently synthesize them.

[0086] Experimental results: see Table 3 and Figure 5 .

[0087] Table 3 Comparison of reference signal synthesis quality under jitter conditions (mean of 20 Monte Carlo simulations)

[0088] Figure 5The coherent accumulation output gain curve varies with the number of pulses. Method A, due to its equal-interval truncation according to the nominal period, suffers from jitter, causing random time shifts in the reference waveforms relative to the actual waveforms. This results in incomplete phase alignment during coherent synthesis, with a correlation coefficient of only 0.86. Furthermore, the error accumulates with the number of accumulated pulses, reaching a gain of only 13.4 dB at N=64, a loss of 4.6 dB relative to the ideal value, and a main lobe widening of 1.6 times. The present invention first predicts using a time-series model and then detects the actual pulse boundaries within the candidate time window, eliminating the truncation time shift at its source. Cross-correlation screening further eliminates abnormal segments, and time / phase alignment compensates for residual fractional time shifts, achieving a correlation coefficient of 0.995. At N=64, the gain reaches 17.7 dB (with a loss of only 0.3 dB), and the main lobe remains essentially unwidened. Figure 3 The curve obtained by the method of this invention almost coincides with the ideal coherence curve, while the equal-interval resampling curve deviates significantly as N increases.

[0089] Experiment 3: Verification of the effectiveness of reference version updates under dynamic changes in launch parameters Experimental setup: Target pulse width 110μs. Parameters for the first 30 seconds: center frequency 10.000MHz, bandwidth 100kHz, period 10ms; from the 30th second onwards, the center frequency jumps by +8kHz (exceeding the 5kHz threshold), the bandwidth becomes 120kHz (a relative change of 20%, exceeding the 10% threshold), and the period alternates between 8 / 12ms (a relative change of 20% in the mean, exceeding the 5% threshold), triggering parameter status change confirmation. The receiver echo signal-to-noise ratio is 10dB.

[0090] Comparison method: Method A (Fixed Single Version): The version 001 reference sequence generated from 0-30 seconds is used throughout the process without any updates; Method B (the method of this invention): Detect parameter changes by sliding statistical time window, generate version 002, and the receiving station confirms the current valid version by matching sampling points with a peak amplitude ratio ≥4 and a position range ≤2 along the direct path.

[0091] Experimental results: see Table 4.

[0092] Table 4 Comparison of target detection performance before and after parameter jumps (mean of 20 Monte Carlo runs)

[0093] After the parameter jump, the fixed single-version reference sequence mismatches with the new signal waveform, the matched filter output decreases, the direct path matching peak position drifts by 3.8 sampling points (not meeting the consistency condition of ≤2 sampling points), the processing gain loss is 5.2dB, and the detection probability drops from 0.96 to 0.52. The method of this invention confirms the parameter state change and generates version 002 within approximately 3 statistical time windows (approximately 0.5s) after the jump. The receiving station automatically selects the valid version based on the direct path matching result. After the jump, the detection probability remains at 0.94, the gain loss is only 0.4dB, and the matching peak position drifts by 0.5 sampling points. The results show that the reference version update and receiving station version confirmation mechanism effectively reduces the reference version mismatch caused by dynamic parameter changes and transmission delays.

[0094] Experiment 4: Verification of Non-Uniform Slow-Time Doppler Processing under Non-Equally Interval Pulse Sampling Experimental setup: Target Doppler +12.5Hz, distance 300km; pulse arrival time with ±3% periodic random jitter to form non-equal interval slow time sampling; 64 consecutive effective pulses; receiver echo signal-to-noise ratio 10dB.

[0095] Comparison method: Method A (Uniform Equal-Interval FFT): Ignore the actual arrival time, rearrange the 64 pulses into an equally spaced sequence according to the nominal period, and then perform a 64-point FFT; Method B (the method of this invention): Perform non-uniform slow-time Doppler transformation strictly according to the actual arrival time of each pulse (search range ±20Hz, step 0.02Hz), and use 24 training units CFAR detection.

[0096] Experimental results: see Table 5 and Figure 6 .

[0097] Table 5. Comparison of Doppler processing performance (mean of 20 Monte Carlo simulations)

[0098] Figure 6 The Doppler spectrum comparison shows that Method A forces non-uniformly spaced samples to be processed at equal intervals, which is equivalent to introducing random phase errors into the slow time series. This results in a 2.1-fold widening of the main lobe, a side lobe increase to -8.2 dB, an accumulation loss of 2.9 dB, and a Doppler estimation error of 0.42 Hz (an order of magnitude greater than the search step of 0.02 Hz), with a detection probability of only 0.71. The method of this invention calculates phase compensation pulse-by-pulse based on the actual arrival time and performs rotational accumulation, resulting in no main lobe widening, a side lobe increase of -19.4 dB, an accumulation loss of 0.2 dB, and a detection probability of 0.97. Figure 4The solid red line (method of this invention) shows a narrow and sharp main lobe and flat side lobes. The dashed blue line (equal-interval FFT) shows a broadened main lobe and an overall raised side lobe of about 10 dB, which visually verifies the effect of non-uniform slow-time Doppler processing in reducing Doppler broadening and coherence accumulation loss.

[0099] The four experiments correspond to the four beneficial effects of the invention, and all indicators can be stably reproduced by signal-level simulation under the stated parameter conditions: In complex electromagnetic environments (same-frequency + adjacent-frequency aliasing), the method of this invention has a center frequency error of 0.09kHz, a bandwidth error of 1.5kHz, and a pulse detection error rate of 1.3%, which are significantly better than the direct energy detection method (4.8kHz / 21.6kHz / 25.7%), and the stability of parameter extraction is significantly improved. Under ±3% pulse jitter, the correlation coefficient between the reference sequence and the real waveform is 0.995, and the coherent accumulation gain is 17.7dB (ideal 18.0dB) with N=64. This recovers 4.3dB of loss compared to the equal interval resampling method (13.4dB), and the effects of cross-pulse truncation and waveform mismatch are significantly reduced. After the transmission parameters changed, the reference version update kept the detection probability at 0.94 (the fixed single version dropped to 0.52), the processing gain loss was 0.4dB (compared to 5.2dB), and the reference version mismatch was effectively suppressed; Under non-uniform pulse sampling, the non-uniform slow-time Doppler transformation reduces the Doppler estimation error to 0.02 Hz, the sidelobe to −19.4 dB, and the detection probability to 0.97, significantly reducing Doppler broadening and coherence accumulation loss.

[0100] Example 2: In this embodiment, S1 includes: The 3-30MHz high-frequency signal received in the vicinity of the non-cooperative transmitter is divided into multiple sub-band signals by a filter bank; Each sub-band signal is synchronously sampled based on a unified sampling clock and divided into frames according to time to obtain multiple sub-band sampling sequences; Calculate the frame power of each sub-band sampling sequence, construct the channel power feature vector, determine the acquisition threshold based on the background power statistics, and filter the effective sub-bands; Based on the pre-obtained channel delay calibration parameters for each effective sub-band, the sampling position of each sub-band sampling sequence is compensated, and the corresponding channel identifier is associated; The multiple sub-band sampling sequences after compensation are established according to the sampling time and stored continuously to form a first sampling sequence that contains signals from multiple radiation sources and retains the sub-band correspondence.

[0101] In the above scheme, the reference signal acquisition end receives electromagnetic signals in the range of 3 MHz to 30 MHz via a high-frequency receiving antenna. The received signal is pre-limited and out-of-band filtered before being sent to a 27-channel filter bank. The 27 channels are arranged continuously at 1 MHz frequency intervals, with channel 1 corresponding to 3 MHz to 4 MHz, channel 2 corresponding to 4 MHz to 5 MHz, and so on, extending to channel 27 corresponding to 29 MHz to 30 MHz. The passband boundary of each channel coincides with the corresponding sub-band frequency boundary, and the stopband is entered at 40 dB outside the passband. The filter output of each sub-band is converted to a complex baseband via quadrature frequency conversion, thereby converting the received signal from 3 MHz to 30 MHz into 27 independent sub-band signals with known frequency ranges. Among them, channel 7 covers 9 MHz to 10 MHz, channel 8 covers 10 MHz to 11 MHz, and these two channels constitute the subsequent processing frequency range of 9 MHz to 11 MHz in the above embodiment. The 27 sampling channels share a single 40 MHz reference clock. This reference clock is frequency-divided into 4 million sampling pulses per second, which are simultaneously fed into the analog-to-digital converter circuits of each channel. At the start of acquisition, a synchronization pulse of the same frame resets the sampling counters of all 27 channels, ensuring that the first sampling point of each channel corresponds to the same sampling time. Each channel samples continuously at 4 million complex sampling points per second, with each frame consisting of 4096 sampling points. There is a 2048-point overlap between adjacent frames, resulting in a 2048-point difference in the starting sampling positions of adjacent frames. This framing method is consistent with the framing process of the first sampling sequence described above. After framing, the frame power is calculated frame by frame for each channel. For each of the 4096 complex sampling points in a frame, the in-phase and quadrature components are obtained. The power of the sampling point is obtained by multiplying the in-phase component by itself and adding the power of the quadrature component by itself. The power of the 4096 sampling points is accumulated sequentially, and the result is divided by 4096 to obtain the frame power of the current frame for that channel. Under the same frame number, the 27 frame powers are arranged in a fixed order from channel 1 to channel 27 to form the channel power feature vector corresponding to the current frame. The positions in the feature vector maintain a fixed correspondence with the physical channel number, so that each power value can be traced back to a defined sub-band frequency range.

[0102] The acquisition threshold is determined based on the background power of each channel. For any given channel, 256 frames of power are continuously acquired. These 256 frames are then arranged in ascending order of value, and the first 64 frames are taken as the background power sample. The sum of these 64 frame power values ​​is then divided by 64 to obtain the average background power. Subsequently, the difference between each of the 64 frame power values ​​and the average background power is calculated, and the absolute value of each difference is taken. The maximum value among the 64 absolute differences is taken as the background power fluctuation for that channel. The background power fluctuation is multiplied by 3 and then added to the average background power; the resulting power value is determined as the acquisition threshold for that channel. Therefore, the acquisition threshold is determined by the measured background power and its fluctuation range for the current channel, without using a pre-fixed absolute power threshold.

[0103] Valid subbands are determined based on continuous frame power. Each channel is checked for eight consecutive frames, and the frame power of each frame is compared with the acquisition threshold for that channel. If 6, 7, or 8 out of the 8 frames reach the acquisition threshold, the channel is marked as a valid subband. If 0 to 5 frames reach the acquisition threshold, the channel is not added to the current valid subband set. After completing one 8-frame determination, the process moves forward 4 frames and re-executes the next determination, updating the valid subband status every 4 frames. In this embodiment, channels 7 and 8 both meet the above determination conditions, therefore data in the 9 MHz to 11 MHz range enters the aforementioned subsequent frequency band trajectory extraction process. Other channels, as long as they meet the same determination conditions, are also saved as valid subbands, thus preserving other simultaneously occurring radiation source signals in the first sampling sequence.

[0104] The sampling position compensation for each channel is completed based on the channel delay calibration parameters measured before the equipment is put into operation. During calibration, the same broadband pulse signal is simultaneously input to 27 channels via an equal-length power branch, with the pulse sampling position of the first channel as the time reference. Delay measurements are performed on channels 2 through 27 respectively. When measuring a channel, the calibration sequence of that channel is moved point by point within a range of 64 sampling points before and after the first channel. For each moving sampling point, the complex sample values ​​at the overlapping positions of the two sequences are multiplied point by point, with the sample value corresponding to the first channel undergoing conjugation processing. All products are then added together to obtain the amplitude of the accumulated result. The moving position with the largest amplitude corresponds to the integer sampling position difference between the two channels, and this position difference is recorded as the integer delay calibration value of that channel.

[0105] When the maximum correlation amplitude is located between adjacent integer sampling points, the correlation amplitude of the sampling point containing the maximum value and one sampling point before and after it are read. Quadratic interpolation is performed based on these three correlation amplitudes to obtain the offset of the correlation peak relative to the integer sampling point, and this offset is recorded as the fractional sampling delay calibration value. The integer delay calibration value and fractional sampling delay calibration value for each channel are written to non-volatile memory after calibration and bound to the corresponding physical channel number.

[0106] When performing sampling position compensation on effective sub-bands, the integer delay calibration value of that channel is read first. If the channel signal is delayed by 3 sampling points relative to the first channel, the channel sequence is shifted forward by 3 sampling points; if the channel signal is advanced by 2 sampling points relative to the first channel, the channel sequence is shifted backward by 2 sampling points. After integer sampling position compensation, the fractional sampling delay calibration value is read and compensated using a 21st-order fractional delay finite impulse response filter, with the filter's delay direction set opposite to the measured channel's delay direction. After integer and fractional position adjustments, the sampling positions of each effective sub-band are unified to the sampling time reference of the first channel.

[0107] Each compensated valid sub-band is assigned a fixed channel identifier. The channel identifier uses the physical channel number and simultaneously stores the start and end frequencies of that channel. For example, channel 7 stores frequency information of 9 MHz and 10 MHz, and channel 8 stores frequency information of 10 MHz and 11 MHz. The channel identifier, start frequency, and end frequency are stored together with the corresponding sampled data and the correspondence remains unchanged during subsequent reading and processing. A time correspondence is established between different valid sub-bands based on a unified sampling count value. After the frame synchronization pulse arrives, the first sampling time number is set to 0, and thereafter, the sampling time number is incremented by 1 for each sampling pulse. Sample points of valid sub-bands at the same sampling time use the same sampling time number. For continuous storage, a data block is composed of 4096 sample points. Each data block sequentially records the frame start sampling time number, channel identifier, sub-band start frequency, sub-band end frequency, and 4096 time-delay compensated complex sample points; at the start time of the same frame, all valid sub-band data are written sequentially in ascending order of channel number.

[0108] After adopting the above storage method, the sub-band data corresponding to the same sampling time can be directly matched according to the sampling time number, and the continuous data of the same sub-band can be read continuously according to the channel identifier. The first sampling sequence thus formed simultaneously retains the sampling time relationship and the sub-band frequency correspondence. When processing the 9 MHz to 11 MHz data in the subsequent processing, the sampling blocks corresponding to the 7th and 8th channels are directly read, and the processing parameters of 4096-point frame length, 2048-point frame shift, and 4 million complex sampling points per second are continued to be used. Thus, without changing the above subsequent processing conditions, the sub-band selection, time delay calibration, and first sampling sequence construction of the 3 MHz to 30 MHz high-frequency received signal are completed.

[0109] Example 3: S2 includes: The first sampled sequence is divided into frames and windowed frequency domain transform is performed to obtain the time-frequency power matrix; The spectral threshold is determined based on the spectral power and background statistics of each frame, and continuous over-threshold frequency units are extracted to form candidate occupied frequency bands. The candidate frequency band trajectory is formed by associating the frequency overlap rate and center frequency deviation of candidate occupied frequency bands in adjacent frames. Band-limited extraction is performed on the data corresponding to each candidate frequency band trajectory to obtain sub-band sequences, and pulse arrival times are detected based on the adaptive threshold of their envelope sequences. A pulse interval sequence is formed based on the arrival times of adjacent pulses. A time sequence consistency feature value is constructed based on the clustering characteristics of the pulse interval values ​​and the repetition of the interval arrangement pattern. Target frequency band trajectories that meet the consistency threshold are selected, and the center frequency, bandwidth, and pulse arrival time sequence are determined accordingly.

[0110] In the above scheme, step two uses the first sampling sequence formed in step one as the processing object. In this embodiment, the sampling rate of each effective sub-band is maintained at 4 million complex samples per second, with 4096 samples constituting one frame, and 2048 identical samples are retained between two adjacent frames. For the processing range of 9 MHz to 11 MHz mentioned above, the sampling data corresponding to the 7th and 8th channels are read respectively, and the start position of each frame is determined according to the unified sampling time number.

[0111] For each frame, 4096 complex sampling points are sequentially multiplied by a 4096-point Hanning window, and then a 4096-point Fast Fourier Transform is performed. For each frequency unit after the frequency domain transform, its real part is multiplied by itself, its imaginary part is multiplied by itself, and the two multiplication results are added together. The result is recorded as the spectral power of that frequency unit. The sampling rate is 4 million sampling points per second, and the interval between adjacent frequency units corresponding to the 4096-point frequency domain transform is 976.5625 Hz. The spectral power of each frame is arranged according to the frame number, so that the same column always corresponds to the same frequency position, thus obtaining a time-frequency power matrix with time frames as rows and frequency units as columns.

[0112] The spectral threshold for each frame is determined based on its spectral power. For the 1024 frequency units within the 1 MHz sub-band, arranged in ascending order of spectral power, the first 256 spectral power units are taken as background power samples. The average background power of the frame is obtained by summing these 256 background power units and dividing by 256. The difference between each background power unit and the average background power is then calculated, and the absolute value of each difference is summed and divided by 256 to obtain the average background fluctuation. This average fluctuation is multiplied by 4 and added to the average background power; the resulting power value is used as the spectral threshold for the current frame. In this embodiment, for data received between 9 MHz and 11 MHz, the spectral threshold obtained using the above process is converted to -70 dB relative to the full-scale power of the analog-to-digital converter, consistent with the aforementioned spectral occupancy determination criteria. Coefficient 4 is determined by collecting 60 seconds of pulse-free background data before the reference signal acquisition terminal is put into operation. During the calibration period, the average background power and average background fluctuation are calculated according to the same framing method. When the coefficient increases from 1 to 4, the background frequency unit does not form an occupied frequency band consisting of 3 consecutive over-threshold frequency units for 8 consecutive frames. Therefore, 4 is used as the fixed fluctuation coefficient of the spectrum threshold.

[0113] The spectral power of each frequency unit is compared with the spectral threshold of the current frame one by one. When the power of three consecutive frequency units reaches the spectral threshold, the first frequency unit exceeding the threshold is taken as the left starting point of the candidate occupied frequency band, and the check continues in the direction of frequency increase; until a frequency unit below the spectral threshold is found, the last consecutive frequency unit exceeding the threshold is taken as the right ending point. The center frequency corresponding to the left starting point is taken as the left boundary of the candidate occupied frequency band, and the center frequency corresponding to the right ending point is taken as the right boundary. The bandwidth of the candidate occupied frequency band is obtained by subtracting the left boundary frequency from the right boundary frequency; the center frequency of the candidate occupied frequency band is obtained by adding the left boundary frequency and the right boundary frequency and dividing by 2. When there are fewer than three consecutive frequency units exceeding the threshold, no candidate occupied frequency band is formed to avoid transient power fluctuations of a single frequency unit and two adjacent frequency units from entering subsequent trajectory association.

[0114] Adjacent frames are correlated simultaneously based on frequency overlap rate and center frequency deviation. For one candidate occupied frequency band in the previous frame and one candidate occupied frequency band in the next frame, the larger of the two left boundary frequencies is taken as the starting position of the overlapping frequency range, and the smaller of the two right boundary frequencies is taken as the ending position of the overlapping frequency range. When the ending position is higher than the starting position, the overlapping bandwidth is obtained by subtracting the starting frequency from the ending frequency; when the ending position is not higher than the starting position, the overlapping bandwidth is recorded as 0. The bandwidths of the two candidate occupied frequency bands are calculated separately, and the smaller bandwidth is taken. The overlapping bandwidth is then divided by this smaller bandwidth to obtain the frequency overlap rate. The center frequency deviation is obtained by subtracting the center frequency of the candidate occupied frequency band in the previous frame from the center frequency of the candidate occupied frequency band in the next frame and taking the absolute value. When the frequency overlap rate reaches 0.60 and the center frequency deviation does not exceed 1953.125 Hz, the two candidate occupied frequency bands are connected to the same frequency band track. 1953.125 Hz corresponds to two frequency units, representing the allowable center frequency shift range between adjacent frames; a frequency overlap rate of 0.60 indicates that adjacent frames retain at least 60% of the common frequency range of the current narrower frequency band. After these two conditions are met consecutively for eight frames, the corresponding trajectory is determined as a candidate frequency band trajectory; if neither of the two conditions is met simultaneously in any adjacent frame, the current trajectory is terminated at that frame position.

[0115] For the formed candidate frequency band trajectory, the center frequencies of each valid frame within the trajectory are summed sequentially and then divided by the number of valid frames to obtain the trajectory center frequency; the bandwidths of each valid frame are summed sequentially and then divided by the number of valid frames to obtain the trajectory average bandwidth. The corresponding data in the first sampling sequence is digitally down-converted according to the obtained trajectory center frequency, shifting the candidate signal to the zero-frequency position. The down-converted data is then band-limited using a 129th-order finite impulse response filter. The total passband width of the filter is 1.20 times the average bandwidth of the trajectory, and the initial total stopband width is 1.40 times the average bandwidth of the trajectory, thus obtaining a sub-band sequence that retains only the signal components corresponding to the candidate frequency band trajectory.

[0116] The envelope power is calculated point-by-point for each sample of the sub-band sequence. For each complex sample point, the in-phase component and the quadrature component are obtained. The two components are multiplied by themselves and then added together; the resulting value is the envelope power for that sample point. The envelope detection threshold is updated every 4096 sample points. Specifically, the current 4096 envelope powers are arranged in ascending order, and the first 1024 envelope powers are taken as background samples. The sum of these 1024 background samples is divided by 1024 to obtain the average envelope background power. The difference between each background sample and the average envelope background power is then calculated, and the absolute value is taken. The sum of these 1024 absolute values ​​is divided by 1024 to obtain the average envelope background fluctuation. The average envelope background fluctuation is multiplied by 4 and then added to the average envelope background power; the result is determined as the envelope detection threshold for the current frame. In this embodiment, the envelope detection threshold calculated above is converted to -55 dB relative to the full-scale power, consistent with the pulse detection conditions described above. The coefficient 4 in the envelope threshold is determined using the aforementioned 60-second background calibration data. Under the condition that the pulse start point must not be triggered by the background power in 8 consecutive sampling points, the smallest integer coefficient 4 that prevents the background data from generating pulse detection results is fixed.

[0117] When the envelope power reaches the envelope detection threshold for eight consecutive sampling points, the first sampling point out of those eight is designated as the pulse arrival position. After the pulse arrives, when the envelope power falls below the envelope detection threshold for eight consecutive sampling points, the first sampling point out of those eight is designated as the pulse end position. The pulse duration is calculated based on the number of sampling points between the start and end sampling points and the sampling rate of 4 million sampling points per second. Detection results with a duration less than 100 microseconds or greater than 20 milliseconds are deleted, and the remaining detection results are saved in chronological order of arrival.

[0118] For each candidate frequency band trajectory, 33 valid pulse arrival times are continuously obtained. The arrival times of two adjacent pulses are subtracted to obtain one pulse interval, resulting in 32 pulse intervals. These 32 pulse intervals are then sequentially clustered. First, the first pulse interval is used to establish the first interval group, which is then used as the initial center value for that group. When processing subsequent pulse intervals, the absolute value of the difference between the current pulse interval and the center value of each existing interval group is calculated, and then divided by the corresponding interval group center value. If the calculation result does not exceed 0.02, the current pulse interval is assigned to the interval group with the smallest calculation result; if the calculation results of all existing interval groups exceed 0.02, a new interval group is established. For each pulse interval assigned, all interval values ​​within that group are summed and divided by the number of intervals in the group to redetermine the group's center value. After all 32 pulse intervals have been classified, interval groups containing four or more pulse intervals are identified as valid interval groups. The number of pulse intervals contained in all valid interval groups is counted, and this number is divided by 32. The result is taken as the degree of pulse interval clustering. A valid interval group must contain at least 4 pulse intervals because subsequent examination of the arrangement patterns of interval lengths from 1 to 8 shows that when 8 intervals constitute one cycle, 32 pulse intervals can form 4 complete repetitions. Therefore, 4 occurrences constitute the minimum number of repetitions required to form a valid interval state.

[0119] The interval arrangement pattern is judged according to the original time order of the pulse intervals. First, a single pulse interval is used as a candidate period length. The interval group to which the first pulse interval belongs is compared with the interval group to which the second pulse interval belongs, and then subsequent adjacent positions are compared sequentially. Subsequently, 2, 3, up to 8 pulse intervals are used as candidate period lengths. When using 2 intervals as candidate period lengths, the interval group to which the first interval belongs is compared with the third, the second with the fourth, and so on. Other candidate period lengths are processed in the same way. For each candidate period length, the number of times corresponding positions belong to the same interval group is counted, and then divided by the total number of comparisons actually performed under that candidate period length to obtain the arrangement repetition degree of that period length. After all 1 to 8 candidate period lengths have been calculated, the arrangement repetition degree with the largest value is taken as the interval arrangement repetition degree of the trajectory.

[0120] The degree of pulse interval clustering and the degree of interval repetition are compared, and the smaller of the two values ​​is taken as the timing consistency feature value. The timing consistency threshold is set to 0.93. This threshold is determined based on the fact that two intervals out of 32 pulse intervals are allowed to deviate from the established timing relationship. The value obtained by dividing 30 effective intervals by 32 is 0.9375, so 0.93 is taken as the judgment threshold. When the timing consistency feature value reaches 0.93, and at least 30 pulse intervals out of 32 pulse intervals belong to the effective interval group, the corresponding candidate frequency band trajectory is determined as the target frequency band trajectory; if the above conditions are not met, the candidate frequency band trajectory is deleted.

[0121] After the target frequency band trajectory is determined, 32 frames of spectrum corresponding to the trajectory are continuously acquired. The power of the 32 spectra at the same frequency position is added together and divided by 32 to obtain the average power at that frequency position, and this forms the 32-frame average power spectrum. The spectral threshold of the average power spectrum is re-determined according to the aforementioned spectral background statistical process, and the frequency units that continuously reach this threshold are taken as the target effective frequency band. The leftmost frequency of the target effective frequency band is taken as the left boundary, and the rightmost frequency is taken as the right boundary. The two boundary frequencies are added together and divided by 2 to obtain the center frequency of the target signal. The target signal bandwidth is obtained by subtracting the left boundary frequency from the right boundary frequency. The pulse arrival times that have passed the timing consistency screening are continuously saved in the order of occurrence to form the pulse arrival time series used in step three to establish the pulse timing model.

[0122] Example 4: In this embodiment, S3 includes: The pulse arrival time series is subtracted to obtain the pulse interval series, and a pulse time series model is established based on the distribution characteristics and arrangement relationship of the pulse intervals; The arrival time window of candidate pulses is predicted using the pulse timing model, and the actual boundary of each candidate pulse is determined based on the envelope adaptive threshold. Candidate reference sequences are truncated according to the actual boundary, the normalized cross-correlation between candidate reference sequences is calculated, reference sequences are filtered according to the correlation threshold, and time-domain alignment and phase compensation are performed according to the cross-correlation peak position and cross-correlation peak complex phase, respectively. Coherent synthesis is performed on the time- and phase-aligned reference sequences to obtain the reference signal sequence; The reference signal sequence is orthogonally downconverted based on the center frequency, and the decimation parameters are determined based on the bandwidth. The reference IQ sequence is then obtained through decimation filtering.

[0123] In the above scheme, step three uses the center frequency, bandwidth, and pulse arrival time sequence output in step two as the processing basis, and reads the complex sampling data corresponding to the target frequency band trajectory from the first sampling sequence. This embodiment continues to use a sampling rate of 4 million complex sampling points per second, and the sampling position is determined by a unified sampling time number for both pulse detection and reference sequence truncation.

[0124] Thirty-three pulse arrival times are read sequentially. The arrival time of each pulse is subtracted from the arrival time of the previous pulse, resulting in 32 pulse intervals. These 32 pulse intervals are then categorized using the interval aggregation method determined above. In this embodiment, at least 30 pulse intervals are grouped into the same effective interval group, and the repetition rate reaches 0.93 when the arrangement period length is 1. Therefore, the pulse intervals corresponding to this effective interval group are taken as the current timing state. All pulse intervals within this group are summed, and then divided by the number of pulse intervals within that group to obtain the pulse interval baseline value for the current timing state.

[0125] The arrival time of the first pulse out of 33 pulses is taken as the timing starting point. The predicted time of the second pulse is obtained by adding one pulse interval reference value to the timing starting point, and the predicted time of the third pulse is obtained by adding two pulse interval reference values ​​to the timing starting point. Subsequent predicted times are increased by the corresponding number of pulse interval reference values ​​according to the pulse sequence number, thus establishing the correspondence between the pulse sequence number and the predicted arrival time. The candidate pulse arrival time window is determined based on the deviation of the detected pulses from the predicted time. First, the time difference between the actual arrival time and the corresponding predicted arrival time of each of the aforementioned 33 pulses is calculated, and the absolute value of the time difference is taken. In this embodiment, the maximum time difference does not exceed 0.02 times the pulse interval reference value. 0.03 times the pulse interval reference value is added to the maximum time difference as a boundary margin, and finally, 0.05 times the pulse interval reference value before and after the predicted time is determined as the search range. Therefore, the total width of the arrival time window of each candidate pulse is 0.10 times the pulse interval reference value. Subsequent pulses are all searched according to this time window.

[0126] The envelope power is recalculated within each candidate pulse arrival time window. For each complex sampling point, the in-phase component and quadrature component are obtained separately. The result of multiplying the in-phase component by itself is added to the result of multiplying the quadrature component by itself, and this is taken as the envelope power of that sampling point. The envelope detection threshold is updated once every 4096 sampling points. The 4096 envelope powers are arranged in ascending order of value, and the first 1024 envelope powers are taken as background samples. The sum of the 1024 background samples is divided by 1024 to obtain the average background power. The difference between the 1024 background samples and the average background power is calculated. The absolute value of each difference is taken, and the sum of the 1024 absolute values ​​is divided by 1024 to obtain the average background fluctuation. The average background fluctuation is multiplied by 4 and then added to the average background power. The resulting power value is taken as the envelope detection threshold for the current data segment.

[0127] In this embodiment, coefficient 4 follows the background calibration results described above. Before formal acquisition, the reference signal acquisition terminal continuously records 60 seconds of pulse-free background data, with 4096 sampling points per frame. Detection is performed using the background average power plus 1, 2, 3, and 4 times the background average fluctuation amount, respectively. When using 4 times the background average fluctuation amount, no consecutive 8 sampling points in the background data reach the detection threshold; therefore, 4 is determined as the fixed fluctuation coefficient for the envelope threshold. The envelope detection threshold obtained in this embodiment corresponds to -55 dB relative to the full-scale power of the analog-to-digital converter, consistent with the aforementioned embodiment.

[0128] Within the candidate pulse arrival time window, when the envelope power reaches or exceeds the envelope detection threshold for eight consecutive sampling points, the first of these eight sampling points is determined as the actual pulse start boundary. Detection continues forward from the start boundary; when the envelope power is below the envelope detection threshold for eight consecutive sampling points, the first of these eight sampling points is determined as the actual pulse termination boundary. The number of sampling points occupied by the pulse is obtained by subtracting the sampling point number of the start boundary from the sampling point number of the termination boundary. This number is then divided by 4 million sampling points per second to obtain the pulse duration. Detection results are deleted if the pulse duration is less than 100 microseconds or greater than 20 milliseconds.

[0129] Based on the determined actual pulse start and end boundaries, 32 candidate reference sequences are extracted from the complex sampled data corresponding to the target frequency band trajectory. Due to the differences in sampling boundaries between pulse durations, the number of sampling points in the shortest sequence among the 32 candidate reference sequences is used as the length for subsequent correlation calculations, and the same number of sampling points are retained starting from the start boundary of each candidate reference sequence.

[0130] The sequence energy of each of the 32 candidate reference sequences is calculated. For each candidate reference sequence, the in-phase and quadrature components of each complex sampling point are multiplied by themselves, and the two multiplication results are added together to obtain the energy of that sampling point. The energies of all sampling points in the sequence are added together to obtain the total sequence energy. The total energies of the 32 candidate reference sequences are compared, and the candidate reference sequence with the largest total energy value is determined as the reference sequence for correlation calculation. The normalized cross-correlation of the remaining candidate reference sequences is calculated with the reference sequence. The position of the sequence to be compared relative to the reference sequence is changed point by point within a range of 64 sampling points. For each sampling point moved, the complex sample value in the current overlapping interval of the two sequences is taken. The corresponding sample value of the reference sequence is conjugated and multiplied with the sequence to be compared point by point, and all products are added together to obtain the amplitude of the accumulated result. The total energy of the two sequences in the current overlapping interval is then calculated. The square root of the product of the two total energies is then taken, and the amplitude of the aforementioned correlation accumulation result is divided by the square root value to obtain the normalized cross-correlation value at the current position. After traversing the range of 64 sampling points before and after, the largest normalized cross-correlation value is determined as the correlation peak of the candidate reference sequence.

[0131] The correlation threshold is set at 0.80. This threshold is determined during the initialization phase of the reference signal acquisition terminal based on non-pulse region data. Using the initial reference sequence as a reference, 16 segments of non-pulse data of the same length as the reference sequence are continuously extracted outside the adjacent pulse time windows, and the normalized cross-correlation peak value is calculated according to the aforementioned method. In this embodiment, the maximum value among the 16 non-pulse data correlation peak values ​​is 0.70. Adding 0.10 to this maximum value as the correlation decision interval yields 0.80, which is then saved as the correlation threshold used in this acquisition process. Candidate reference sequences with correlation peak values ​​reaching or exceeding 0.80 are retained, while those below 0.80 are deleted. After obtaining 16 candidate reference sequences that meet this condition in chronological order, alignment processing begins; if the number is less than 16 segments, processing continues with subsequent pulses until 16 sequences meeting the condition are obtained.

[0132] Time alignment is performed based on the position of the cross-correlation peak. When the cross-correlation peak is located at an integer sampling position, the number of sampling points between the peak position and the zero-delay position of the reference sequence is calculated, and the corresponding candidate reference sequence is moved in the opposite direction to make its cross-correlation peak coincide with the zero-delay position of the reference sequence. When the cross-correlation peak is between two integer sampling positions, the correlation values ​​at the sampling position of the peak and one sampling position before and after it are read. A quadratic curve is fitted with the sampling position as the x-axis and the correlation value as the y-axis, and the position corresponding to the maximum value of the quadratic curve is taken as the actual correlation peak position. The fractional sampling deviation is obtained by subtracting the integer sampling position from this position, and then the deviation is compensated by a 21st-order fractional-delay finite-impulse response filter. After completing the integer sampling shift and fractional sampling compensation, the 16 reference sequences adopt the same time base.

[0133] Phase compensation is determined using the complex results corresponding to the cross-correlation peaks. The real and imaginary parts of this complex result are obtained separately. The arctangent is calculated by dividing the absolute value of the imaginary part by the absolute value of the real part to obtain the fundamental phase angle. The quadrant is then determined based on the positive and negative states of the real and imaginary parts, and the fundamental phase angle is corrected to a complete complex phase angle. Subsequently, all complex sampling points in the corresponding candidate reference sequence are rotated in the opposite direction of this complex phase angle to ensure that the cross-correlation peaks of the 16 reference sequences have a consistent complex phase.

[0134] After time alignment and phase compensation, coherent synthesis of the 16 reference sequences is performed. First, a common sampling interval with valid data for all 16 sequences is determined. For the first sampling position within this common sampling interval, the complex sample values ​​of the 16 sequences at that position are added together and then divided by 16. The result is used as the first sampling point of the reference signal sequence. The remaining sampling positions are processed in the same way until all sampling points within the common sampling interval are synthesized, thus obtaining the reference signal sequence.

[0135] After the reference signal sequence is formed, orthogonal down-conversion is performed based on the center frequency obtained in step two. Since channel identifiers and corresponding frequency ranges have been reserved for each sub-band, the center frequency of the current reference signal sub-band is first obtained based on the channel identifier. Then, the target center frequency determined in step two is subtracted from the center frequency of this sub-band to obtain the frequency offset of the target signal in the current complex baseband sub-band. The digital quadrature local oscillator generates in-phase and quadrature components point-by-point according to this frequency offset and a sampling clock of 4 million samples per second. These components are then multiplied with the reference signal sequence point-by-point to move the target signal center to the zero-frequency position. The decimation parameters are obtained from the target signal bandwidth determined in step two. The down-converted baseband signal first passes through a 129th-order finite impulse response filter, where the passband extends from zero frequency towards both the positive and negative frequency directions to 0.55 times the target signal bandwidth, and the stopband starts at 0.70 times the target signal bandwidth. The decimation factor starts from an integer 1 and increases sequentially. Each increment is calculated by dividing 4 million samples per second by the current decimation factor to obtain the decimated sampling rate. This sampling rate is then compared to 2.5 times the target signal bandwidth. The decimation factor is retained if the decimated sampling rate reaches or exceeds 2.5 times the target signal bandwidth; it stops increasing if it falls below 2.5 times the target signal bandwidth. The previous integer that meets the condition is used as the final decimation factor.

[0136] After baseband filtering is completed, sampling points are retained at equal intervals according to a determined decimation factor, thereby reducing the sampling rate of the reference data. Since band-limited filtering is completed before decimation, frequency components outside the Nyquist frequency are suppressed after decimation and do not enter the effective bandwidth of the target signal. After actual boundary truncation, correlation filtering, time alignment, phase compensation, coherent synthesis of 16 sequences, quadrature downconversion, band-limited filtering, and decimation, a reference quadrature baseband sequence is obtained for subsequent reference version updates and matching filtering at the receiving station.

[0137] Example 5: Furthermore, S4 includes: Extract the distribution and arrangement features of center frequency, bandwidth and pulse interval within a continuous sliding time window, and construct a parameter state feature vector; Calculate the degree of state change between the parameter state feature vectors corresponding to adjacent time windows, and determine the position of parameter state change based on the preset change threshold; The parameter stability intervals are divided according to the positions of adjacent parameter state changes, and a set of state parameters including center frequency, bandwidth, and pulse interval distribution and arrangement parameters are determined for each parameter stability interval; The reference signal is re-triggered and processed based on the state parameter set corresponding to the stable interval of each parameter, an updated reference IQ sequence is generated, and the corresponding baseband sampling rate is recorded. The updated reference IQ sequence is associated with the corresponding state parameter set and parameter stability interval, and a version identifier is assigned to form a set of reference IQ sequences arranged in the order of parameter state changes.

[0138] In the above scheme, step four uses the center frequency, bandwidth, and pulse arrival time series obtained in step two as the basis for parameter state detection, and calls the original complex sampling data used in step three to form the reference IQ sequence. Parameter state detection is performed using a continuous sliding method, with each time window consisting of 32 valid pulses. The first time window contains the 1st to 32nd valid pulses, the second time window contains the 17th to 48th valid pulses, and thereafter, a new time window is formed by moving forward 16 valid pulses each time. A statistical length of 32 valid pulses is used, consistent with the processing scale of 32 pulse intervals in steps two and three; a sliding step size of 16 valid pulses is used, ensuring that adjacent time windows retain 16 common pulses to determine whether parameter changes can continue.

[0139] For each time window, the center frequency, bandwidth, pulse interval distribution, and pulse interval arrangement features are extracted. The center frequency is obtained by reading the center frequencies corresponding to each effective pulse within the time window, summing the 32 center frequencies sequentially, and then dividing by 32. The result is used as the center frequency of the current time window. The bandwidth is processed in the same way: the bandwidths corresponding to the 32 effective pulses are summed and then divided by 32. The result is used as the bandwidth of the current time window.

[0140] The pulse interval distribution follows the classification criteria determined in step two. The 32 pulse intervals corresponding to the current time window are obtained in chronological order and then arranged in ascending order of interval value. The first pulse interval is used to establish the first interval group, with this interval value as the center value of the group. The remaining pulse intervals are compared sequentially with the existing interval groups. First, the difference between the current pulse interval and the center value of the interval group is calculated. Then, the absolute value of this difference is taken and divided by the corresponding interval group center value. When the result does not exceed 0.02, the current pulse interval is assigned to that interval group. If two interval groups simultaneously meet the 0.02 condition, the pulse interval is assigned to the interval group with the smallest calculated value. If none of the interval groups meet this condition, a new interval group is established based on the current pulse interval. For each additional pulse interval, all interval values ​​within the group are summed and divided by the number of intervals in the group to redetermine the center value of the interval group. After classifying the 32 pulse intervals, only interval groups containing four or more pulse intervals are retained as valid interval groups. Four pulse intervals constitute 0.125 of 32 pulse intervals, corresponding to the minimum number of complete repetitions allowed when 1 to 8 intervals form an arrangement period in step two. Each effective interval group is arranged in ascending order of its center value, and the center value of each interval group and the number of pulse intervals contained within that group are recorded sequentially. Then, the number of pulse intervals in each group is divided by 32 to obtain the proportion of the corresponding interval group within the current time window, thus characterizing the distribution of pulse intervals.

[0141] The pulse interval arrangement features are processed according to the 1 to 8 candidate interval periods used in step two. When the candidate period is 1, the interval groups to which the 1st interval belongs are compared sequentially with the 2nd interval, and the interval groups to which the 2nd interval belongs are the same. When the candidate period is 2, the interval groups to which the 1st interval belongs are compared are compared, the interval groups to which the 3rd interval belongs are compared, and the interval groups to which the 4th interval belongs are compared, and so on. When the candidate period increases from 3 to 8, the two pulse intervals that are separated by a corresponding number of positions are compared in the same way. For each candidate period, the number of comparisons to which the interval groups belong is counted, and then divided by the total number of comparisons actually completed in that candidate period to obtain the degree of arrangement repetition. The degree of arrangement repetition corresponding to the 8 candidate periods is compared, and the candidate period with the largest value is taken as the arrangement period of the current time window, and the degree of repetition corresponding to that period is recorded.

[0142] The parameters of the current time window are saved in the following order: center frequency, bandwidth, number of effective interval groups, center value of each effective interval group, proportion of each effective interval group, arrangement period, and degree of repetition. This forms the parameter state feature vector of the current time window. Each time window uses the same parameter order, allowing direct comparison of corresponding parameters in adjacent time windows.

[0143] The degree of state change between adjacent time windows is jointly determined by changes in center frequency, bandwidth, pulse interval distribution, and arrangement characteristics. The center frequency change is first calculated by subtracting the center frequency of the previous time window from the center frequency of the subsequent time window, taking the absolute value of the difference, and then dividing this absolute value by 5 kHz. A result of 1 indicates that the center frequency change between adjacent time windows reaches 5 kHz. This 5 kHz threshold is determined by the frequency resolution in step two. In step two, the interval between adjacent frequency units is 976.5625 Hz, and five consecutive frequency units correspond to 4882.8125 Hz. Based on the parameter recording precision, 5 kHz is taken as the threshold for center frequency state change.

[0144] Bandwidth changes are determined using relative changes. The bandwidth of the previous time window is subtracted from the bandwidth of the subsequent time window. The absolute value of this difference is then divided by the bandwidth of the previous time window to obtain the bandwidth change ratio. This ratio is then divided by 0.10 to obtain the bandwidth change component. When the bandwidth change ratio reaches 0.10, the corresponding bandwidth change component reaches 1. 0.10 is used to limit a continuous 10% bandwidth change between adjacent time windows, maintaining consistency with the parameter state change conditions described above.

[0145] The comparison of pulse interval distribution first checks the number of effective interval groups. When the number of effective interval groups changes between two adjacent time windows, the change component of the pulse interval distribution is directly recorded as 1. When the number of effective interval groups is the same, the groups are compared one by one in ascending order of their center values. For interval groups at the same sequential position, the center value of the interval group in the later time window is subtracted from the center value of the interval group in the previous time window, and the absolute value of the difference is taken. This difference is then divided by the center value of the interval group in the previous time window to obtain the change ratio of the interval center. Additionally, the number ratios of corresponding interval groups in the two time windows are subtracted, and the absolute value is taken to obtain the change value of the number ratio.

[0146] Divide the change ratio of the interval center by 0.05 to obtain the center change result of the interval group; divide the change value of the quantity ratio by 0.125 to obtain the quantity change result of the interval group. Compare the two results and take the larger value as the change value of the interval group. Then compare the change values ​​of all effective interval groups and take the maximum value as the pulse interval distribution change component. Among them, 0.05 corresponds to a 5% change in the pulse interval center value, which is consistent with the pulse interval state change condition used above; 0.125 is obtained by dividing 4 pulse intervals by 32 pulse intervals. When 4 pulse intervals in a valid interval group undergo a change in their affiliation, the quantity ratio change condition is met.

[0147] The arrangement feature is first compared with the arrangement periods of adjacent time windows. When the arrangement period changes, the change component of the arrangement feature is recorded as 1; when the arrangement periods are the same, the repetition degree of the arrangement in the later time window is subtracted from the repetition degree of the earlier time window, the absolute value of the difference is taken, and then divided by 0.125. The result is taken as the change component of the arrangement feature. 0.125 corresponds to a change in the arrangement relationship of 4 intervals out of 32 pulse intervals, thus maintaining the same judgment criterion as the minimum repetition count of the effective interval group.

[0148] After completing the above four change calculations, the center frequency change component, bandwidth change component, pulse interval distribution change component, and arrangement characteristic change component are compared. The component with the largest value is taken as the state change degree of two adjacent time windows. The state change threshold is set to 1. As can be seen from the aforementioned normalization process, when the center frequency change reaches 5 kHz, the bandwidth change reaches 10%, the pulse interval center value change reaches 5%, the number of effective interval groups changes, the ratio of the number of interval groups changes by 0.125, or the arrangement relationship changes by 0.125, the corresponding change component reaches 1, thus satisfying the parameter state change judgment condition.

[0149] To exclude version switching caused by abnormal pulses within a single time window, the change location is not immediately determined after the state change degree reaches or exceeds 1; instead, subsequent time windows are calculated. When the state change degree of three consecutive adjacent time windows all reaches or exceeds 1, it is confirmed that the parameter state has changed, and the starting pulse position of the next time window corresponding to the first of the three consecutive threshold exceedances is determined as the parameter state change location. This process is consistent with the condition mentioned above for confirming parameter changes after three consecutive statistical time windows. After obtaining the parameter state change locations in chronological order, parameter stability intervals are divided based on adjacent parameter state change locations. The first parameter stability interval is from the start of the data to the first parameter state change location; the second parameter stability interval is from the first parameter state change location to the second parameter state change location; and so on, with the last parameter state change location to the end of the currently processed data constituted the last parameter stability interval. Pulses belonging to the parameter state change location are assigned to the next parameter stability interval, ensuring that there are no duplicate pulse numbers between adjacent stability intervals.

[0150] For each stable interval of a parameter, the state parameter set is recalculated. Instead of directly using the parameters of a single sliding time window, the center frequencies of all valid pulses within the stable interval are summed and divided by the number of valid pulses to obtain the center frequency of that stable interval. Similarly, the bandwidth of the stable interval is obtained by summing all bandwidths and dividing by the number of valid pulses. All pulse intervals within the interval are reclassified into effective interval groups according to a 0.02 classification condition, and the center value and proportion of each effective interval group are recorded. Then, the repetition degree is recalculated and arranged according to candidate periods of 1 to 8 intervals. The period length with the highest repetition degree and its corresponding repetition degree are taken as the arrangement parameters for that stable interval. The center frequency, bandwidth, center value of the effective interval group, proportion of the effective interval group, arrangement period, and arrangement repetition degree together constitute the state parameter set for that stable interval.

[0151] The update of the reference IQ sequence is performed within the corresponding parameter stability interval. The pulse arrival time window is predicted according to the pulse interval parameter of this stability interval, and the actual pulse boundary is re-acquired using the envelope detection method determined in step three. Candidate reference sequences are truncated according to the actual boundary, and a normalized cross-correlation threshold of 0.80 is used for screening. Sixteen candidate reference sequences satisfying the correlation conditions are obtained in chronological order. Subsequently, integer sampling position compensation and 21st-order fractional sampling delay compensation are performed based on the cross-correlation peak position, and phase compensation is performed based on the complex phase of the cross-correlation peak. The 16 time- and phase-aligned reference sequences are summed point by point and divided by 16 to obtain the reference signal sequence corresponding to this stability interval. The reference signal sequence is orthogonally down-converted according to the center frequency of the state parameter set, and the decimation parameters are determined according to the bandwidth of this stability interval. The decimation factor starts at 1 and increases sequentially. Each increase is calculated by dividing the current decimation factor by 4 million samples per second to obtain the decimated sampling rate. If the decimated sampling rate still reaches or exceeds 2.5 times the current bandwidth, the decimation factor continues to increase. The increase stops when the next integer decimation factor causes the sampling rate to fall below 2.5 times the bandwidth, and the previous integer is used as the final decimation factor. The value obtained by dividing 4 million samples per second by this final decimation factor is used as the baseband sampling rate corresponding to the updated reference IQ sequence and is saved along with the reference IQ sequence.

[0152] Each parameter stability interval generates a corresponding version. The first parameter stability interval corresponds to version 001, the second parameter stability interval corresponds to version 002, and so on, incrementing the version number by 1 for each new parameter state change. Each version simultaneously records the reference IQ sequence, baseband sampling rate, center frequency, bandwidth, effective interval group center value and its proportion, arrangement period, arrangement repetition degree, stability interval start pulse number, and stability interval end pulse number. The versions are stored sequentially according to the chronological order of the parameter stability intervals, thus forming a reference IQ sequence set. When the receiving station acquires any version, it can directly read the center frequency, bandwidth, pulse interval distribution, arrangement parameters, and baseband sampling rate corresponding to the reference IQ sequence, and perform baseband conversion and matched filtering processing accordingly.

[0153] Example 6: S5 includes: A set of reference IQ sequences with version identifiers and associated center frequency, bandwidth and baseband sampling rate is sent to the receiving station, and a second sampling sequence containing direct components and target echoes is acquired. Based on the center frequency, bandwidth and baseband sampling rate of each version, the second sampling sequence is subjected to quadrature downconversion and decimation filtering to obtain the second baseband IQ sequence corresponding to the reference IQ sequence of each version. Matched filter coefficients were constructed using the conjugate inversion sequences of the reference IQ sequences for each version, and matched filtering was performed on the corresponding second baseband IQ sequences to obtain the matched response sequences. Extract the direct path matching peaks of continuous pulses from each matching response sequence, and generate version consistency evaluation values ​​based on the peak amplitude to background response ratio and cross-pulse position dispersion. The reference IQ sequence corresponding to the version with the highest version consistency evaluation value that meets the preset threshold is determined as the current valid reference IQ sequence, and the corresponding second baseband IQ sequence is determined as the current received baseband IQ sequence.

[0154] In the above scheme, the receiving station receives the set of reference IQ sequences formed in step four. Each reference IQ sequence is transmitted along with a version identifier, center frequency, bandwidth, baseband sampling rate, and number of effective sampling points, and a fixed correspondence is established according to the version identifier. The receiving station continuously samples the received signal using the same time base as the reference signal acquisition end, with an initial sampling rate of 4 million complex sampling points per second. The resulting second sampling sequence includes the direct component propagating directly from the non-cooperative transmitter to the receiving station, as well as the target echo reaching the receiving station after being scattered by the target.

[0155] The receiving station sequentially reads the center frequency, bandwidth, and baseband sampling rate corresponding to each version, and generates a second baseband IQ sequence for each version. When processing a version, a digital quadrature local oscillator (DQO) is generated based on the center frequency of that version. The DQO is updated point-by-point according to 4 million sampling moments per second. Each complex sampling point in the second sampling sequence is multiplied by the corresponding DQO at that moment, shifting the center frequency of the signal corresponding to that version to the zero-frequency position.

[0156] After completing the quadrature downconversion, a 129th-order finite impulse response (FIR) filter is configured according to the bandwidth recorded for this version. The filter passband extends from zero frequency towards both the positive and negative frequency directions to 0.55 times the bandwidth of this version, and the stopband starts at 0.70 times the bandwidth of this version. These filtering parameters are consistent with those used when generating the corresponding version's reference IQ sequence, ensuring that the reference IQ sequence and the received baseband data have the same effective spectral range. The decimation factor is directly determined by the initial sampling rate and the baseband sampling rate of this version. Specifically, 4 million samples per second are divided by the baseband sampling rate recorded for this version, and the integer result is used as the decimation factor. Before decimation, a 129th-order band-limited filter is performed, and then samples are retained at equal intervals from the filter output according to the decimation factor. For example, if the baseband sampling rate recorded for this version is 500,000 complex samples per second, 4,000,000 divided by 500,000 equals 8, and one sample is retained from every eight consecutive filter output samples. The sampling rate of the processed data is the same as that of the reference IQ sequence for this version, and this data is recorded as the second baseband IQ sequence corresponding to the current version.

[0157] The matched filter coefficients are directly generated from the reference IQ sequence of this version. First, all complex sampling points of the reference IQ sequence are read in the original sampling order. Each complex sampling point is conjugated. Then, the order of the conjugated sampling points is reversed so that the last sampling point of the original reference IQ sequence corresponds to the first position of the matched filter coefficient, and the original first sampling point corresponds to the last position of the matched filter coefficient. This yields the matched filter coefficients for this version.

[0158] The second baseband IQ sequence and the matched filter coefficients are progressively multiplied and added at each sampling position. For each sampling point moved, a data segment with the same length as the number of valid sampling points in the reference IQ sequence is taken. Each complex sampling point within this data segment is multiplied point-by-point by the corresponding matched filter coefficient, and all products are summed. The resulting complex result is used as the matching response at the current sampling position. This process is repeated continuously along the second baseband IQ sequence to form the matching response sequence for this version. Each version uses 16 consecutive valid pulses to evaluate the consistency with the currently received signal. The processing start point for each pulse is determined based on the pulse arrival time sequence obtained in step two, and a 20-millisecond search interval is taken from this start point. The 20-millisecond search length follows the direct path detection range described above. The current baseband sampling rate is multiplied by 0.02 seconds, and the resulting integer number of sampling points is used as the direct path search length for the corresponding version. Therefore, different baseband sampling rate versions all use a 20-millisecond search range on the time scale.

[0159] Within each direct path search interval, the amplitude of the complex value of the matching response is calculated point by point, and the amplitudes of all sampling positions are compared. The sampling point with the largest amplitude is determined as the direct path matching peak of that pulse. The same process is performed on 16 pulses in sequence to obtain the peak amplitude of 16 direct path matching peaks and 16 peak positions.

[0160] Background response is extracted after the end position of the search interval for each pulse's direct path. 1024 matching response sampling points are continuously read. The squares of the in-phase and quadrature components are calculated for each complex sampling point. The sum of these two squares yields the power at that point. The sum of all 1024 power values ​​is divided by 1024, and the square root of the result is taken to obtain the background root mean square amplitude of the pulse. After processing 16 pulses, the sum of the 16 direct path matching peak amplitudes is divided by 16 to obtain the average peak amplitude of the direct path for this version. The sum of the 16 background root mean square amplitudes is divided by 16 to obtain the average background response amplitude. The ratio of peak amplitude to background response is obtained by dividing the average peak amplitude of the direct path by the average background response amplitude. The threshold for determining the peak amplitude to background response ratio is set to 4. This threshold is determined using 60 seconds of calibration data before the receiving station begins detection. During calibration, the non-cooperative transmitter was kept in a receptive state. Thirty-two pulses were continuously selected from the direct path data where no target echo was found. The peak amplitude of the direct path and the root mean square amplitude of the background were calculated using the method described above. Candidate ratio thresholds were successively set to 2, 3, and 4 for testing. When the threshold was set to 4, all 32 direct path pulses formed peak values ​​that met the conditions, while no response reaching this ratio was formed within the 1024-point background interval. Therefore, 4 was saved as the amplitude threshold used for version determination.

[0161] The cross-pulse position dispersion is calculated based on the positions of 16 direct path matching peaks. The 16 peak positions are converted into sampling point offsets relative to their respective pulse processing start points. The maximum and minimum values ​​are then obtained from these offsets, and the position dispersion is calculated by subtracting the minimum value from the maximum. If the position dispersion does not exceed two sampling points, the current version is considered to meet the direct path peak position consistency condition. The two-sampling-point position threshold is determined based on the receiver link calibration results. Before the receiving station is put into operation, 32 pulses are continuously processed using a fixed direct path signal. After completing unified sampling clock calibration and the aforementioned down-conversion, filtering, and decimation, the position variation range of the 32 direct path matching peaks relative to the pulse start points is two sampling points. Therefore, two sampling points are determined as the cross-pulse position dispersion threshold. This threshold is fixed and used for direct path peak position determination in all subsequent versions. To simultaneously reflect the stability of peak amplitude and peak position in direct path matching, amplitude evaluation results and position evaluation results are calculated separately. The amplitude evaluation result is obtained by dividing the peak amplitude to background response ratio by 4. When the peak amplitude to background response ratio is 4, the amplitude evaluation result is 1, and when the ratio is greater than 4, the evaluation result is greater than 1.

[0162] The location evaluation result is determined based on the location dispersion. First, the location dispersion is increased by 1, then 3 is divided by the increased value. When the location dispersion is 0, the location evaluation result is 3; when the location dispersion is 1 sampling point, the location evaluation result is 1.5; when the location dispersion is 2 sampling points, the location evaluation result is 1; when the location dispersion reaches 3 sampling points, the location evaluation result drops to 0.75. Thus, a location threshold of 2 sampling points corresponds to an evaluation result of 1.

[0163] The amplitude evaluation result and the position evaluation result are compared, and the smaller value is taken as the consistency evaluation value of this version. The consistency evaluation threshold is set to 1. Only when the peak amplitude to background response ratio reaches 4 and the cross-pulse position dispersion does not exceed 2 sampling points, will both evaluation results simultaneously reach 1, and thus the consistency evaluation value will reach or exceed 1. This process ensures that both amplitude and position conditions participate in version selection; if either condition is not met, the corresponding version cannot be determined as a valid version.

[0164] The same process is followed to calculate the consistency evaluation value of all versions in the reference IQ sequence set. First, versions with a consistency evaluation value less than 1 are deleted. Then, the consistency evaluation values ​​of the remaining versions are compared, and the version with the highest value is determined as the current valid version. If two versions have the same consistency evaluation value, the peak amplitude to background response ratio of the two versions is compared, and the version with the larger ratio is determined as the current valid version. If the peak amplitude to background response ratio is still the same, the version with the larger version number is used, thus obtaining a unique current valid version. After determining the current valid version, the corresponding reference IQ sequence is determined as the current valid reference IQ sequence, and the second baseband IQ sequence generated using the center frequency, bandwidth, and baseband sampling rate of this version is determined as the current received baseband IQ sequence. This completes the matching determination between the reference version and the actual direct path signal of the receiving station, providing corresponding received baseband data for subsequent target echo matched filtering and slow-time Doppler processing using the current valid reference IQ sequence.

[0165] Example 7: S6 includes: Based on the current valid reference IQ sequence, the matched filter coefficients are constructed, and the continuous pulses in the current received baseband IQ sequence are matched and filtered to obtain multiple matched response sequences. The actual slow time position corresponding to each matched response sequence is determined based on the pulse arrival time sequence, and the delay-non-uniform slow time response data is organized according to the propagation delay dimension and the actual slow time position. For each propagation delay unit, a non-uniform Doppler transformation is performed on the delay-non-uniform slow-time response data based on the corresponding actual slow-time position to obtain the range-Doppler response matrix; The target detection threshold is determined based on the background response statistics of the distance-Doppler response matrix, and target response units that exceed the target detection threshold are filtered out. Based on the delay difference of the target response unit relative to the matching peak of the direct path and its Doppler position, the target propagation distance difference information and target Doppler information are determined, and the target detection results are obtained.

[0166] In the above scheme, step six uses the currently valid reference orthogonal baseband sequence determined in step five and the corresponding currently received baseband sequence for target detection. Each time, 64 consecutive valid pulses are processed. The arrival times of the 64 pulses are directly read from the pulse arrival time sequence formed in step two and corrected in step three. Each pulse maintains its actual time position, and no equal-interval reconstruction is performed on the pulse intervals.

[0167] First, matched filter coefficients are constructed based on the current valid reference orthogonal baseband sequence. All complex sampling points in the reference sequence are read in sampling time order. The sign of the orthogonal component of each sampling point is changed while the in-phase component remains unchanged, completing the complex conjugation process. Then, the conjugated sampling points are rearranged in the opposite direction to the original sampling order, so that the last sampling point of the original reference sequence is at the beginning of the matched filter coefficients, and the first sampling point of the original reference sequence is at the end of the matched filter coefficients, thus forming the matched filter coefficients corresponding to the current reference waveform.

[0168] For each of the 64 valid pulses, a corresponding data segment is read from the current received baseband sequence. When processing one pulse, the number of sampling points of the matched filter coefficients is used as the sliding data length. Starting from the first sampling position of the pulse data segment, an equal-length data segment is extracted. The extracted data is multiplied point-by-point by the complex sample value of the matched filter coefficients at the same index position, and all products are added together. The resulting complex number is used as the matched response at that position. The starting position of the data is then moved forward by one sampling point and the calculation is repeated until all valid positions of the pulse data segment are completed, thus forming a matched response sequence. The 64 pulses are processed in the same way to obtain 64 matched response sequences. The 64 matched response sequences are arranged according to a uniform propagation delay position. Each matched response sequence uses the starting sampling position of the corresponding pulse data segment as the delay starting point and divides the propagation delay units sequentially according to the sampling point order. The time interval between two adjacent propagation delay units is determined by the baseband sampling rate of the current valid version, specifically 1 second divided by the baseband sampling rate. For example, when the current baseband sampling rate is 500,000 complex sampling points per second, the time interval between two adjacent propagation delay units is 2 microseconds.

[0169] The slow-time position is not replaced by the pulse sequence number, but is determined by the actual arrival time of each pulse. The arrival time of the first pulse out of 64 pulses is taken as the starting point of the slow time, and the slow-time position of the first pulse is recorded as 0. The slow-time position of the second pulse is obtained by subtracting the actual arrival time of the first pulse from the actual arrival time of the second pulse. The slow-time positions of the remaining pulses are obtained by subtracting the arrival time of the first pulse from their own actual arrival time. In this way, each propagation delay unit corresponds to 64 complex matched responses with definite actual time positions, thereby forming delay-nonuniform slow-time response data.

[0170] Non-uniform Doppler processing is performed independently for each propagation delay unit. In this embodiment, the Doppler search range is set to -20 Hz to +20 Hz, and the frequency interval between adjacent search positions is 0.02 Hz, thus a total of 2001 Doppler search positions are set. For one Doppler search position, the complex matched response of the current propagation delay unit on 64 pulses is read sequentially. When processing the first pulse, its slow time position is 0, and the phase compensation angle is 0. When processing the remaining pulses, the current Doppler search frequency is multiplied by the actual time difference between the pulse and the first pulse. The resulting value represents the number of phase cycles accumulated within that time difference. This value is then multiplied by 360 degrees to obtain the corresponding phase angle, and rotation compensation is performed on the complex matched response of the pulse in the opposite direction. After completing the phase compensation for 64 pulses, the 64 complex results are added together.

[0171] The in-phase and quadrature components are obtained from the summed complex result. The value obtained by multiplying the in-phase component by itself is added to the value obtained by multiplying the quadrature component by itself, and this is taken as the response power of the current propagation delay unit at the current Doppler search position. The search is completed sequentially at frequency intervals of 0.02 Hz from -20 Hz to +20 Hz. The same process is then performed on all propagation delay units to form a range-Doppler response matrix arranged by propagation delay position and Doppler frequency position.

[0172] The target detection threshold is calculated based on the actual background response around the target unit. For a target unit in the matrix, two adjacent guard units are skipped in front of the target unit along the propagation delay direction, followed by eight consecutive training units; similarly, two guard units are skipped behind the target unit along the propagation delay direction, followed by eight consecutive training units. In the Doppler direction, one guard unit is skipped on the negative frequency side, followed by four consecutive training units; similarly, one guard unit is skipped on the positive frequency side, followed by four consecutive training units. Thus, each target unit uses a fixed 24 training units to calculate the background power. Target determination is not performed at matrix edges where 24 training units cannot be completely obtained.

[0173] The response powers of all 24 training units are summed and then divided by 24 to obtain the average background power at the target location. The target detection threshold is obtained by multiplying this average background power by 12. The coefficient 12 is determined using 60 seconds of target-free background data before the receiving station is put into operation. During calibration, the range-Doppler response matrix is ​​regenerated according to the 64-pulse length, 0.02 Hz Doppler interval, and 24 training units used in actual detection. The ratio between the response power and the corresponding average background power at each non-direct path location is calculated. The maximum ratio obtained from the 60-second calibration data is 10.6. 10.6 is rounded up to obtain 11, and then 1 is added as a detection margin. The final threshold coefficient is determined to be 12, and this value is stored in the receiving station's detection parameters.

[0174] During actual detection, the response power of the unit to be detected is compared with the corresponding target detection threshold. When the response power of the unit to be detected reaches or exceeds the target detection threshold, the unit is identified as a target candidate response unit; if it is lower than the target detection threshold, the unit is not retained. The direct path matching peak position determined in step five is used as a propagation delay reference position and is not involved in the target candidate response determination.

[0175] When matching responses for the same target occupy adjacent cells in the range-Doppler matrix, these adjacent candidate response cells are merged. Candidate response cells that are adjacent in the propagation delay direction, Doppler direction, or diagonal direction are assigned to the same response region. The response power of all candidate response cells within the same response region is compared, and only the cell with the highest response power value is retained as the target response location for that region, thus ensuring that one response region corresponds to one target detection result.

[0176] After determining the target response location, the propagation delay unit number of the target response is read, and the propagation delay unit number of the direct path matching peak within the same 64-pulse processing cycle is also read. The propagation delay unit number of the direct path matching peak is subtracted from the propagation delay unit number of the target response to obtain the difference in the number of propagation delay units between the two; this number is then multiplied by 1 second and divided by the unit time length obtained from the current baseband sampling rate to obtain the delay difference between the target response and the direct path matching peak.

[0177] Multiplying this delay difference by 299,792,458 meters per second yields the increase in distance between the propagation path of the non-cooperative transmitter after being scattered by the target and the direct propagation path of the non-cooperative transmitter to the receiving station. Since this value represents the path difference between the scattered propagation path and the direct propagation path, it is not divided by 2 according to the round-trip propagation relationship of a monostatic radar.

[0178] The target's Doppler information is directly determined by the Doppler search position where the target response is located. The Doppler search starts at -20 Hz, increasing by 0.02 Hz for each additional search position. After determining the target response's position, the difference in index between that position and the first search position is multiplied by 0.02 Hz, and then added to -20 Hz to obtain the target's corresponding Doppler frequency. For example, if the target response is located at the 751st search position, the 751st position differs from the first position by 750 search intervals. Multiplying 750 by 0.02 Hz gives 15 Hz, and adding 15 Hz to -20 Hz yields -5 Hz. Therefore, the target's Doppler frequency is -5 Hz.

[0179] Finally, the propagation delay unit corresponding to the target response, the delay difference relative to the matching peak of the direct path, the propagation distance difference, the Doppler frequency, and the response power are output as the detection results for the target. When there are two or more non-adjacent target response regions in the same range-Doppler response matrix, the corresponding propagation distance difference and Doppler frequency are calculated according to the above processing procedure, and target detection results are formed respectively.

[0180] Example 8: The present invention also provides a system applicable to the high-frequency radar network detection method based on blind extraction of signals from non-cooperative transmitters as described above, comprising: The front-end acquisition device is deployed in the vicinity of the non-cooperative emission source to acquire a first sampling sequence containing signals from multiple radiation sources. The first sampling sequence is processed in a frame-by-frame frequency domain. Candidate frequency band trajectories are formed based on the overlap relationship of the frequency bands occupied by adjacent frames. The target frequency band trajectory is selected based on the pulse interval structure of the sub-band sequence corresponding to each candidate frequency band to obtain the center frequency, bandwidth and pulse arrival time sequence. The front-end acquisition device is also used to establish a pulse timing model based on the pulse arrival time sequence, extract candidate reference sequences according to the detected actual pulse boundaries, perform cross-correlation screening, time and phase alignment and coherent synthesis on the candidate reference sequences, and perform down-conversion and decimation filtering to obtain a reference IQ sequence; update the reference IQ sequence according to the changes in signal parameters within a continuous time window to form a set of reference IQ sequences with version identifiers and corresponding signal parameters; A communication link for transmitting the reference IQ sequence set to the receiving station; The receiving station is used to acquire a second sampling sequence containing the direct component and the target echo. It performs baseband conversion on the second sampling sequence according to the corresponding signal parameters of each version, and performs matched filtering using the reference IQ sequence of each version. It determines the current valid reference IQ sequence and the corresponding received baseband IQ sequence based on the amplitude of the matching peak of the direct path and the consistency of the cross-pulse position. The receiving station is also used to perform matched filtering on the received baseband IQ sequence using the currently valid reference IQ sequence, perform non-uniform slow-time Doppler transformation according to the pulse arrival time sequence, and obtain the target detection result based on the delay difference of the target response relative to the matching peak of the direct path and the Doppler position.

[0181] In the above scheme, the front-end acquisition equipment is located in the vicinity of the non-cooperative transmitter. The front-end acquisition equipment includes a high-frequency receiving antenna, a filter bank, a unified sampling clock, an analog-to-digital converter, a digital processing unit, and a data storage unit. The high-frequency receiving antenna receives electromagnetic signals in the range of 3 MHz to 30 MHz. The filter bank is divided into 27 channels with a 1 MHz passband width; channel 1 corresponds to 3 MHz to 4 MHz, and channel 27 corresponds to 29 MHz to 30 MHz. All channels share a 40 MHz reference clock, which is divided into 4 million sampling moments per second. The same frame synchronization pulse is used as the sampling start time, ensuring that each channel corresponds to the same sampling time at the same sampling number.

[0182] The complex sampled data output from each channel is arranged into a frame of 4096 sampling points, with adjacent frames shifted forward by 2048 sampling points. Each frame's data is multiplied by a 4096-point Hanning window and then subjected to a 4096-point frequency domain transformation. The squared real part and squared imaginary part of each frequency unit are added together to obtain the spectral power of that frequency unit. The 1024 spectral powers within the 1 MHz sub-band are arranged in ascending order, and the first 256 powers are used to calculate the background average power. The average absolute deviation of these 256 background powers from the background average power is then calculated. This average is multiplied by 4 and added to the background average power; the resulting power value is used as the spectral threshold for the current frame. When three consecutive frequency units reach this threshold, they are merged into the same candidate occupancy frequency band.

[0183] Candidate occupied frequency bands in two adjacent frames are connected according to their frequency overlap range and center frequency deviation. First, the frequency widths commonly covered by the two bands are obtained, then divided by the smaller of the two band widths to get the frequency overlap rate. Simultaneously, the center frequencies of the two bands are subtracted and the absolute value is taken to obtain the center frequency deviation. When the frequency overlap rate reaches 0.60 and the center frequency deviation does not exceed 1953.125 Hz, the two bands are connected into the same trajectory. After this condition is met for 8 consecutive frames, the corresponding trajectory is retained as a candidate frequency band trajectory.

[0184] For each candidate frequency band trajectory, the front-end acquisition device sums the center frequencies of each frame within the trajectory and divides by the number of valid frames to obtain the trajectory center frequency; it also sums the bandwidths of each frame and divides by the number of valid frames to obtain the trajectory bandwidth. Digital down-conversion is performed according to the trajectory center frequency, and a 129th-order finite impulse response filter is used to extract the corresponding sub-band. The envelope power is calculated point-by-point for each sample of the sub-band sequence. Using 4096 envelope powers as one statistical unit, the first 1024 sorted powers are used to calculate the background average power and the average absolute deviation. The average absolute deviation is multiplied by 4 and added to the background average power to obtain the envelope detection threshold. When the envelope power reaches the detection threshold for 8 consecutive sampling points, the first sampling point is recorded as the pulse arrival position; after the pulse begins, when the envelope power is below the detection threshold for 8 consecutive sampling points, the first sampling point is recorded as the pulse end position. The front-end acquisition device continuously acquires 33 pulse arrival times, subtracting the previous pulse arrival time from the next pulse arrival time to form a 32-pulse interval. The 32 pulse intervals are categorized according to a 2% relative deviation condition, and their repetition relationships are detected and arranged according to interval lengths of 1 to 8. When at least 30 pulse intervals enter the effective interval group and the timing consistency characteristic value reaches 0.93, the corresponding candidate frequency band trajectory is determined as the target frequency band trajectory.

[0185] After the target frequency band trajectory is determined, the front-end acquisition device adds up the power at the same frequency position in the corresponding 32 frames of spectrum of the trajectory and divides by 32 to form the average power spectrum, and then re-determines the spectrum threshold according to the aforementioned background statistical process. The leftmost frequency that continuously reaches the threshold is taken as the left boundary of the target frequency band, and the rightmost frequency is taken as the right boundary; the center frequency is obtained by adding the two boundary frequencies and dividing by 2, and the bandwidth is obtained by subtracting the left boundary frequency from the right boundary frequency. At the same time, the arrival times of the confirmed pulses are saved in chronological order.

[0186] The front-end acquisition device establishes a pulse timing model using the arrival times of 33 consecutive pulses. The difference between adjacent pulses yields 32 pulse intervals. The interval values ​​belonging to the current valid interval group are summed and divided by the number of intervals in that group to obtain the pulse interval reference value. Using the arrival time of the first pulse as the time reference, the predicted arrival times of subsequent pulses are incremented by the pulse interval reference value according to the pulse sequence number. A search range of 0.05 times the pulse interval reference value is retained before and after each predicted time. Within this range, the actual pulse start and end boundaries are redefined according to the aforementioned envelope threshold.

[0187] Thirty-two candidate reference sequences were extracted from the first sampling sequence according to the actual boundaries. The energy of all complex sampling points of each sequence was calculated and accumulated, and the sequence with the largest total energy was determined as the reference sequence. The remaining candidate reference sequences were subjected to normalized cross-correlation calculation with the reference sequence within a range of 64 sampling points before and after them, and the sequence was retained when the correlation peak reached 0.80. After obtaining 16 reference sequences that met the conditions in chronological order, integer sampling position correction was performed based on the cross-correlation peak position, and a 21st-order fractional-delay finite impulse response filter was used to compensate for the fractional sampling delay. Subsequently, the phase of each sequence was rotated according to the complex phase of the cross-correlation peak, so that the 16 reference sequences maintained a consistent time and phase reference at the same sampling position. After alignment, within the common effective sampling interval of the 16 reference sequences, the 16 complex sampling values ​​at the same sampling position were added and divided by 16, thus forming the reference signal sequence. The reference signal sequence was orthogonally down-converted according to the target center frequency, and then band-limited processing was performed using a 129th-order finite impulse response filter. The sampling multiple starts from 1 and increases incrementally. Each increase is calculated by dividing the current sampling multiple by 4 million sampling points per second. The sampling multiple is only increased when the obtained sampling rate reaches 2.5 times the target bandwidth. If the next integer multiple results in a sampling rate that is lower than 2.5 times the target bandwidth, the previous integer multiple is used to complete the sampling, and the baseband sampling rate after sampling is recorded.

[0188] The front-end acquisition device uses 32 consecutive valid pulses to form a parameter statistical time window, with adjacent time windows shifting by 16 pulses. For each time window, the center frequency, bandwidth, pulse interval distribution, and interval arrangement parameters are calculated. When the center frequency changes by 5 kHz, the bandwidth changes by 10%, the pulse interval center value changes by 5%, the effective interval group composition changes, or the degree of repetition changes by 0.125 between adjacent time windows, the corresponding state change component reaches 1. The maximum value among the change components is taken as the state change degree. When the state change degree reaches 1 for three consecutive time windows, the starting pulse position of the first threshold-exceeding time window is determined as the parameter state change position. A parameter stability interval is formed between adjacent parameter state change positions. For each parameter stability interval, the center frequency, bandwidth, pulse interval distribution, and arrangement parameters are recalculated, and a reference orthogonal baseband sequence is regenerated according to the processing order of actual boundary truncation, 0.80 cross-correlation filtering, time alignment, phase compensation, 16-segment coherent synthesis, orthogonal downconversion, and decimation filtering. The first parameter stability interval is assigned version 001, and subsequent version numbers are added sequentially according to the parameter state change order. Each version simultaneously saves the center frequency, bandwidth, baseband sampling rate, pulse interval distribution, arrangement parameters, and the start and end pulse numbers of the stable interval, thus forming a set of reference sequences.

[0189] The communication link sends the reference sequence set and corresponding parameters to the receiving station. This embodiment uses a gigabit Ethernet fiber optic link for transmission. Each version of the data packet sequentially writes the version number, center frequency, bandwidth, baseband sampling rate, number of valid sampling points in the reference sequence, start pulse number of the stable interval, and end pulse number, followed by the corresponding complex sampled data. The sending end calculates a checksum for the data packet and appends it to the end of the data. The receiving station recalculates the checksum of the received data. If the two checksums match, the version is written to the version buffer; otherwise, the data packet is deleted and not used as input for subsequent matching processing.

[0190] The receiving station continuously acquires a second sampling sequence at 4 million complex sampling points per second. The second sampling sequence includes the direct component propagating directly from the non-cooperative transmitter to the receiving station, as well as the target echo entering the receiving station after being scattered by the target. For each reference version in the version buffer, the receiving station reads the center frequency of that version, performs orthogonal down-conversion, then performs band-limiting processing using a 129th-order finite impulse response filter according to the corresponding bandwidth, and determines the decimation factor according to the baseband sampling rate recorded for that version, so that the resulting second baseband orthogonal sequence has the same sampling rate as the corresponding reference sequence.

[0191] The receiving station first performs conjugation processing on all complex sampling points of each version of the reference sequence, then arranges them in reverse chronological order to form the corresponding matched filter coefficients. The second baseband sequence is then multiplied and added with the matched filter coefficients in a step of one sampling point to obtain the matched response sequence for that version. Sixteen consecutive valid pulses are selected, and a 20-millisecond direct path search interval is set starting from the arrival time of each pulse. The matched response with the largest amplitude within each search interval is taken as the direct path matching peak.

[0192] The average peak amplitude of the direct path is obtained by summing the amplitudes of the 16 matching peaks and dividing by 16. After each pulse passes through the direct path search interval, 1024 matching response sampling points are continuously acquired. The root mean square amplitude of these 1024 points is calculated, and the average background response amplitude is obtained by summing the average peak amplitudes of the 16 pulses and dividing by 16. The peak amplitude to background response ratio is then obtained by dividing the average peak amplitude of the direct path by the average background response amplitude.

[0193] Simultaneously, the positions of the 16 direct path matching peaks relative to their respective pulse origins are compared, and the cross-pulse position dispersion is obtained by subtracting the minimum sampling point number from the maximum sampling point number. The peak amplitude to background response ratio is divided by 4 to obtain the amplitude evaluation result; the position evaluation result is obtained by dividing 3 by the value after increasing the position dispersion by 1; the smaller value of the two evaluation results is taken as the version consistency evaluation value. When the version consistency evaluation value reaches 1 and is the largest among all valid versions, the corresponding reference sequence is determined as the current valid reference sequence, and the second baseband orthogonal sequence corresponding to this version is determined as the current receiving baseband sequence. The receiving station then continuously selects 64 valid pulses for target processing. Matched filter coefficients are constructed using the conjugate inversion result of the current valid reference sequence, and matched filtering is performed on the 64 pulses respectively to obtain 64 matched response sequences. Each matched response sequence is arranged according to the propagation delay sampling position, and the actual arrival time of each of the 64 pulses is subtracted from the arrival time of the first pulse to obtain 64 actual slow time positions.

[0194] For each propagation delay unit, Doppler searches are performed sequentially from -20 Hz to +20 Hz at 0.02 Hz intervals. For each search frequency, the frequency is multiplied by the actual slow time of 64 pulses, converted into the corresponding phase angle, and the complex matched responses of each pulse are rotated in opposite directions. The 64 compensated complex responses are summed, and the squares of the in-phase and quadrature components of the accumulated results are summed to obtain the response power of the current delay unit at the current Doppler position. After calculating all delay positions and 2001 Doppler positions, a range-Doppler response matrix is ​​formed. During target detection, 24 training units are used to calculate the background power for each target unit. Eight training units are taken before and after the delay direction, with two guard units set on each side; four training units are taken on each side of the Doppler direction, with one guard unit set on each side. The response power of the 24 training units is summed and divided by 24 to obtain the average background power, which is then multiplied by 12 to form the target detection threshold. When the response power of the unit to be detected reaches the target detection threshold, it is retained as the target response unit.

[0195] Adjacent target response units along the propagation delay direction, Doppler direction, and diagonal direction are merged. Within each adjacent region, the unit with the highest response power is retained. The propagation delay unit number corresponding to the matching peak of this target response unit and the direct path is read. The sampling time is obtained by multiplying the difference between the two numbers by 1 second and dividing by the current baseband sampling rate, yielding the delay difference between the target response and the direct path. This delay difference is then multiplied by 299,792,458 meters per second to obtain the increased propagation distance of the target scattering path relative to the direct propagation path. The frequency value corresponding to the Doppler search position of the target response is used as the target Doppler information, and together with the propagation distance difference and response power, forms the target detection result.

[0196] Example 9: This embodiment provides an apparatus suitable for the high-frequency radar network detection method based on blind extraction of signals from non-cooperative transmitters described above, comprising: The sampling module is used to acquire the first sampling sequence of signals from multiple radiation sources in the vicinity of a non-cooperative emission source; The frequency band filtering module is used to perform frequency domain transformation on the first sampling sequence in frames, form candidate frequency band trajectories according to the overlap relationship of the frequency bands occupied by adjacent frames, and filter the target frequency band trajectory according to the pulse interval structure of the sub-band sequence corresponding to the candidate frequency band to obtain the center frequency, bandwidth and pulse arrival time sequence. The reference generation module is used to establish a pulse timing model based on the pulse arrival time sequence and detect the actual pulse boundary. It then extracts a candidate reference sequence according to the actual pulse boundary and obtains a reference IQ sequence after cross-correlation screening, time and phase alignment, coherent synthesis, down-conversion and decimation filtering. The version update module is used to determine the position of parameter state change based on the changes in center frequency, bandwidth and pulse interval characteristics within a continuous time window, and update the reference IQ sequence to obtain a set of reference IQ sequences with version identifiers and corresponding signal parameters. The version confirmation module is used to perform baseband conversion on the second sampling sequence collected by the receiving station according to the signal parameters corresponding to each version, perform matched filtering using the reference IQ sequence of each version, and determine the current valid reference IQ sequence and the corresponding receiving baseband IQ sequence based on the amplitude of the matching peak of the direct path and the consistency of the cross-pulse position. The target detection module is used to perform matched filtering on the received baseband IQ sequence using the currently valid reference IQ sequence, perform non-uniform slow-time Doppler transformation according to the pulse arrival time sequence, and obtain the target detection result based on the delay difference of the target response relative to the matching peak of the direct path and the Doppler position.

[0197] In the above scheme, the device transmits data in the following processing order: sampling module, frequency band filtering module, reference generation module, version update module, version confirmation module, and target detection module. The sampling module outputs the first sampling sequence and channel identifier; the frequency band filtering module outputs the center frequency, bandwidth, and pulse arrival time sequence; the reference generation module outputs the reference IQ sequence and the corresponding baseband sampling rate; the version update module outputs a set of reference IQ sequences with version identifiers; the version confirmation module outputs the currently valid reference IQ sequence and the corresponding received baseband IQ sequence; and the target detection module generates the target propagation distance difference and target Doppler information based on the above data. Each module uses a unified sampling time number and pulse number, ensuring that data from different processing stages can be correlated according to the same pulse and the same sampling position.

[0198] The sampling module receives high-frequency signals in the 3 MHz to 30 MHz range from the vicinity of a non-cooperative transmitter. The received signal is divided into 27 1 MHz sub-bands by a 27-channel filter bank. The first channel corresponds to 3 MHz to 4 MHz, extending sequentially to the 27th channel, which corresponds to 29 MHz to 30 MHz. The 27 channels share a 40 MHz reference clock, which is then divided by ten to form a sampling clock with 4 million complex samples per second. Each channel simultaneously receives the frame synchronization signal, and the arrival time of the frame synchronization signal is used as the starting point for the sampling number.

[0199] Each sub-band sampling sequence is divided into 4096 sampling points per frame, with adjacent frames shifted forward by 2048 sampling points. The sampling module calculates the frame power for each frame by calculating the squares of the in-phase and quadrature components of the 4096 complex sampling points, adding the two squares to obtain the power of each sampling point, and then summing the power of the 4096 sampling points and dividing by 4096. For each channel, 256 frame powers are continuously acquired, arranged in ascending order, and the power of the first 64 frames is summed and divided by 64 to obtain the background average power. The absolute value of the difference between the power of these 64 frames and the background average power is then calculated, and the maximum value is taken as the background power fluctuation. The background power fluctuation is multiplied by 3 and added to the background average power. The result is used as the acquisition threshold for that channel. When the frame power of at least 6 out of 8 consecutive frames reaches the acquisition threshold, the corresponding channel is retained as a valid sub-band.

[0200] The effective subbands are compensated for sampling positions according to pre-stored channel delay calibration parameters. Integer sampling delays are compensated by shifting the corresponding number of sampling points forward or backward, while fractional sampling delays are compensated using a 21st-order fractional delay finite impulse response filter. After compensation, the channel number, subband start frequency, subband end frequency, and sampling time number are written into the corresponding data block and stored sequentially in chronological order, thus forming the first sampling sequence. The frequency band filtering module reads the effective subband data from the first sampling sequence, multiplies every 4096 sampling points by a 4096-point Hanning window, and performs a 4096-point frequency domain transformation. The spectral power of each frequency unit after the frequency domain transformation is obtained by adding the square of the real part and the square of the imaginary part. When 4 million sampling points per second are transformed using a 4096-point frequency domain transformation, the frequency interval between adjacent frequency units is 976.5625 Hz.

[0201] The 1024 spectral powers within the 1 MHz subband are arranged in ascending order. The first 256 powers are summed and divided by 256 to obtain the average background power of the current frame. The absolute value of the difference between each of these 256 powers and the average background power is then calculated, summed, and divided by 256 to obtain the average background fluctuation. This average fluctuation is multiplied by 4 and added to the average background power to obtain the spectral threshold for the current frame. When the spectral power of three consecutive frequency units reaches this threshold, these consecutive frequency units are merged into a candidate occupied frequency band.

[0202] When associating the trajectories of two adjacent candidate occupied frequency bands, the bandwidth of the common coverage area of ​​the two frequency bands is first obtained. Then, this common bandwidth is divided by the smaller value of the bandwidths of the two candidate occupied frequency bands, and the resulting value is used as the frequency overlap rate. At the same time, the center frequencies of the two candidate occupied frequency bands are subtracted and the absolute value is taken to obtain the center frequency deviation. When the frequency overlap rate reaches 0.60 and the center frequency deviation does not exceed 1953.125 Hz, the two candidate occupied frequency bands are assigned to the same trajectory. After eight consecutive frames meet the above conditions, a candidate frequency band trajectory is formed.

[0203] For the candidate frequency band trajectory, the center frequencies of each frame within the trajectory are summed and divided by the number of valid frames to obtain the trajectory center frequency; the bandwidths of each frame are summed and divided by the number of valid frames to obtain the trajectory bandwidth. Down-conversion is performed according to the trajectory center frequency, and then a 129th-order finite impulse response filter is used for band-limited extraction. The envelope power is calculated point-by-point for each sample of the sub-band sequence, and 4096 envelope powers are used as one statistical segment. The first 1024 sorted powers are summed and divided by 1024 to obtain the average envelope background power. The average absolute value of the difference between the 1024 powers and the average background power is then calculated. This average value is multiplied by 4 and added to the average envelope background power to obtain the pulse detection threshold.

[0204] When the envelope power reaches the pulse detection threshold for 8 consecutive sampling points, the first sampling point is determined as the pulse arrival position; after the pulse starts, when 8 consecutive sampling points are below the pulse detection threshold, the first sampling point is determined as the pulse end position. After obtaining 33 consecutive pulse arrival times, the arrival times of adjacent pulses are subtracted sequentially to form 32 pulse intervals. The 32 pulse intervals are classified according to a 2% relative deviation condition. For the current pulse interval, the absolute value of the difference between it and the center value of an existing interval group is calculated, and then divided by the corresponding center value. If the result does not exceed 0.02, it is assigned to that interval group; if all interval groups exceed 0.02, a new interval group is established. Interval groups containing 4 or more pulse intervals are determined as valid interval groups. Subsequently, the repetition relationship is checked according to interval lengths from 1 to 8. When at least 30 pulse intervals are assigned to valid interval groups, and the temporal consistency characteristic value determined by the degree of interval clustering and the degree of repetition reaches 0.93, the corresponding candidate frequency band trajectory is determined as the target frequency band trajectory.

[0205] The frequency band filtering module accumulates the power at the same frequency position in the 32 frames of spectrum corresponding to the target frequency band trajectory, and then divides by 32 to obtain the average spectrum. Based on the aforementioned spectrum threshold calculation method, the effective frequency band of the target is redefined, with the leftmost effective frequency as the left boundary and the rightmost effective frequency as the right boundary. The center frequency is obtained by adding the two boundary frequencies and dividing by 2. The bandwidth is obtained by subtracting the left boundary frequency from the right boundary frequency, and the arrival times of the effective pulses are output to the reference generation module in chronological order.

[0206] The reference generation module continuously reads 33 pulse arrival times and subtracts adjacent arrival times to obtain 32 pulse intervals. The pulse intervals within the current valid interval group are summed and divided by the number of intervals in the group to obtain the pulse interval baseline value. Using the arrival time of the first pulse as the time baseline, the pulse interval baseline value is continuously increased according to the pulse sequence number to form the predicted arrival times of subsequent pulses. A search range of 0.05 times the pulse interval baseline value is set before and after each predicted arrival time, and within this range, the aforementioned envelope detection threshold is used to redetermine the actual start and end boundaries of the pulse.

[0207] Thirty-two candidate reference sequences were extracted according to the actual boundaries. The squares of the in-phase and orthogonal components of all complex sampling points in each sequence were summed, and this summation was applied to all sampling points to obtain the total energy of each candidate reference sequence. The sequence with the highest total energy was selected as the baseline sequence. Other candidate reference sequences were moved point-by-point within a range of 64 sampling points. At each position, the complex sampling values ​​of the overlapping intervals of the two sequences were correlated and accumulated. The result was then normalized using the square root of the product of the energies of the two overlapping sequences, and the maximum value among all moving positions was taken as the normalized cross-correlation peak value. Candidate reference sequences with a correlation peak value reaching 0.80 were retained.

[0208] Sixteen candidate reference sequences satisfying relevant conditions are obtained in chronological order. Integer sampling shift is performed based on the position of the cross-correlation peaks. When the cross-correlation peaks are between integer sampling positions, the fractional sampling deviation is determined using the correlation results of the peak point and one sampling point before and after it, and compensated using a 21st-order fractional-delay finite impulse response filter. Subsequently, the phase difference is determined based on the complex phase of the cross-correlation peaks, and reverse phase compensation is performed on each sequence. After the 16 candidate reference sequences are aligned in time and phase, they are summed point by point within a common effective sampling interval. The accumulated result at each position is then divided by 16 to obtain the reference signal sequence. The reference generation module performs orthogonal down-conversion based on the target center frequency, and then band-limiting processing is performed using a 129th-order finite impulse response filter. The sampling multiple starts from 1 and increases incrementally. Each time, 4 million sampling points per second are divided by the current sampling multiple. When the resulting sampling rate reaches 2.5 times the target bandwidth, the sampling multiple continues to increase. When the next integer multiple causes the resulting sampling rate to be lower than 2.5 times the target bandwidth, the previous integer is used as the final sampling multiple. This forms a reference IQ sequence and records the corresponding baseband sampling rate.

[0209] The version update module uses 32 consecutive valid pulses as one time window, with adjacent time windows shifting by 16 pulses. Within each time window, the center frequency, bandwidth, center value of the effective interval group, proportion of each interval group, arrangement period, and arrangement repetition degree are calculated. The center frequency change condition is met when the absolute value of the center frequency difference between adjacent time windows reaches 5 kHz; the bandwidth change condition is met when the absolute value of the bandwidth difference divided by the bandwidth of the previous time window reaches 0.10; and the pulse timing change condition is met when the pulse interval center value changes by 0.05, the number of effective interval groups changes, the proportion of interval groups changes by 0.125, or the arrangement repetition degree changes by 0.125.

[0210] Each change is normalized according to its corresponding threshold, and the result with the largest value is taken as the state change degree. When the state change degree reaches 1 for three consecutive time windows, the starting pulse position of the first time window exceeding the threshold is determined as the parameter state change position. A parameter stability interval is formed between adjacent parameter state change positions, and the center frequency, bandwidth, pulse interval distribution and arrangement parameters are recalculated within each stability interval. Then, the reference generation module is called to regenerate the corresponding reference IQ sequence.

[0211] The first stable interval of the parameter is assigned version 001, and subsequent version numbers are sequentially increased according to the time order of parameter state changes. Each version simultaneously saves the reference IQ sequence, center frequency, bandwidth, baseband sampling rate, pulse interval distribution parameters, arrangement parameters, and the start and end pulse numbers of the stable interval, forming a reference IQ sequence set.

[0212] The version confirmation module reads the second sampling sequence generated by the receiving station. For each reference IQ version, orthogonal downconversion is performed based on the center frequency of that version. Band-limiting processing is then implemented using a 129th-order finite impulse response filter based on the bandwidth of that version. The decimation factor is obtained by dividing 4 million samples per second by the baseband sampling rate recorded for that version, and decimation is completed, forming a second baseband IQ sequence with the same sampling rate as the reference IQ sequence of that version. Subsequently, the current version reference IQ sequence is conjugated point-by-point and reversed to form the matched filter coefficients. The second baseband IQ sequence is then multiplied and added with the matched filter coefficients in a step of one sample point to obtain the matched response sequence. Sixteen valid pulses are continuously read, and a 20-millisecond direct path search interval is set forward from the arrival time of each pulse. The matched response with the largest amplitude within each interval is determined as the direct path matching peak.

[0213] The average peak amplitude of the direct path is obtained by summing the peak amplitudes of the 16 matched peaks and dividing by 16. After each direct path search interval, 1024 matched response sampling points are continuously read. The squares of the in-phase components and quadrature components of each sampling point are summed, and the average of the 1024 results is calculated and the square root is taken to obtain the background root mean square amplitude of the pulse. The average background root mean square amplitude is obtained by summing the 16 background root mean square amplitudes and dividing by 16. The peak amplitude to background response ratio is obtained by dividing the average peak amplitude of the direct path by the average background response amplitude.

[0214] Simultaneously, the positions of the 16 direct path matching peaks relative to their respective pulse origins are compared, and the cross-pulse position dispersion is obtained by subtracting the minimum sampling point position from the maximum sampling point position. The peak amplitude to background response ratio is divided by 4 to obtain the amplitude evaluation result, and the position evaluation result is obtained by dividing 3 by the position dispersion plus 1. The smaller value of the two evaluation results is taken as the version consistency evaluation value. Versions with an evaluation value of 1 are added to the valid version set. The version with the largest evaluation value is selected from the valid version set, and its reference IQ sequence is determined as the current valid reference IQ sequence. The corresponding second baseband IQ sequence is determined as the current received baseband IQ sequence. The target detection module continuously acquires 64 valid pulses. Based on the current valid reference IQ sequence, the conjugate inversion matched filter coefficients are reconstructed, and matched filtering is performed on the received baseband data corresponding to the 64 pulses to obtain 64 matched response sequences. Each matched response sequence is arranged according to the propagation delay unit corresponding to the sampling position, and the actual arrival time of each pulse is subtracted from the arrival time of the first pulse to obtain 64 actual slow time positions.

[0215] Specifically, for each propagation delay unit, a Doppler search is performed sequentially from -20 Hz to +20 Hz at 0.02 Hz intervals. When processing a search frequency, the frequency is multiplied by 64 actual slow times, then multiplied by 360 degrees to obtain the corresponding phase compensation angle. The complex matched responses of each pulse are then rotated in opposite directions and accumulated. The squares of the in-phase component and the squares of the quadrature component of the accumulated result are added to obtain the response power of the current propagation delay unit and the current Doppler position. After calculating 2001 Doppler positions and all propagation delay units, a range-Doppler response matrix is ​​formed.

[0216] The target detection module obtains 24 training units for each target unit. Eight units are taken before and after the propagation delay direction, with two guard units retained on each side of the target unit. Four units are taken on each side of the Doppler direction, with one guard unit retained on each side. The response power of the 24 training units is summed and divided by 24 to obtain the background average power. This background average power is then multiplied by 12 to obtain the target detection threshold. When the power of a target unit reaches this threshold, it is retained as a target response unit. Adjacent target response units are connected and merged, retaining the unit with the highest response power in each connected region. The propagation delay unit number of the target response unit and the direct path matching peak is read. The sampling time is obtained by multiplying the difference between the two numbers by 1 second and dividing by the current baseband sampling rate to obtain the delay difference between the target response and the direct path matching peak. This delay difference is then multiplied by 299,792,458 meters per second to obtain the increased propagation distance of the target scattering path relative to the direct path. The frequency corresponding to the Doppler search position of the target response is used as the target Doppler information, and together with the propagation distance difference and response power, forms the target detection result.

[0217] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A high-frequency radar network detection method based on blind extraction of signals from non-cooperative transmitters, characterized in that, include: S1. Collect a first sampling sequence containing signals from multiple radiation sources in the vicinity of a non-cooperative emission source; S2. Perform frequency domain transformation on the first sampling sequence in frames, and form candidate frequency band trajectories based on the overlap relationship of the frequency bands occupied by adjacent frames; perform envelope pulse detection on the sub-band sequences corresponding to each candidate frequency band, and filter the target frequency band trajectory based on the consistency of the pulse interval structure to obtain the center frequency, bandwidth and pulse arrival time sequence. S3. Establish a pulse timing model based on the pulse arrival time sequence and determine candidate arrival time windows, and detect the actual pulse boundary within the time window; Candidate reference sequences are extracted according to the actual pulse boundaries, and after cross-correlation screening, time and phase alignment and coherent synthesis, down-conversion and decimation filtering are performed according to the center frequency and bandwidth to obtain the reference IQ sequence. S4. Determine the position of parameter state change based on the changes in center frequency, bandwidth and pulse interval characteristics within the continuous time window, and update the reference IQ sequence accordingly to form a set of reference IQ sequences with version identifiers and corresponding signal parameters. S5. Send the reference IQ sequence set to the receiving station, perform baseband conversion on the second sampling sequence of the receiving station according to the corresponding signal parameters of each version, and perform matched filtering using the reference IQ sequences of each version. Determine the current valid reference IQ sequence and the corresponding receiving baseband IQ sequence based on the amplitude of the matching peak of the direct path and the consistency of the cross-pulse position. S6. Perform matched filtering on the received baseband IQ sequence using the current valid reference IQ sequence, perform non-uniform slow-time Doppler transformation based on the pulse arrival time sequence, and obtain the target detection result based on the delay difference of the target response relative to the matching peak of the direct path and the Doppler position.

2. The high-frequency radar network detection method based on blind extraction of signals from non-cooperative transmitters according to claim 1, characterized in that, S1 includes: The 3-30MHz high-frequency signal received in the vicinity of the non-cooperative transmitter is divided into multiple sub-band signals by a filter bank; Each sub-band signal is synchronously sampled based on a unified sampling clock and divided into frames according to time to obtain multiple sub-band sampling sequences; Calculate the frame power of each sub-band sampling sequence, construct the channel power feature vector, determine the acquisition threshold based on the background power statistics, and filter the effective sub-bands; Based on the pre-obtained channel delay calibration parameters for each effective sub-band, the sampling position of each sub-band sampling sequence is compensated, and the corresponding channel identifier is associated; The multiple sub-band sampling sequences after compensation are established according to the sampling time and stored continuously to form a first sampling sequence that contains signals from multiple radiation sources and retains the sub-band correspondence.

3. The high-frequency radar network detection method based on blind extraction of signals from non-cooperative transmitters according to claim 1, characterized in that, S2 includes: The first sampled sequence is divided into frames and windowed frequency domain transform is performed to obtain the time-frequency power matrix; The spectral threshold is determined based on the spectral power and background statistics of each frame, and continuous over-threshold frequency units are extracted to form candidate occupied frequency bands. The candidate frequency band trajectory is formed by associating the frequency overlap rate and center frequency deviation of candidate occupied frequency bands in adjacent frames. Band-limited extraction is performed on the data corresponding to each candidate frequency band trajectory to obtain sub-band sequences, and pulse arrival times are detected based on the adaptive threshold of their envelope sequences. A pulse interval sequence is formed based on the arrival times of adjacent pulses. A time sequence consistency feature value is constructed based on the clustering characteristics of the pulse interval values ​​and the repetition of the interval arrangement pattern. Target frequency band trajectories that meet the consistency threshold are selected, and the center frequency, bandwidth, and pulse arrival time sequence are determined accordingly.

4. The high-frequency radar network detection method based on blind extraction of signals from non-cooperative transmitters according to claim 1, characterized in that, S3 includes: The pulse arrival time series is subtracted to obtain the pulse interval series, and a pulse time series model is established based on the distribution characteristics and arrangement relationship of the pulse intervals; The arrival time window of candidate pulses is predicted using the pulse timing model, and the actual boundary of each candidate pulse is determined based on the envelope adaptive threshold. Candidate reference sequences are truncated according to the actual boundary, the normalized cross-correlation between candidate reference sequences is calculated, reference sequences are filtered according to the correlation threshold, and time-domain alignment and phase compensation are performed according to the cross-correlation peak position and cross-correlation peak complex phase, respectively. Coherent synthesis is performed on the time- and phase-aligned reference sequences to obtain the reference signal sequence; The reference signal sequence is orthogonally downconverted based on the center frequency, and the decimation parameters are determined based on the bandwidth. The reference IQ sequence is then obtained through decimation filtering.

5. The high-frequency radar network detection method based on blind extraction of signals from non-cooperative transmitters according to claim 1, characterized in that, S4 includes: Extract the distribution and arrangement features of center frequency, bandwidth and pulse interval within a continuous sliding time window, and construct a parameter state feature vector; Calculate the degree of state change between the parameter state feature vectors corresponding to adjacent time windows, and determine the position of parameter state change based on the preset change threshold; The parameter stability intervals are divided according to the positions of adjacent parameter state changes, and a set of state parameters including center frequency, bandwidth, and pulse interval distribution and arrangement parameters are determined for each parameter stability interval; The reference signal is re-triggered and processed based on the state parameter set corresponding to the stable interval of each parameter, an updated reference IQ sequence is generated, and the corresponding baseband sampling rate is recorded. The updated reference IQ sequence is associated with the corresponding state parameter set and parameter stability interval, and a version identifier is assigned to form a set of reference IQ sequences arranged in the order of parameter state changes.

6. The high-frequency radar network detection method based on blind extraction of signals from non-cooperative transmitters according to claim 1, characterized in that, S5 includes: A set of reference IQ sequences with version identifiers and associated center frequency, bandwidth and baseband sampling rate is sent to the receiving station, and a second sampling sequence containing direct components and target echoes is acquired. Based on the center frequency, bandwidth and baseband sampling rate of each version, the second sampling sequence is subjected to quadrature downconversion and decimation filtering to obtain the second baseband IQ sequence corresponding to the reference IQ sequence of each version. Matched filter coefficients were constructed using the conjugate inversion sequences of the reference IQ sequences for each version, and matched filtering was performed on the corresponding second baseband IQ sequences to obtain the matched response sequences. Extract the direct path matching peaks of continuous pulses from each matching response sequence, and generate version consistency evaluation values ​​based on the peak amplitude to background response ratio and cross-pulse position dispersion. The reference IQ sequence corresponding to the version with the highest version consistency evaluation value that meets the preset threshold is determined as the current valid reference IQ sequence, and the corresponding second baseband IQ sequence is determined as the current received baseband IQ sequence.

7. The high-frequency radar network detection method based on blind extraction of signals from non-cooperative transmitters according to claim 1, characterized in that, S6 includes: Based on the current valid reference IQ sequence, the matched filter coefficients are constructed, and the continuous pulses in the current received baseband IQ sequence are matched and filtered to obtain multiple matched response sequences. The actual slow time position corresponding to each matched response sequence is determined based on the pulse arrival time sequence, and the delay-non-uniform slow time response data is organized according to the propagation delay dimension and the actual slow time position. For each propagation delay unit, a non-uniform Doppler transformation is performed on the delay-non-uniform slow-time response data based on the corresponding actual slow-time position to obtain the range-Doppler response matrix; The target detection threshold is determined based on the background response statistics of the distance-Doppler response matrix, and target response units that exceed the target detection threshold are filtered out. Based on the delay difference of the target response unit relative to the matching peak of the direct path and its Doppler position, the target propagation distance difference information and target Doppler information are determined, and the target detection results are obtained.

8. A system applicable to the high-frequency radar network detection method based on blind extraction of signals from non-cooperative transmitters as described in any one of claims 1-7, characterized in that, include: The front-end acquisition device is deployed in the vicinity of the non-cooperative emission source to acquire a first sampling sequence containing signals from multiple radiation sources. The first sampling sequence is processed in a frame-by-frame frequency domain. Candidate frequency band trajectories are formed based on the overlap relationship of the frequency bands occupied by adjacent frames. The target frequency band trajectory is selected based on the pulse interval structure of the sub-band sequence corresponding to each candidate frequency band to obtain the center frequency, bandwidth and pulse arrival time sequence. The front-end acquisition device is also used to establish a pulse timing model based on the pulse arrival time sequence, extract candidate reference sequences according to the detected actual pulse boundaries, perform cross-correlation screening, time and phase alignment and coherent synthesis on the candidate reference sequences, and perform down-conversion and decimation filtering to obtain a reference IQ sequence. The reference IQ sequence is updated according to the changes in signal parameters within a continuous time window, forming a set of reference IQ sequences with version identifiers and corresponding signal parameters; A communication link for transmitting the reference IQ sequence set to the receiving station; The receiving station is used to acquire a second sampling sequence containing the direct component and the target echo. It performs baseband conversion on the second sampling sequence according to the corresponding signal parameters of each version, and performs matched filtering using the reference IQ sequence of each version. It determines the current valid reference IQ sequence and the corresponding received baseband IQ sequence based on the amplitude of the matching peak of the direct path and the consistency of the cross-pulse position. The receiving station is also used to perform matched filtering on the received baseband IQ sequence using the currently valid reference IQ sequence, perform non-uniform slow-time Doppler transformation according to the pulse arrival time sequence, and obtain the target detection result based on the delay difference of the target response relative to the matching peak of the direct path and the Doppler position.

9. An apparatus applicable to the high-frequency radar network detection method based on blind extraction of signals from non-cooperative transmitters as described in any one of claims 1-7, characterized in that, include: The sampling module is used to acquire the first sampling sequence of signals from multiple radiation sources in the vicinity of a non-cooperative emission source; The frequency band filtering module is used to perform frequency domain transformation on the first sampling sequence in frames, form candidate frequency band trajectories according to the overlap relationship of the frequency bands occupied by adjacent frames, and filter the target frequency band trajectory according to the pulse interval structure of the sub-band sequence corresponding to the candidate frequency band to obtain the center frequency, bandwidth and pulse arrival time sequence. The reference generation module is used to establish a pulse timing model based on the pulse arrival time sequence and detect the actual pulse boundary. It then extracts a candidate reference sequence according to the actual pulse boundary and obtains a reference IQ sequence after cross-correlation screening, time and phase alignment, coherent synthesis, down-conversion and decimation filtering. The version update module is used to determine the position of parameter state change based on the changes in center frequency, bandwidth and pulse interval characteristics within a continuous time window, and update the reference IQ sequence to obtain a set of reference IQ sequences with version identifiers and corresponding signal parameters.

10. The apparatus for the high-frequency radar network detection method based on blind extraction of signals from non-cooperative transmitters according to claim 9, characterized in that, Also includes: The version confirmation module is used to perform baseband conversion on the second sampling sequence collected by the receiving station according to the signal parameters corresponding to each version, perform matched filtering using the reference IQ sequence of each version, and determine the current valid reference IQ sequence and the corresponding receiving baseband IQ sequence based on the amplitude of the matching peak of the direct path and the consistency of the cross-pulse position. The target detection module is used to perform matched filtering on the received baseband IQ sequence using the currently valid reference IQ sequence, perform non-uniform slow-time Doppler transformation according to the pulse arrival time sequence, and obtain the target detection result based on the delay difference of the target response relative to the matching peak of the direct path and the Doppler position.

Images