A Radio Reconnaissance Method and System Based on Multi-Channel Parallel Processing and Intelligent Clustering

Abstract

This invention belongs to the field of electronic search technology and relates to a radio reconnaissance method and system based on multi-channel parallel processing and intelligent clustering. The method includes: performing polyphase filtering and digital channelization processing on broadband radio signals to obtain adjacent and overlapping channel sequences; performing square-law detection and constant false alarm rate detection on each channel sequence to generate initial pulse descriptors; establishing candidate correspondences within adjacent channels based on similarity of arrival time and noise consistency, and determining cross-channel merging relationships based on frequency range continuity and frequency-time relationship slope consistency to generate complete pulse descriptors; performing hierarchical clustering and sorting based on angle of arrival and pulse width; and performing multi-level differential analysis on the clustering results to extract repetition frequency sequences and determine repetition frequency types. The technical solution of this application can maintain the stability and consistency of cross-channel pulse merging and sorting results under broadband and high-density conditions, and improve the accuracy of radio search processing in complex electromagnetic environments.

Description

Technical Field

[0001] This invention belongs to the field of electronic search technology, and specifically relates to a radio reconnaissance method and system based on multi-channel parallel processing and intelligent clustering. Background Technology

[0002] In existing technologies, radio search tasks typically rely on broadband digital reception and signal processing links. This involves channelizing intercepted signals, pulse detection, and parameter extraction, followed by pulse sorting and repetition rate analysis to meet the needs of detecting, separating, and identifying radiation sources in complex electromagnetic environments. However, existing radio search processing methods have some significant shortcomings in cross-subband consistency processing and sorting process organization in broadband high-density pulse scenarios.

[0003] In practical applications, existing processing links often have a certain degree of parameter measurement, sorting, and identification capabilities. However, when the pulse energy crosses multiple adjacent sub-bands or frequency drift occurs near the sub-band boundary, the detection results output from different sub-bands are prone to exhibiting weak consistency in parameter characterization and weak stability of the merging discrimination criteria. This, in turn, affects the sequence purity and statistical reliability of the subsequent sorted objects. At the same time, under conditions of increased pulse density, diverse radiation source types, and frequent changes in repetition frequency morphology, the clustering and repetition frequency sequence extraction in the sorting process are prone to exhibiting large fluctuations in the discrimination boundary and weak consistency in the processing results. This adversely affects the stability and comparability of repetition frequency type discrimination.

[0004] Therefore, existing technologies often suffer from problems such as weak stability of the discrimination criteria when performing consistency merging of cross-subband pulses in broadband radio search, and weak consistency of sorting and repetition rate type discrimination results under high-density, multi-type radiation source conditions. These are the shortcomings of existing technologies.

[0005] In view of this, it is very necessary to provide a radio reconnaissance method and system based on multi-channel parallel processing and intelligent clustering to solve the above-mentioned defects in the prior art. Summary of the Invention

[0006] The purpose of this invention is to address the shortcomings of the prior art, namely, the weak stability of the discrimination criteria when performing consistency merging of cross-subband pulses in broadband radio search, and the weak consistency of the sorting and repetition frequency type discrimination results under high-density multi-type radiation source conditions. This invention provides a radio reconnaissance method and system based on multi-channel parallel processing and intelligent clustering to solve the above-mentioned technical problems.

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] In a first aspect, embodiments of this application provide a radio reconnaissance method based on multi-channel parallel processing and intelligent clustering, comprising:

[0009] Polyphase filtering and digital channelization processing are performed on broadband radio signals to obtain adjacent and overlapping channel sequences.

[0010] Square-law detection and constant false alarm rate (CFAR) detection are performed on each channel sequence to generate an initial pulse descriptor. CFAR detection includes generating threshold statistical parameters and associating them with the initial pulse descriptor. The initial pulse descriptor includes arrival time, pulse width, amplitude, and frequency range.

[0011] Based on the time-of-arrival proximity constraint and the noise consistency constraint corresponding to the threshold statistical parameters, candidate correspondences of initial pulse descriptors are established in adjacent channels, and cross-channel merging relationships are determined within the candidate correspondences based on the continuity of frequency range and the consistency of frequency-time relationship slope.

[0012] Initial pulse descriptors that satisfy the cross-channel merging relationship are merged, and the frequency range, pulse width, and frequency-time relationship slope are recalculated to generate complete pulse descriptors;

[0013] A two-dimensional histogram of arrival angle is constructed based on the angle of arrival of the complete pulse descriptor and spatial clustering is performed to group the complete pulse descriptors into the corresponding spatial clusters. Within the spatial clusters, a pulse width histogram is constructed based on the pulse width and subdivision clustering is performed to obtain the pulse width clusters.

[0014] A multi-level difference histogram is constructed for the pulse width cluster, and threshold determination, harmonic detection, sequence search and iterative elimination are performed to output the repetition frequency sequence. The repetition frequency type is determined based on the repetition frequency sequence, which includes fixed repetition frequency, staggered repetition frequency and jitter repetition frequency.

[0015] By adopting the above technical solution, through multi-channel parallel processing and combining clustering and repetition frequency analysis, stable merging and orderly sorting of pulses in broadband radio signals can be achieved. This can enhance the consistency of pulse parameter characterization and sequence extraction, and meet the requirements of stable cross-channel pulse consistency merging discrimination criteria in broadband radio search, as well as stable consistency of sorting and repetition frequency type discrimination results under high-density multi-type conditions.

[0016] The implementation involves first performing polyphase filtering and digital channelization on the broadband signal to form multiple channel sequences that are adjacent and have overlapping frequency bands, supporting subsequent detection processing in parallel. Then, detection is performed on each channel sequence to generate initial pulse descriptors containing arrival time, pulse width, amplitude, and frequency range. Simultaneously, statistical information related to the detection process is introduced as a consistency reference. Based on this, candidate correspondences are established within adjacent channels by combining time proximity and statistical consistency. Cross-channel merging is determined based on frequency range continuity and the consistency of the frequency-time relationship slope, ensuring stable and consistent merging results. Subsequently, descriptors meeting the merging conditions are fused, and key parameters are recalculated to form complete descriptors. Further statistical aggregation is performed based on the angle of arrival, and the aggregation results are subdivided and clustered by pulse width, resulting in a clear structure for the sorted objects. Finally, differential statistics and sequence analysis are used to obtain the repetition frequency sequence and complete type discrimination, ensuring consistency of the overall processing results in complex environments.

[0017] Preferably, the step of performing polyphase filtering and digital channelization processing on broadband radio signals to obtain each channel sequence includes: constructing a polyphase filter bank based on the prototype filter and performing parallel filtering and extraction on the broadband radio signals to obtain each channel sequence, and associating each channel sequence with a channel frequency identifier, which is used to determine the continuity of the frequency range.

[0018] By adopting the above technical solution, parallel filtering and extraction of multiphase filter banks are used to associate channel frequency identifiers with each channel sequence, so as to achieve unified calibration of channel frequency position. This can enhance the alignment and consistency of cross-channel frequency range continuity determination and reduce the impact of frequency characterization fluctuations at the boundary frequency band on the merging determination.

[0019] As a preferred embodiment, the step of establishing candidate correspondences of initial pulse descriptors in adjacent channels based on arrival time proximity constraints and noise consistency constraints corresponding to threshold statistical parameters includes: constructing a sliding time window in the set of initial pulse descriptors in adjacent channels, establishing candidate correspondences for initial pulse descriptors that satisfy arrival time proximity constraints within the sliding time window, and filtering based on noise consistency constraints corresponding to threshold statistical parameters, wherein the noise consistency constraints include the threshold statistical parameters satisfying a preset tolerance condition.

[0020] By adopting the above technical solution, a sliding time window is used to establish candidate correspondences in adjacent channels and combined with the tolerance screening of threshold statistical parameters, the time domain constraints and noise statistical constraints of candidate matching are coordinated, which can improve the reliability and stability of candidate correspondences and reduce the interference of mismatched pairings and frequency repetition analysis consistency caused by random noise fluctuations.

[0021] Preferably, the step of determining the cross-channel merging relationship within the candidate correspondence based on the continuity of the frequency range and the consistency of the frequency-time relationship slope includes:

[0022] Within the candidate correspondence, calculate the frequency range endpoint difference between each initial pulse descriptor word, calculate the frequency range overlap, and determine the frequency range continuity.

[0023] Within the candidate correspondence, a frequency-time relationship is constructed based on the arrival time and midpoint frequency of each initial pulse descriptor word. The frequency-time relationship is linearly fitted and the slope residual of the frequency-time relationship is calculated. The consistency of the slope of the frequency-time relationship is determined based on the slope residual of the frequency-time relationship.

[0024] When the frequency range continuity determination result meets the continuity determination condition, and the frequency-time relationship slope consistency determination result meets the consistency determination condition, the cross-channel merging relationship is determined.

[0025] By adopting the above technical solution, the continuity of frequency range is determined by endpoint difference and overlap, and the slope consistency is determined by the slope residual obtained by linear fitting. This achieves a dual consistency check of cross-channel merging relationship, which can enhance the stability and interpretability of cross-channel merging discrimination criteria and improve the accuracy of pulse merging under frequency drift and boundary segmentation conditions.

[0026] Preferably, when there are multiple candidate correspondences that can be used to determine the cross-channel merging relationship, a candidate correspondence is selected from the multiple candidate correspondences as the cross-channel merging relationship according to the priority rule of the frequency-time relationship slope residual. Consistency verification is performed on the candidate correspondences that are not selected, and the candidate correspondences are updated.

[0027] By adopting the above technical solution, the slope residual priority rule is used to select the best among multiple candidate correspondences and to perform consistency verification and update on the unselected correspondences. This achieves stable decision-making and dynamic correction in candidate conflict scenarios, which can suppress the risk of erroneous chain merging under multi-pulse dense conditions and improve the overall consistency and comparability of cross-channel merging results.

[0028] Preferably, the steps of recalculating the frequency range, pulse width, and frequency-time relationship slope include:

[0029] Perform a union operation on the frequency ranges of each initial pulse descriptor that satisfies the cross-channel combining relationship to obtain the recalculated frequency range;

[0030] For each initial pulse descriptor that satisfies the cross-channel combining relationship, an interval operation is performed on the arrival time to obtain the recalculated pulse width;

[0031] Within the recalculated frequency range, a linear fit is performed based on the arrival time of each initial pulse descriptor that satisfies the cross-channel combining relationship and the midpoint frequency of the frequency range to obtain the slope of the recalculated frequency-time relationship.

[0032] By adopting the above technical solution, the unified re-estimation of key parameters is achieved by using frequency range union operation, arrival time interval operation and linear fitting within the recalculation range. This can improve the overall consistency and continuity of the parameter representation after merging, and enhance the support stability of the complete pulse descriptor for subsequent clustering and sequence extraction.

[0033] Preferably, the steps of constructing a two-dimensional histogram of arrival angles based on the complete pulse descriptor and performing spatial clustering include:

[0034] Within the two-dimensional space of the angle of arrival, the azimuth direction bin width and the pitch direction bin width are set according to the angle of arrival measurement error range corresponding to the complete pulse description word, and the two-dimensional bin of the angle of arrival is constructed.

[0035] The complete pulse description words falling into each two-dimensional bin at the angle of arrival are counted to form a two-dimensional histogram of the angle of arrival.

[0036] Based on the adjacency relationship between the two-dimensional bins of the angle of arrival, spatial clustering is performed, and connected component merging is performed on the two-dimensional bins of the angle of arrival with non-zero counts to generate spatial clusters;

[0037] For two-dimensional bins with arrival angles located at the boundaries of spatial clusters, a re-determination is performed based on the angular distance between the arrival angles corresponding to the bin centers to adjust the affiliation of spatial clusters.

[0038] By adopting the above technical solution, combining the angle of arrival measurement error to set two-dimensional bins and performing connected component merging and boundary re-determination based on adjacency relationship, adaptive calibration of spatial cluster boundaries can be achieved, which can improve the stability and consistency of spatial cluster affiliation and reduce the impact of angle measurement error and boundary binning effect on the comparability of spatial clustering results.

[0039] Preferably, the steps of constructing a multi-level difference histogram for the pulse width cluster and performing threshold determination, harmonic detection, sequence search, and iterative elimination include:

[0040] When the differential histogram of the pulse width cluster exhibits multi-peak competition, the differential level is switched and a higher-order differential histogram is constructed. Threshold determination is performed based on the higher-order differential histogram to obtain candidate repetition frequencies. Here, multi-peak competition means that the ratio of the peak value of the main peak to the peak value of the secondary peak in the differential histogram meets the preset ratio threshold condition.

[0041] Harmonic detection based on integer multiple relationship consistency index is performed on candidate repetition frequencies, and the repetition frequencies are divided into a retention set and a rejection set;

[0042] For the retained set, a sequence search is performed within the pulse width cluster according to the arrival time interval corresponding to the candidate repetition frequency to obtain the pulse set that is consistent with the candidate repetition frequency;

[0043] Pulses matching the candidate repetition frequency are removed from the pulse width cluster and the difference histogram is updated. The process of difference histogram construction, threshold determination, harmonic detection, and sequence search is repeated until the repetition frequency separation within the pulse width cluster is completed.

[0044] By adopting the above technical solution, multi-level differential statistics with differential number switching and a consistency index based on integer multiple relationship are used for harmonic detection and iterative elimination, which realizes robust frequency repetition separation under multi-peak competition conditions. This can enhance the discrimination stability of frequency repetition candidates and the purity of sequence extraction, and improve the consistency of discrimination results for fixed frequency repetition, uneven frequency repetition and jittery frequency repetition.

[0045] Secondly, embodiments of this application also provide a radio reconnaissance system based on multi-channel parallel processing and intelligent clustering, comprising:

[0046] The multi-channelization module is used to perform polyphase filtering and digital channelization processing on broadband radio signals to obtain adjacent and overlapping channel sequences.

[0047] The detection generation module is used to perform square law detection and constant false alarm rate (CFAR) detection on each channel sequence and generate an initial pulse descriptor. CFAR detection includes generating threshold statistical parameters and associating them with the initial pulse descriptor. The initial pulse descriptor includes arrival time, pulse width, amplitude, and frequency range.

[0048] The candidate correspondence module is used to establish candidate correspondences of initial pulse descriptors in adjacent channels based on the arrival time proximity constraint and the noise consistency constraint corresponding to the threshold statistical parameters.

[0049] The merging determination module is used to determine the cross-channel merging relationship within the candidate correspondence based on the continuity of the frequency range and the consistency of the frequency-time relationship slope.

[0050] The parameter recalculation module is used to merge initial pulse descriptors that satisfy the cross-channel merging relationship, and recalculate the frequency range, pulse width and frequency-time relationship slope to generate a complete pulse descriptor;

[0051] The spatial clustering module is used to construct a two-dimensional histogram of arrival angles based on the complete pulse descriptor and perform spatial clustering to group the complete pulse descriptors into the corresponding spatial clusters;

[0052] The subdivision clustering module is used to construct pulse width histograms based on pulse width within spatial clusters and perform subdivision clustering to obtain pulse width clusters;

[0053] The frequency repetition separation module is used to construct a multi-level differential histogram for pulse width clusters and perform threshold determination, harmonic detection, sequence search and iterative elimination, output the frequency repetition sequence, and determine the frequency repetition type based on the frequency repetition sequence. The frequency repetition types include fixed frequency repetition, staggered frequency repetition and jittery frequency repetition.

[0054] Preferably, the frequency repetition rate separation module includes:

[0055] The level switching submodule is used to switch the difference level and construct a higher-order difference histogram when the difference histogram of the pulse width cluster presents multi-peak competition. Multi-peak competition is defined as the ratio of the peak value of the main peak to the peak value of the secondary peak in the difference histogram satisfying a preset ratio threshold condition.

[0056] The threshold determination submodule is used to perform threshold determination based on the high-order difference histogram to obtain candidate repetition frequencies;

[0057] The harmonic detection submodule is used to perform harmonic detection on candidate repetition frequencies based on the consistency index of integer multiple relationship, and divide them into a retention set and a rejection set;

[0058] The sequence search submodule is used to perform a sequence search within the pulse width cluster for the retained set according to the arrival time interval corresponding to the candidate repetition frequency, so as to obtain a pulse set that is consistent with the candidate repetition frequency.

[0059] The iterative elimination submodule is used to remove pulses that match the candidate repetition frequency from the pulse width cluster and update the difference histogram. It then repeats the difference histogram construction, threshold determination, harmonic detection, and sequence search until the repetition frequency separation within the pulse width cluster is completed.

[0060] As can be seen from the above technical solutions, the present invention has the following advantages:

[0061] This application provides a radio reconnaissance method and system based on multi-channel parallel processing and intelligent clustering. By combining multi-channel parallel processing with clustering and repetition frequency analysis, it achieves stable merging and orderly sorting of pulses in broadband radio signals. This enhances the consistency of pulse parameter characterization and sequence extraction, and meets the requirements of stable cross-channel pulse consistency merging discrimination criteria in broadband radio search, as well as stable consistency of sorting and repetition frequency type discrimination results under high-density multi-type conditions. Attached Figure Description

[0062] To more clearly illustrate the technical solution of the present invention, the accompanying drawings used in the description will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0063] Figure 1This is a flowchart of a radio reconnaissance method based on multi-channel parallel processing and intelligent clustering provided by the present invention;

[0064] Figure 2 This is a block diagram of the principle of a radio reconnaissance system based on multi-channel parallel processing and intelligent clustering provided by the present invention.

[0065] The modules include: 1. Multi-factor authentication module, 2. Detection and generation module, 3. Candidate correspondence module, 4. Merging and judgment module, 5. Parameter recalculation module, 6. Spatial clustering module, 7. Subdivision clustering module, and 8. Repetition frequency separation module. Detailed Implementation

[0066] Various embodiments of this disclosure are described more fully below with reference to the accompanying drawings. This disclosure may have various embodiments, and adjustments and changes may be made therein. However, it should be understood that there is no intention to limit the various embodiments of this disclosure to the specific embodiments disclosed herein, but rather this disclosure should be understood to cover all adjustments, equivalents, and / or alternatives falling within the spirit and scope of the various embodiments of this disclosure.

[0067] In the following, the terms “comprising” or “may include”, which may be used in various embodiments of this disclosure, indicate the presence of the disclosed functions, operations, or elements, and do not limit the addition of one or more functions, operations, or elements. Furthermore, as used in various embodiments of this disclosure, the terms “comprising,” “having,” and their cognates are intended only to indicate a particular feature, number, step, operation, element, component, or combination of the foregoing, and should not be construed as primarily excluding the presence of one or more other features, numbers, steps, operations, elements, components, or combinations of the foregoing, or the possibility of adding one or more combinations of the foregoing.

[0068] It should be noted that, in various embodiments of this disclosure, the expression "or" or "at least one of A and / or B" includes any combination or all combinations of the words listed simultaneously. For example, the expression "A or B" or "at least one of A and / or B" may include A, may include B, or may include both A and B.

[0069] It should be noted in advance that, in order to facilitate a clear and accurate description of the technical solutions in the embodiments of this application, the following is a brief explanation of some terms and related technologies involved in the embodiments of this application:

[0070] For broadband radio search applications characterized by wide pulse signal spectrum coverage, high pulse density, and diverse radiation source types, existing processing methods generally exhibit weak performance in terms of cross-channel pulse consistency merging and discrimination, as well as the stability of sorting and repetition frequency type discrimination results under high-density conditions. This makes it difficult to meet the practical requirements for stable separation and reliable feature discrimination of radiation sources in complex electromagnetic environments. This application proposes a radio reconnaissance method and system based on multi-channel parallel processing and intelligent clustering. By performing multi-channel parallel processing on broadband signals and combining cross-channel merging, hierarchical clustering of spatial and parameter dimensions, and repetition frequency analysis, the pulses possess a more stable consistency foundation during cross-channel merging and sorting. This enables ordered sorting and repetition frequency type discrimination of pulse signals, improving the stability and consistency of signal processing results in broadband radio search scenarios and meeting the application requirements for radiation source search and analysis in complex environments.

[0071] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0072] like Figure 1 As shown in the figure, this embodiment provides a radio reconnaissance method based on multi-channel parallel processing and intelligent clustering, including:

[0073] Step S1: Perform polyphase filtering and digital channelization processing on the broadband radio signal to obtain adjacent and overlapping channel sequences;

[0074] Step S2: Perform square law detection and constant false alarm rate (CFAR) detection on each channel sequence to generate an initial pulse descriptor. CFAR detection includes generating threshold statistical parameters and associating them with the initial pulse descriptor. The initial pulse descriptor includes arrival time, pulse width, amplitude, and frequency range.

[0075] Step S3: Based on the arrival time proximity constraint and the noise consistency constraint corresponding to the threshold statistical parameters, establish candidate correspondences of initial pulse descriptors in adjacent channels, and determine cross-channel merging relationships within the candidate correspondences based on the frequency range continuity and the frequency-time relationship slope consistency.

[0076] Step S4: Merge the initial pulse descriptors that satisfy the cross-channel merging relationship, and recalculate the frequency range, pulse width, and frequency-time relationship slope to generate a complete pulse descriptor;

[0077] Step S5: Construct a two-dimensional histogram of arrival angle based on the angle of arrival of the complete pulse descriptor and perform spatial clustering to group the complete pulse descriptors into the corresponding spatial clusters. Within the spatial clusters, construct a pulse width histogram based on the pulse width and perform subdivision clustering to obtain pulse width clusters.

[0078] Step S6: Construct a multi-level differential histogram for the pulse width cluster and perform threshold determination, harmonic detection, sequence search and iterative elimination to output the repetition frequency sequence. Based on the repetition frequency sequence, determine the repetition frequency type, which includes fixed repetition frequency, staggered repetition frequency and jittery repetition frequency.

[0079] This embodiment employs multiphase filtering digital channelization and multi-channel parallel processing to effectively carry broadband radio signals in the frequency domain and provide a parallel foundation for subsequent detection and analysis. By introducing threshold statistics obtained from constant false alarm rate (CFAR) detection into the channel detection stage and associating them with pulse descriptors, and combining arrival time similarity and noise consistency constraints during cross-channel merging, the correspondence between cross-channel pulses has a more stable basis for discrimination. By further utilizing frequency range continuity and frequency-time relationship slope consistency to determine cross-channel merging relationships and recalculating key parameters after merging, the resulting pulse description results maintain consistency in parameter representation. Through spatial statistical clustering based on arrival angle combined with pulse width dimension subdivision clustering, the pulse sorting results under high-density conditions present a clear hierarchical structure. By constructing multi-level differential statistics on the clustering results and performing sequence analysis, the repetition frequency sequence is stably output and different repetition frequency types are distinguished, thereby improving the overall stability and consistency of pulse merging, sorting, and repetition frequency analysis results in broadband radio search scenarios, meeting the application requirements for reliable search and analysis of radiation sources in complex electromagnetic environments.

[0080] Hereinafter, steps S1 to S6 will be specifically described according to embodiments of this application.

[0081] In step S1, the core task is to convert the received broadband radio signal into a multi-subband sequence that can be processed in parallel, while maintaining real-time processing constraints and frequency domain coverage integrity, and to establish a frequency domain correlation between adjacent subbands that can be used for subsequent cross-channel consistency judgment. The input of this step is the broadband radio signal and its sampling clock information, and the output is a set of adjacent channel sequences with overlapping frequency bands and a channel frequency identifier that is consistent with each channel sequence, so as to support the subsequent quantitative determination of frequency range continuity.

[0082] In this embodiment, the broadband radio signal is down-converted and analog-to-digital converted by the radio frequency front-end to form a discrete sampling sequence, which can be denoted as: ,in This is the index of the sampling points, and the sampling rate is... To map this sequence into multiple sub-band channels, a polyphase filter bank structure based on a prototype filter can be used to implement polyphase filtering digital channelization processing of broadband radio signals, including parallel filtering and decimation, thereby obtaining each channel sequence. The impulse response of the prototype filter is denoted as... The filter length is .

[0083] Specifically, the prototype filter is arranged according to the number of phases. Perform multiphase decomposition, making the multiphase components for:

[0084]

[0085] in, Indicates the phase index, This represents the index after extraction of the multiphase components. Through the above decomposition, the filtering operation can be performed... The phase components are expanded in parallel and naturally coupled with the decimation operation to reduce the computational burden on a single path and stabilize latency.

[0086] In some embodiments of this application, to form frequency band overlap between adjacent sub-bands, a frequency band smaller than 1000 MHz can be used on the channelization output side. The extraction factor is used to implement oversampling subband partitioning. For example, the extraction factor can be... This allows the equivalent bandwidth coverage of adjacent channels to overlap, achieving blind-zone-free coverage and continuous cross-channel energy expression. For the first... Channels ( The channelized output can be constructed based on a combination of complex exponential shift and low-pass filtering to obtain the channel sequence. The calculation relationship can be written as:

[0087]

[0088] in, This is the extracted time index. For the first The digital center angle frequency corresponding to each channel Let be the imaginary unit. To ensure that the center frequencies of the channel are evenly distributed, we can let . ,in, The number of channels is the same as the number of phases. Through the above shifting, filtering and decimation process, adjacent and overlapping channel sequences are obtained, so that the energy and parameter estimation results of the same broadband signal can be continuously expressed in the form of multi-channel sequences when it crosses multiple adjacent sub-bands.

[0089] In some embodiments of this application, constructing a polyphase filter bank based on a prototype filter and performing parallel filtering and decimation on broadband radio signals can be further implemented as a hardware-oriented parallel pipeline approach: simultaneously processing the input sequence by Phase interleaving method input Multiphase components of the road The corresponding multiply-accumulate channels, each channel output at the decimation point Samples were taken and aggregated at the location. Channel output In engineering implementation, this structure can distribute multiply-accumulate operations across multiple parallel resources and embed decimation operations into the output sampling cycle, thereby reducing the output data rate without sacrificing frequency domain coverage.

[0090] Furthermore, at the output end, each channel sequence is... A frequency-related identifier is attached, namely the channel frequency identifier. For example, the channel frequency identifier can be taken as the channel center frequency. It satisfies:

[0091]

[0092] in, For reference starting frequency, The center frequency interval between adjacent channels is given by the sampling rate and the number of channels. Under this definition, each channel sequence can be recorded at the data structure level. This enables the association of channel frequency identifiers with each channel sequence, allowing subsequent processing stages to directly reference frequency domain location information without altering the sequence itself.

[0093] It should be further explained that when channel frequency identifiers are used to determine frequency range continuity, the frequency coverage range corresponding to each channel sequence can be considered. Represented as a symmetrical interval around the center frequency:

[0094]

[0095] in, For the first The equivalent passband width of each channel. Because the decimation factor is... Oversampling subbands are formed, and the intervals of adjacent channels are... and The existence of overlap provides a clear interval basis for determining the continuity of the frequency range. For example, when... , At that time, the center frequency interval between adjacent channels is The equivalent coverage of a single channel is acceptable. This results in continuous coverage in the frequency domain while retaining moderate overlap to accommodate the actual filter transition band and measurement errors.

[0096] Thus, step S1 completes the structured transformation from a broadband discrete sampling sequence to a polyphase filtered channelized output sequence, forming a multi-channel output system that uses polyphase decomposition to support parallel filtering, decimation factors to control the data rate and frequency band overlap, and channel frequency identifiers to bind frequency domain positions. This enables subsequent processes to perform consistency determination based on frequency range continuity between adjacent channels and provides a stable and traceable channel sequence input basis for cross-channel joint processing.

[0097] In step S2, the core task is to convert the channel sequences output in step S1 from continuous sampling sequences into pulse-level data representations that can be directly entered into subsequent cross-channel correspondence, merging, and clustering processes, and to maintain the adaptive detection capability for changes in noise background under complex electromagnetic environments. The input of this step is each channel sequence and its corresponding channel frequency identifier, and the output is a set of initial pulse descriptors organized by channel. Each initial pulse descriptor carries threshold statistical parameters that are consistent with the detection threshold generation process, so as to carry out noise consistency constraint screening in the subsequent cross-channel candidate correspondence stage.

[0098] In this embodiment, square-law detection and constant false alarm rate (CFAR) detection can be performed on each channel sequence. The complex sampling points of each channel sequence are converted into a power sequence to reduce the impact of phase perturbation on the detection criterion and to provide a uniform input for threshold calculation. For example, the detection of the power sequence can be performed using a sliding window method. The sliding window consists of a protection zone and a reference zone. The protection zone is used to prevent pulse energy leakage into noise statistics, and the reference zone is used to characterize the statistical level of local noise background at that moment. At each sliding window position, threshold statistical parameters are first formed based on the reference zone samples, and then the detection threshold is obtained from the threshold statistical parameters. The current unit to be detected is compared with the detection threshold to output whether it is a pulse candidate point.

[0099] Specifically, after digital down-conversion and channelization, each channel sequence typically manifests as a complex sampled sequence, with each sampling point consisting of in-phase and quadrature components. Square-law detection is used to convert this complex sampled sequence into a power domain sequence, thereby reducing the impact of phase perturbations, carrier frequency residuals, and instantaneous phase jumps on the detection criteria, and ensuring that subsequent threshold estimation and pulse envelope extraction have unified dimensions. For example, square-law detection can perform amplitude squaring on each sampling point to obtain the instantaneous power value, which can be achieved by:

[0100]

[0101] in, Indicates in-phase components, Indicates orthogonal components, This represents the instantaneous power value at the corresponding sampling point. The power sequence output by the square-law detection can be used as the input for constant false alarm rate (CFAR) detection, allowing threshold statistics and threshold comparisons to be completed in the power domain, thereby avoiding detection instability caused by phase changes.

[0102] It should be noted that, to suppress high-frequency fluctuations in the power sequence caused by thermal noise and stabilize pulse edge determination, short-time smoothing or energy integration processing can be performed on the power sequence after square-law detection. The smoothing window length is related to the channel sampling rate, the lower bound of the target pulse width, and the detection response speed constraint. If the window is too short, noise fluctuations will increase the number of false alarms; if the window is too long, the rising and falling edges of the pulse will be widened, affecting the accuracy of arrival time and pulse width measurement. After square-law detection and smoothing, the peak value and duration interval of the power sequence can more stably reflect the energy structure of the pulse event, making the pulse candidate points appear as segments that continuously exceed the threshold on the time axis. This provides a direct basis for subsequently aggregating candidate points into pulse events and extracting arrival time, pulse width, and amplitude.

[0103] For example, when a power sequence continuously exceeds a corresponding threshold within a certain time interval, the sampling position of the first exceedance of the threshold can be used as the location basis for arrival time, the duration of continuous exceedance of the threshold can be used as the location basis for pulse width, and the power peak value or root mean square value within the interval can be used as the location basis for amplitude; the frequency range is then bound to the frequency domain coverage of the channel to which the pulse event belongs. This forms an executable link of "re-sampling point - power sequence - stable envelope - event location," maintaining a data caliber consistent with subsequent threshold statistical parameter generation and pulse descriptor field filling.

[0104] Furthermore, in this step, the threshold statistical parameter is retained only as a directly recordable and reusable numerical value, its engineering meaning being "the statistically representative value of the reference area power". For example, the threshold statistical parameter can be the mean power of the reference area, or, in the presence of strong interference spikes, the median power of the reference area. When the mean power of the reference area is used as the threshold statistical parameter, the detection threshold can be obtained by multiplying this mean by a scaling factor, where the scaling factor... Based on the preset false alarm probability Given the length of the reference area, the following can be used:

[0105]

[0106] in, This represents the number of samples in the reference area. The detection threshold obtained based on this proportionality coefficient can adaptively change with the statistics of the reference area, keeping the false alarm level stable under different noise backgrounds.

[0107] In this embodiment, the "pulse candidate points" output by constant false alarm rate (CFAR) detection need to be further organized into "pulse events" to generate pulse-level parameters. Specifically, connectivity aggregation can be performed on candidate points that continuously exceed the detection threshold to obtain the start and end points of each pulse event. To suppress fragmentation caused by noise spikes, minimum duration constraints and short-discontinuity bridging rules can be further introduced into the connectivity aggregation, so that a single pulse can still be merged into the same pulse event even when there is local fading or sampling jitter. Subsequently, parameters such as arrival time, pulse width, and amplitude are extracted from each pulse event, and the frequency range of the event is determined together with the channel frequency identifier, thereby completing the generation of the pulse descriptor.

[0108] In some embodiments of this application, the following parameter specifications can be used when generating the initial pulse descriptor: arrival time is taken as the sampling timestamp corresponding to the start of the pulse event; pulse width is taken as the time difference between the end and start of the pulse event; amplitude is taken as the peak or root mean square value of the power sequence within the pulse event; and frequency range is taken as the coverage frequency band of the channel to which the pulse event belongs. When multiple non-contiguous pulse events occur in the same channel within the same time window, corresponding initial pulse descriptors are generated for each event and enqueued in chronological order. For example, in a broadband high-density pulse scenario, the reference area length can be set to tens to hundreds of sampling points to balance threshold stability and response speed, and the preset false alarm probability can be set to... arrive The magnitude is adjusted to suit different alarm sensitivity requirements, and the above parameters are adjusted according to the task mode configuration.

[0109] Furthermore, threshold statistics can be correlated to initial pulse descriptors. This correlation can be achieved by adding a corresponding threshold statistics field to each initial pulse descriptor, ensuring the field value matches the threshold statistics when the pulse event is first identified as a candidate point. When a pulse event spans multiple sliding window positions, the threshold statistics can be taken as the threshold statistics at the event's starting point, or a representative value of the threshold statistics within the event interval can be used to enhance robustness. This correlation allows subsequent processing to directly utilize threshold statistics to characterize local noise levels without repeating the threshold estimation process, and supports noise consistency screening between adjacent channels.

[0110] Furthermore, the arrival time, pulse width, amplitude, and frequency range in the initial pulse descriptor can be organized using a fixed-length field structure to ensure that arrival time similarity constraints, frequency range continuity judgments, and statistical consistency screening can be directly performed in subsequent cross-channel candidate correspondence and merging determinations. At the same time, the threshold statistical parameters, as additional fields of the same origin as the detection process, are output along with the initial pulse descriptor, so that the pulse-level descriptions remain comparable when the noise background changes.

[0111] Thus, step S2 completes the conversion from each channel sequence to the initial pulse descriptor set, forming a pulse detection and parameter generation system that uses square-law detection to achieve unified power domain expression, constant false alarm rate detection to achieve threshold adaptation, and threshold statistical parameter binding to achieve traceable noise level. This provides a stable pulse-level input basis and consistency constraint for subsequent cross-channel candidate correspondence and merging determination.

[0112] In step S3, the core task is to establish a reliable cross-channel correspondence link between the initial pulse descriptor sets of adjacent channels, and to further converge this link from "possible correspondence" to a "mergeable" deterministic relationship. The input of this step is the initial pulse descriptor set of adjacent channels and its threshold statistical parameters and frequency range information, and the output is the cross-channel merging relationship set, so that subsequent merging and parameter recalculation can be performed on multiple channel segments belonging to the same broadband pulse, while suppressing erroneous correspondence and erroneous merging caused by noise fluctuations, threshold fluctuations or accidental pulses arriving at the same time.

[0113] Specifically, the overall approach to candidate correspondence and merging determination can be divided into two levels of convergence: first, candidate correspondences are formed based on temporal and noise consistency; then, merging relationships are formed based on frequency domain continuity and frequency-time consistency. During this convergence process, constraints based on arrival time proximity and noise consistency corresponding to threshold statistical parameters can be used to converge cross-channel correspondences from full matching to a finite candidate set, thereby controlling computational load and reducing the probability of accidental matching in high-density pulse scenarios. Based on this, candidate correspondences of initial pulse descriptors are established in adjacent channels, ensuring that each initial pulse descriptor is associated only with objects that are temporally adjacent and have compatible noise backgrounds. Finally, within the candidate correspondences, cross-channel merging relationships are determined based on frequency range continuity and frequency-time relationship slope consistency, elevating cross-channel merging determination from "local congruence" to a "physically continuous" consistency judgment.

[0114] In some embodiments of this application, to ensure the executability and timing consistency of candidate correspondences, candidate correspondences can be established sequentially between adjacent channels. Specifically, a sliding time window is constructed in the initial pulse descriptor set of adjacent channels. The width of the sliding time window can be matched with the temporal resolution of the pulse detection output and the alignment error caused by cross-channel segmentation, so that multiple initial pulse descriptors corresponding to the same physical pulse in adjacent channels can fall into the same window simultaneously. The sliding time window can advance window by window along the direction of increasing arrival time, and at each window position, matching calculations are performed only on objects within the window, thereby avoiding the increase in complexity caused by global pairwise comparisons.

[0115] Based on this, within each sliding time window, candidate correspondences are established for initial pulse descriptors that satisfy the arrival time similarity constraint. The arrival time similarity constraint can be implemented as a determination that "the arrival time difference does not exceed the preset time tolerance". This time tolerance can be jointly determined by the channelization decimation beat, the jitter range triggered by the detection threshold, and the upper bound of the inter-channel group delay difference.

[0116] Furthermore, to ensure that candidate correspondences simultaneously reflect the consistency of the noise background, screening can be performed based on noise consistency constraints corresponding to threshold statistics. Within the candidate correspondence, the threshold statistics carried by the initial impulse descriptors on both sides are compared, and candidate correspondences generated under significantly different noise background conditions are eliminated. This reduces erroneous correspondences caused by noise abrupt changes, sudden interference, or threshold estimation shifts. For example, the noise consistency constraint includes the threshold statistics satisfying a preset tolerance condition. Specifically, the absolute or relative difference of the threshold statistics can satisfy a tolerance threshold, and the tolerance threshold can be adjusted according to the reference window length or the degree of environmental noise fluctuation to achieve a balance between robustness and recall.

[0117] After screening candidate correspondences, it is necessary to further determine the frequency domain continuity and frequency-time consistency within the candidate correspondences to converge "suspicious correspondences" into "mergeable relationships". Specifically, the determination of frequency range continuity revolves around endpoint difference and overlap: when calculating the frequency range endpoint difference between each initial pulse descriptor within the candidate correspondences, the frequency range of each initial pulse descriptor is represented as an interval formed by the lower endpoint and the upper endpoint, and the endpoint difference is calculated to describe the degree of connection between the boundaries of the two intervals; at the same time, the frequency range overlap is calculated to describe the overlap ratio of the two intervals in the frequency domain, thereby distinguishing different situations such as "adjacent connection", "partial overlap", and "obvious separation".

[0118] For example, endpoint difference and overlap can be expressed in the following form:

[0119]

[0120] in, and These represent the lower and upper endpoints of the frequency range of the initial pulse descriptors on both sides within the candidate correspondence, respectively. Indicates the difference between endpoints. The degree of overlap is indicated. Based on the above quantization results, the continuity of the frequency range can be determined. For example, the continuity determination condition is "the difference between endpoints does not exceed the preset frequency tolerance and the degree of overlap meets the preset range", where the frequency tolerance can be associated with the channel spacing, the filter transition band width and the upper limit of the frequency estimation error to ensure that the determination is consistent with the engineering error caliber.

[0121] It should be further explained that frequency range continuity only characterizes whether the frequency domain is connected, and is still insufficient to rule out the accidental situation of "connecting exactly but not being the same pulse". Therefore, frequency-time consistency needs to be introduced for secondary judgment. Specifically, within the candidate correspondence, a frequency-time relationship is constructed based on the arrival time of each initial pulse descriptor and the midpoint frequency of the frequency range. The arrival time of each initial pulse descriptor in the candidate correspondence and its midpoint frequency form a set of point pairs, and the frequency-time relationship is represented by this set of point pairs. The midpoint frequency is taken as the average of the upper and lower endpoints of the frequency range to reduce instability caused only by endpoint perturbations.

[0122] Furthermore, to perform a linear fit on the frequency-time relationship and calculate the slope residual, a linear fit can be performed on the aforementioned set of point pairs to obtain the fitted slope and calculate the degree of deviation of each point pair from the fitted line. This deviation is used as the slope residual to measure consistency. For example, the slope residual can be expressed in the root mean square form of the fitted residual as follows:

[0123]

[0124] in, Indicates the first candidate in the corresponding relationship The arrival time of the initial pulse description word, This indicates the corresponding midpoint frequency. and These represent the slope and intercept obtained from the linear fitting, respectively. Indicates the number of point pairs. This represents the slope residual.

[0125] Based on this slope residual, the consistency of the frequency-time relationship slope can be determined. For example, the consistency determination condition is "the slope residual does not exceed the preset residual threshold", and the residual threshold is associated with the frequency estimation error, the arrival time quantization error and the time span of the candidate correspondence, so as to ensure that the determination remains stable under different pulse widths and different channel segmentation patterns.

[0126] Based on this, when the frequency range continuity determination result meets the continuity determination condition and the frequency-time relationship slope consistency determination result meets the consistency determination condition, the cross-channel merging relationship can be determined and recorded as an element of the "cross-channel merging relationship set" for direct use when performing cross-channel merging and parameter recalculation in the future.

[0127] Furthermore, in a high-density pulse environment, the same initial pulse descriptor may simultaneously form candidate correspondences with multiple objects within a sliding time window, resulting in competition for merging relationships. In this case, a priority and verification mechanism needs to be introduced to ensure that one-to-one or one-to-many merging constraints do not conflict.

[0128] Specifically, since the slope of the frequency-time relationship reflects the consistency between time and frequency, when there are multiple candidate correspondences that can be used to determine cross-channel merging relationships, the residual of the frequency-time relationship slope can be used as the priority ranking criterion. Under this priority rule, the candidate correspondence with the smaller slope residual can be regarded as a more credible merging relationship. The candidate correspondence with the smallest slope residual and simultaneously satisfying the continuity condition is selected as the current merging relationship, thus selecting one candidate correspondence from multiple candidate correspondences as the cross-channel merging relationship. For the remaining unselected candidate correspondences, a consistency review is performed. After removing occupied objects, their continuity and slope consistency judgment is recalculated. If the conditions are still met after the review, they are converted into candidate correspondences with other unoccupied objects or retained to enter the next window position for further judgment. Finally, the candidate correspondences are updated to reflect the object occupancy status, review results, and the remaining candidate set, so that the matching and judgment of subsequent window positions can continue to advance under consistency constraints, avoiding the spread of conflicts accumulated across windows.

[0129] Thus far, step S3 organizes cross-channel matching through a sliding time window and achieves controllable convergence of candidate correspondences based on arrival time similarity and noise consistency. Within the candidate correspondence range, the merging relationship is further determined based on frequency range continuity and frequency-time slope consistency. At the same time, in the multi-candidate competition scenario, a priority selection and consistency verification mechanism based on slope residuals is introduced, forming a deterministic judgment system from "pulse detection output" to "cross-channel merging relationship", providing a reliable relationship input basis and conflict constraint boundary conditions for subsequent cross-channel merging and parameter recalculation.

[0130] In step S4, the core task is to implement the cross-channel merging relationship determined in step S3 into an executable merging operation, and to converge the initial pulse descriptor fragments scattered in adjacent channels into a single pulse-level description result. The input of this step is the cross-channel merging relationship and the set of initial pulse descriptors it covers, and the output is the complete pulse descriptor set, so that subsequent clustering and sorting based on angle of arrival and pulse width can be performed at the "complete pulse" granularity.

[0131] In this embodiment, initial pulse descriptors that satisfy cross-channel merging relationships need to be merged. Multiple initial pulse descriptors within the same cross-channel merging relationship can be considered as segmented observations of the same physical pulse on different channels, and merged according to the associated links given by the cross-channel merging relationship. The merged records retain only one pulse-level entry to avoid the same physical pulse being repeatedly counted in subsequent clustering statistics. Simultaneously, to ensure consistent parameters after merging, key fields need to be recalculated during the merging process to satisfy the joint consistency expression of cross-channel segments.

[0132] In some embodiments of this application, when recalculating the frequency range, pulse width, and frequency-time relationship slope, the frequency range can be used as the main boundary of the frequency domain coverage, the pulse width as the main boundary of the time duration, and the frequency-time relationship slope as the main consistency quantity characterizing the frequency change trend over time. All three are determined by the same segment covered by the same cross-channel merging relationship to ensure that the physical meaning used for subsequent clustering and sorting is stable and consistent.

[0133] Specifically, frequency range recalculation prioritizes coverage integrity to avoid omitting the boundaries of segmented segments. Therefore, a union operation is performed on the frequency ranges of each initial pulse descriptor that satisfies the cross-channel merging relationship to obtain the recalculated frequency range. For example, the frequency range of each initial pulse descriptor can be regarded as multiple intervals, and the recalculated frequency range can be taken as the joint coverage interval of these intervals. The lowest and highest frequency endpoints of the joint coverage interval are used as the endpoints of the recalculated frequency range. When there is partial overlap between cross-channel segments, the union operation will naturally eliminate duplicate coverage and retain the overall boundary, so that the frequency range is neither narrowed nor expanded without basis.

[0134] Furthermore, pulse width recalculation prioritizes the closure of time boundaries to avoid pulse widths being too short due to single-channel threshold jitter. Therefore, interval operations are performed on the arrival times of each initial pulse descriptor that satisfies the cross-channel merging relationship to obtain the recalculated pulse width. For example, the pulse event time intervals corresponding to each initial pulse descriptor can be uniformly mapped onto the same time axis, with the earliest arrival time taken as the start time of the recalculation and the latest end time taken as the end time of the recalculation, and the recalculated pulse width is obtained accordingly. When some segments are slightly truncated at the event boundaries due to threshold fluctuations within a single channel, this interval operation can use the boundaries of adjacent channel segments to complete the time coverage, thereby improving the stability of the pulse width.

[0135] Furthermore, the recalculated frequency-time relationship slope needs to be consistent with the recalculated frequency range to avoid trend deviation caused by fitting on local segments. Therefore, within the recalculated frequency range, a linear fit can be performed based on the arrival time of each initial pulse descriptor that satisfies the cross-channel merging relationship and the midpoint frequency of the frequency range to obtain the recalculated frequency-time relationship slope. For example, the arrival time of each initial pulse descriptor within the cross-channel merging relationship and its midpoint frequency can be combined to form a set of point pairs. A linear fit is performed on the set of point pairs under the constraint of the recalculated frequency range, and the slope in the fitting result is used as the recalculated frequency-time relationship slope. When the number of point pairs is insufficient to support stable fitting, the slope can be set to zero or set to an unusable flag, and the recalculated frequency range and pulse width can be retained to ensure that the output record is still available and does not introduce unstable estimation.

[0136] In this embodiment of the application, when generating a complete pulse descriptor, a merged record can be used as the carrier, and the recalculated frequency range, recalculated pulse width, and recalculated frequency-time relationship slope can be written into the record. At the same time, the threshold statistical parameter caliber associated with the cross-channel segment is inherited to maintain traceable noise background. The complete pulse descriptor corresponds one-to-one with the fields used in subsequent clustering and sorting, so that the cross-channel merging output can be directly used as the input for subsequent steps without having to trace back to the original segment again.

[0137] At this point, step S4 completes the conversion of cross-channel merging relationships into complete pulse descriptors, forming a parameter recalculation system that eliminates duplicate counts through merging operations, ensures frequency domain coverage through frequency range union, closes pulse width boundaries through time interval operations, and recalculates the slope of the frequency-time relationship through linear fitting. This provides a complete and consistent pulse-level input foundation for subsequent clustering and sorting based on angle of arrival and pulse width.

[0138] In step S5, the core task is to coarsely sort the complete pulse descriptors according to the spatial direction characteristics of the radiation source, and then further subdivide them according to the pulse width characteristics within each spatial group, so that pulse sets of different systems or different emission sources under the same spatial direction can be further separated. The input of this step is the complete pulse descriptor set and its angle of arrival and pulse width fields, and the output is the pulse width cluster set, which provides a structured input for subsequent frequency repetition separation and type determination on a purer set.

[0139] In this embodiment, spatial coarse sorting is achieved using two-dimensional histogram clustering. A two-dimensional histogram of arrival angles can be constructed based on the arrival angles of complete pulse descriptors and spatial clustering can be performed. The arrival angle two-dimensional space can be constructed using azimuth and elevation angles, and the landing point of each complete pulse descriptor in the two-dimensional space can be mapped to the corresponding bin, thereby forming a count statistics and performing connected component aggregation on the count distribution.

[0140] Specifically, the binning structure should be consistent with the angle of arrival (AHA) measurement error caliber to avoid either excessively fine binning that would fragment clusters or excessively coarse binning that would mix different radiation sources. Therefore, within the two-dimensional AHA space, the azimuth and pitch bin widths can be set according to the AHA measurement error range corresponding to the complete pulse descriptor, thus constructing two-dimensional AHA binning. The bin width can be set to be no less than the AHA measurement error range, and the bin width can be increased synchronously as the measurement error increases to maintain stable classification. For example, the bin width can be configured to be on the order of 2°, and the two-dimensional space can be divided into a fixed number of sub-intervals for connected component clustering.

[0141] After binning is completed, the complete pulse descriptors falling into each two-dimensional bin at the angle of arrival can be counted to form a two-dimensional histogram of the angle of arrival. For example, the number of occurrences can be accumulated in each bin, and bins with non-zero counts can be retained as candidate clustering units. When some bins have extremely low counts that are clearly caused by noise or isolated pulses, a minimum counting threshold can be set to temporarily exclude them from subsequent connected component merging, thereby reducing the interference of fragment clusters on subsequent subdivision. For example, the minimum counting threshold can be set to the order of 5 records to filter out accidental occurrences.

[0142] Furthermore, spatial cluster generation relies on adjacency relationships rather than global distance calculations to ensure that the clustering process can be implemented in real time. Therefore, spatial clustering can be performed based on the adjacency relationships between two-dimensional bins with angles of arrival. Connectivity components can be merged for two-dimensional bins with non-zero angles of arrival, and spatial clusters can be generated. For example, connectivity components can be constructed using bin adjacency rules of four or eight adjacencies. Bins that are adjacent to each other and have non-zero counts can be merged into the same connectivity component, and the connectivity component can be used as a spatial cluster. This approach can express the meaning of "similar directions" on a two-dimensional histogram with local connectivity and is convenient for keeping the computational scale under control when the amount of data increases.

[0143] It should be further explained that boundary binning often leads to unstable cluster assignments due to measurement errors, angle quantization, and environmental reflections. Therefore, boundary re-determination can be introduced to improve the stability of cluster boundaries. Specifically, for two-dimensional bins located at the arrival angles of spatial cluster boundaries, re-determination is performed based on the angular distance between the arrival angles corresponding to the bin centers to adjust the spatial cluster assignment relationships. For example, the angular distance from the center of the boundary bin to the center of the adjacent spatial cluster can be calculated, and its assignment can be adjusted when the angular distance meets the re-determination threshold condition, so that the boundary bins converge to more reasonable cluster centers, thereby reducing misclassification caused by inter-cluster overlap.

[0144] After spatial clusters are generated, records need to be collected into clusters for further subdivision. Therefore, complete pulse descriptors can be collected into corresponding spatial clusters. Each complete pulse descriptor is written into the spatial cluster set according to the spatial cluster identifier of the bin to which the record belongs, making each spatial cluster a pulse set to be subdivided. Subsequently, pulse width subdivision is performed within each spatial cluster, and a pulse width histogram is constructed based on the pulse width within the spatial cluster, and subdivision clustering is performed to obtain pulse width clusters. For example, histogram statistics can be performed on the pulse width field within the spatial cluster, setting a pulse width binning step that matches the pulse width measurement error and resolution. Bins with a percentage reaching a threshold can be selected on the histogram, or adjacent bins can be merged to form subdivision categories, thereby further splitting pulse sets of different pulse width systems under the same spatial direction. For example, the pulse width binning step can be on the order of 5%, and categories with a histogram percentage exceeding 10% can be used as output categories to suppress false subdivision categories caused by abnormal pulses.

[0145] Thus far, step S5 has completed the two-level clustering transformation from complete pulse descriptors to pulse width clusters, forming a hierarchical clustering system that uses the merging of connected domains in the two-dimensional histogram of arrival angle to achieve coarse spatial sorting, the boundary angle distance to determine the stable cluster affiliation, and the pulse width histogram within the spatial cluster to achieve finer clustering. This provides a structured cluster input basis for subsequent frequency repetition separation and type determination on purer sets.

[0146] In step S6, the core task is to extract a stable pulse repetition interval structure within each pulse width cluster and separate the multiple repetition frequency components that are mixed within the same cluster. The input of this step is the arrival time sequence within the pulse width cluster, and the output is the repetition frequency sequence, which determines the repetition frequency type so that the subsequent radiation source parameter set can contain repetition frequency features that can be used for identification and tracking.

[0147] In this embodiment, frequency repetition separation uses a multi-level differential histogram as the main method. A multi-level differential histogram is constructed for each pulse width cluster, and threshold determination, harmonic detection, sequence search, and iterative elimination are performed. First, a sequence is formed within the pulse width cluster by arrival time. Differential statistics are performed on adjacent arrival time intervals to form a differential histogram, and threshold determination is performed on the differential histogram to generate candidate repetition frequencies. When multiple candidate peaks are obtained from the threshold determination, harmonic detection is then performed to suppress spurious peaks with integer multiple relationships. Sequence search is then performed on the retained candidates to verify whether the candidate repetition frequency can form a sufficiently long pulse link in the original sequence. Successfully verified pulse links are eliminated, and the process returns to differential histogram reconstruction until the remaining pulses within the cluster are insufficient to form a stable frequency repetition structure.

[0148] For example, threshold determination can be achieved by:

[0149]

[0150] in, Time interval The threshold for judgment at the place, This represents the total number of pulses. For difference fractions, and It is a positive number less than 1, used to control the rate and magnitude of threshold decay over time.

[0151] In some embodiments of this application, when the differential histogram of the pulse width cluster exhibits multi-peak competition (e.g., the ratio of the peak value of the main peak to the peak value of the secondary peak in the differential histogram meets a preset ratio threshold condition), the differential level is switched and a higher-order differential histogram is constructed. Threshold determination is performed based on the higher-order differential histogram to obtain candidate repetition frequencies. Specifically, multi-peak competition can be understood as "the competition intensity of two or more peaks is close, making it difficult to determine the dominant repetition frequency based solely on the first-order differential histogram." When the ratio threshold condition is met, the differential level is switched from a lower order to a higher order to reduce the interference of unevenness, jitter, and aliasing on the peak shape, making candidate repetition frequencies more easily highlighted. The construction of the higher-order differential histogram is still completed according to the differential interval statistics, and the threshold determination is still performed according to the same threshold function form to maintain consistency in the determination chain.

[0152] Furthermore, to suppress harmonic spurious peaks caused by integer multiple relationships, harmonic detection based on the consistency index of integer multiple relationships is performed on candidate repetition frequencies, dividing them into a retention set and a rejection set. For example, the consistency of integer multiple relationships between each candidate repetition frequency can be checked. If a candidate repetition frequency can be explained by another candidate repetition frequency through an integer multiple relationship and the consistency index meets the threshold condition, it is included in the rejection set. The retention set retains candidates that are more likely to correspond to the basic repetition frequency components, so as to reduce the probability of false locking in subsequent sequence searches.

[0153] After harmonic detection, the repeatability of the candidates needs to be verified in the original arrival time series. Therefore, for the retained set, a sequence search can be performed within the pulse width cluster according to the arrival time interval corresponding to the candidate repetition frequency to obtain a pulse set consistent with the candidate repetition frequency. For example, the candidate repetition frequency can be used as the expected interval, and the subsequent pulses that meet the interval tolerance can be searched in the sequence from the starting pulse along the time axis to form the pulse link corresponding to the candidate repetition frequency. When the link length reaches the preset minimum number condition, the link is output as a pulse set consistent with the candidate repetition frequency to ensure that the output repetition frequency has sufficient statistical support.

[0154] It should be further explained that frequency repetition separation requires the elimination of multiple components to avoid duplicate counting. Therefore, pulses consistent with candidate repetition frequencies can be removed from the pulse width cluster and the difference histogram updated. The process of difference histogram construction, threshold determination, harmonic detection, and sequence search can be repeated until the frequency repetition separation within the pulse width cluster is completed. For example, pulses that have been assigned to a certain frequency repetition link can be removed from the current cluster sequence and the difference histogram can be reconstructed to reveal the frequency repetition structure of the remaining pulses. This process is repeated iteratively until the number of remaining pulses is insufficient to form a stable link or the histogram order reaches a preset upper limit, thereby completing the separation of multiple frequency components within the cluster.

[0155] Once the repetition frequency sequence is determined, it can be output. At this point, each separated candidate repetition frequency and its corresponding pulse set can be output together, and auxiliary information such as the difference level number and link length can be recorded for subsequent consistency maintenance. Subsequently, the repetition frequency type is determined based on the repetition frequency sequence, including fixed repetition frequency, staggered repetition frequency, and jitter repetition frequency. For example, the type can be determined based on the stability and multi-valued structure of the repetition frequency sequence: when the repetition frequency sequence shows a single stable interval and small fluctuations within the tolerance, it can be determined as a fixed repetition frequency; when the repetition frequency sequence shows a finite number of discrete intervals and can form a repeating combination pattern, it can be determined as a staggered repetition frequency; when the repetition frequency sequence shows continuous diffusion or random fluctuation characteristics within the tolerance range, it can be determined as a jitter repetition frequency. The type result is output along with the repetition frequency sequence to support subsequent identification and tracking.

[0156] At this point, step S6 completes the repetition frequency separation and type determination within the pulse width cluster, forming a repetition frequency sorting system that generates candidates using multi-level differential histogram threshold determination, enhances separability by triggering higher-order differentials through multi-peak competition, suppresses harmonic pseudo-peaks using integer multiple relationship consistency index, verifies the link through sequence search, and achieves multi-component decomposition through iterative elimination. This results in repetition frequency sequences and repetition frequency type results that can be used for radiation source identification and parameter set output, providing a reliable time-series feature input basis for subsequent parameter aggregation and display control.

[0157] In summary, this method achieves wideband, blind-zone-free parallel reception and real-time detection through multiphase filtering and digital channelization. It combines threshold statistical parameter constraints and arrival time similarity constraints to suppress noise fluctuations and accidental matching. Based on the determination of frequency range continuity and frequency-time relationship slope consistency, it completes reliable cross-channel pulse merging and recalculates key parameters. Furthermore, it improves pulse set purity through spatial clustering of the angle of arrival two-dimensional histogram and subdivided clustering of the pulse width histogram. It employs multi-level differential histogram threshold determination, harmonic suppression, sequence search, and iterative elimination to extract repetition frequency sequences and distinguish between fixed, uneven, and jitter types. This significantly reduces misclassification and missed classification in complex, high-density electromagnetic environments, improves the integrity of wide-pulse parameter measurement and the stability of repetition frequency identification, enhances the adaptive capability and real-time performance of the processing flow, thereby improving the accuracy of radiation source identification and the reliability of situational awareness, and reducing the cost of manual intervention and subsequent analysis.

[0158] It should be noted that, although the embodiments in this application are based on... Figure 1 Steps S1 to S6 are described sequentially, but this does not mean that steps S1 to S6 must be performed in a strict order. The reason this embodiment follows this order is... Figure 1The order in which steps S1 to S6 are described is provided to facilitate understanding of the technical solutions of the embodiments of this application by those skilled in the art. In other words, in the embodiments of this application, the order of steps S1 to S6 can be appropriately adjusted according to actual needs.

[0159] In some embodiments of this application, the "radio reconnaissance method based on multi-channel parallel processing and intelligent clustering" is applied to non-cooperative radio signal search and processing scenarios in complex electromagnetic environments, with the broadband acquisition link outputting discrete sampling sequences. The method runs on a parallel processing architecture. The front end completes real-time channelization and pulse detection, while the back end completes cross-channel merging, clustering and sorting, and frequency repetition identification, forming a complete processing process from broadband input to radiation source parameter set output.

[0160] The complete implementation process may include the following steps:

[0161] Step 1: The broadband acquisition link performs down-conversion and analog-to-digital conversion on the spatial electromagnetic signal to obtain a discrete sampling sequence. And set the number of channels. The value is 64, which allows the wideband input to be divided into 64 adjacent and overlapping channels simultaneously to achieve blind-zone-free reception.

[0162] Step 2, based on the polyphase filter structure... Perform digital channelization processing and output channel sequences. and for each Associated channel frequency identifier ,in , This is the index of the extracted sequence; the extraction factor is determined according to the structure shown in the diagram. This creates overlapping frequency bands between adjacent channels, providing a frequency domain basis for determining the continuity of subsequent frequency ranges.

[0163] Step 3, for each channel sequence The square-law detection is performed, converting the complex sampling points into a power sequence, which is then used as the input for constant false alarm rate (CFAR) detection. The CFAR detection generates threshold statistics within a sliding window. , and by The detection threshold is obtained, and continuous sampling intervals exceeding the threshold are aggregated into pulse events; an initial pulse descriptor is generated for each pulse event. ,in Including arrival time Pulse width Amplitude With frequency range And threshold statistics parameters Associated Write This is used for subsequent noise consistency constraints.

[0164] Step four: Construct a sliding time window between the initial pulse descriptor sets of adjacent channels; establish candidate correspondences for initial pulse descriptors that fall within the same sliding time window and satisfy the arrival time similarity constraint; compare threshold statistical parameters within the candidate correspondences. And apply noise consistency constraints, the noise consistency constraints are based on The criteria for judgment are to meet the preset tolerance conditions, thereby eliminating random time alignment candidates caused by noise mutations or threshold fluctuations.

[0165] Step 5: Calculate the frequency range endpoint difference and frequency range overlap within the candidate correspondence to determine the frequency range continuity; simultaneously, within the candidate correspondence, calculate the frequency range endpoint difference based on arrival time. With midpoint frequency Construct a frequency-time relationship, perform linear fitting on the frequency-time relationship, and calculate the slope residual of the frequency-time relationship. ,based on The consistency of the frequency-time relationship slope is determined; when the frequency range continuity determination result meets the continuity determination condition and the slope consistency determination result meets the consistency determination condition, the cross-channel merging relationship is determined.

[0166] Step 6: When multiple candidate correspondences can be used to determine cross-channel merging relationships, calculate the slope residual. The priority rule selects candidate correspondences, so that Smaller candidate correspondences are given priority as cross-channel merging relationships; consistency checks are performed on unselected candidate correspondences, and candidate correspondences are updated based on the check results to avoid the same initial pulse descriptor being repeatedly occupied or causing conflict merging in adjacent channels.

[0167] Step 7: Merge the initial pulse descriptors that satisfy the cross-channel merging relationship to generate complete pulse descriptors. Frequency range recalculation affects the coverage of each merge relationship. Performing the union operation yields Pulse width recalculation performs interval operations on the arrival time interval to obtain the recalculated value. and within the recalculated frequency range Linear fitting was performed to obtain the slope of the recalculated frequency-time relationship. ,Will Write ,in Take the earliest arrival time covered by the merge relationship.

[0168] Step 8: Based on the complete pulse description word, construct a two-dimensional histogram of arrival angles in the two-dimensional space of arrival angles and perform spatial clustering; the two-dimensional space of arrival angles uses azimuth and elevation angles as coordinates, and the arrival angle can be denoted as... The bin width is set according to the measurement error range of the angle of arrival, the bin step is 2°, and the two-dimensional space is divided into 32×32 sub-intervals. For each sub-interval, the bin width is set. The counting process forms a two-dimensional histogram. Subintervals with counts greater than 5 are selected as valid regions. Adjacent subintervals with non-zero counts are merged according to their adjacency relationship to generate spatial clusters. For the boundary subintervals of the spatial clusters, the binning center angle is used as the basis for... The angular distance between them is re-evaluated to adjust the cluster affiliation, making the spatial cluster boundaries more stable.

[0169] Step 9: Construct a pulse width histogram based on pulse width within each spatial cluster and perform subdivision clustering. The pulse width histogram is stepped by 5%. The category with a histogram percentage of more than 10% is taken as the output category to obtain the pulse width cluster. This further separates the pulse sets of different pulse width systems in the same spatial direction and reduces abnormal pulse interference.

[0170] Step 10: Construct a multi-level difference histogram for each pulse width cluster and perform threshold determination, harmonic detection, sequence search, and iterative elimination to output the repetition frequency sequence and determine the repetition frequency type; the threshold determination uses an exponential decay threshold function.

[0171]

[0172] in, The time intervals on the horizontal axis of the difference histogram are represented by... Indicates the total number of pulses within the pulse width cluster. Represents the difference fraction. and The value is a positive number less than 1. When the difference histogram shows multi-peak competition, the difference level is switched and a higher-order difference histogram is constructed to improve separability. Multi-peak competition is judged based on the ratio of the peak value of the main peak to the peak value of the secondary peak meeting a preset ratio threshold. For candidate repetition frequencies that exceed the threshold, harmonic detection is performed based on the consistency index of integer multiple relationship to form a retention set and a rejection set. For the retention set, a sequence search is performed according to the arrival time interval corresponding to the candidate repetition frequency. When the number of pulses obtained by the search is greater than 5, the pulse set is identified as a consistent sequence and removed from the cluster. The difference histogram is updated and iterated repeatedly until the number of remaining pulses is less than 5 or the difference level reaches the upper limit. Based on the output repetition frequency sequence, fixed repetition frequency, staggered repetition frequency and jittered repetition frequency are determined, and staggered patterns are merged to form a radiation source parameter set.

[0173] Through the complete implementation process described above, this method enhances the real-time interception capability of wideband through blind-zone-free channelized parallel reception, suppresses false correspondences caused by noise fluctuations through threshold statistical parameter constraints and time proximity constraints, achieves reliable cross-channel wide pulse merging and improves the integrity of parameter measurements through frequency range continuity and frequency-time slope consistency, improves pulse set purity by combining two-dimensional histogram spatial clustering of angle of arrival and pulse width histogram subdivision clustering, and stabilizes and separates complex repetition types such as unevenness and jitter through multi-level differential histogram threshold determination, harmonic suppression, sequence search and iterative elimination. It can significantly reduce false alarms and missed alarms in high-density, strongly overlapping signal environments, improve sorting accuracy, repetition identification stability and processing efficiency, enhance process adaptability and engineering real-time performance, thereby improving the reliability of radiation source identification and situational awareness and reducing the burden of manual analysis.

[0174] It should be understood that the step numbers identified by "Step 1, Step 2" and other similar forms in the above embodiments are only used to distinguish different steps and do not limit the steps to be executed in the order of these numbers. The specific execution order of each step can be adjusted according to its functional requirements and the inherent logic in the actual application scenario. The above step numbers should not be interpreted as a limitation on the implementation process of the embodiments of this application.

[0175] like Figure 2 As shown, the following is an embodiment of a radio reconnaissance system based on multi-channel parallel processing and intelligent clustering provided by this disclosure. This radio reconnaissance system based on multi-channel parallel processing and intelligent clustering belongs to the same inventive concept as the radio reconnaissance methods based on multi-channel parallel processing and intelligent clustering in the above embodiments. For details not described in detail in the embodiments of the radio reconnaissance system based on multi-channel parallel processing and intelligent clustering, please refer to the embodiments of the radio reconnaissance methods based on multi-channel parallel processing and intelligent clustering described above.

[0176] Based on the same concept, another embodiment of this application provides a radio reconnaissance system based on multi-channel parallel processing and intelligent clustering, comprising:

[0177] Multi-channelization module 1 is used to perform multi-phase filtering digital channelization processing on broadband radio signals to obtain adjacent and overlapping channel sequences;

[0178] The detection generation module 2 is used to perform square law detection and constant false alarm rate (CFAR) detection on each channel sequence and generate an initial pulse descriptor. The CFAR detection includes generating threshold statistical parameters and associating them with the initial pulse descriptor. The initial pulse descriptor includes arrival time, pulse width, amplitude, and frequency range.

[0179] Candidate correspondence module 3 is used to establish candidate correspondences of initial pulse descriptors in adjacent channels based on arrival time proximity constraints and noise consistency constraints corresponding to threshold statistical parameters;

[0180] The merging determination module 4 is used to determine the cross-channel merging relationship based on the continuity of the frequency range and the consistency of the slope of the frequency-time relationship within the candidate correspondence relationship;

[0181] The parameter recalculation module 5 is used to merge the initial pulse descriptors that satisfy the cross-channel merging relationship, and recalculate the frequency range, pulse width and frequency-time relationship slope to generate a complete pulse descriptor.

[0182] Spatial clustering module 6 is used to construct a two-dimensional histogram of arrival angles based on the angle of arrival of complete pulse descriptors and perform spatial clustering to group complete pulse descriptors into corresponding spatial clusters;

[0183] The subdivision clustering module 7 is used to construct a pulse width histogram based on pulse width within a spatial cluster and perform subdivision clustering to obtain pulse width clusters;

[0184] The frequency repetition separation module 8 is used to construct a multi-level differential histogram for the pulse width cluster and perform threshold determination, harmonic detection, sequence search and iterative elimination, output the frequency repetition sequence, and determine the frequency repetition type based on the frequency repetition sequence. The frequency repetition types include fixed frequency repetition, staggered frequency repetition and jittery frequency repetition.

[0185] In some embodiments of this application, the frequency repetition rate separation module 8 includes:

[0186] The level switching submodule is used to switch the difference level and construct a higher-order difference histogram when the difference histogram of the pulse width cluster presents multi-peak competition. Multi-peak competition is defined as the ratio of the peak value of the main peak to the peak value of the secondary peak in the difference histogram satisfying a preset ratio threshold condition.

[0187] The threshold determination submodule is used to perform threshold determination based on the high-order difference histogram to obtain candidate repetition frequencies;

[0188] The harmonic detection submodule is used to perform harmonic detection on candidate repetition frequencies based on the consistency index of integer multiple relationship, and divide them into a retention set and a rejection set;

[0189] The sequence search submodule is used to perform a sequence search within the pulse width cluster for the retained set according to the arrival time interval corresponding to the candidate repetition frequency, so as to obtain a pulse set that is consistent with the candidate repetition frequency.

[0190] The iterative elimination submodule is used to remove pulses that match the candidate repetition frequency from the pulse width cluster and update the difference histogram. It then repeats the difference histogram construction, threshold determination, harmonic detection, and sequence search until the repetition frequency separation within the pulse width cluster is completed.

[0191] In summary, this system, through the collaboration of multi-channel information processing module 1, detection and generation module 2, candidate correspondence module 3, merging and judgment module 4, parameter recalculation module 5, spatial clustering module 6, subdivision clustering module 7, and frequency repetition separation module 8, completes parallel detection of broadband radio signals, consistent merging of cross-channel pulses, hierarchical clustering and sorting, and frequency repetition sequence extraction. It achieves orderly separation and frequency repetition type discrimination of high-density multi-type pulse signals, and can maintain the stability and consistency of cross-channel pulse parameter characterization and sorting results in broadband radio search scenarios, providing a stable data processing foundation for reliable identification and analysis of radiation source signals.

[0192] The above-disclosed embodiments are merely preferred embodiments of the present invention, but the present invention is not limited thereto. Any non-creative variations that can be conceived by those skilled in the art, as well as any improvements and modifications made without departing from the principles of the present invention, should fall within the protection scope of the present invention.

Claims

1. A radio reconnaissance method based on multi-channel parallel processing and intelligent clustering, characterized in that, include: Polyphase filtering and digital channelization processing are performed on broadband radio signals to obtain adjacent and overlapping channel sequences. Square-law detection and constant false alarm rate (CFAR) detection are performed on each channel sequence to generate an initial pulse descriptor. CFAR detection includes generating threshold statistical parameters and associating them with the initial pulse descriptor. The initial pulse descriptor includes arrival time, pulse width, amplitude, and frequency range. Based on the time-of-arrival proximity constraint and the noise consistency constraint corresponding to the threshold statistical parameters, candidate correspondences of initial pulse descriptors are established in adjacent channels, and cross-channel merging relationships are determined within the candidate correspondences based on the continuity of frequency range and the consistency of frequency-time relationship slope. Initial pulse descriptors that satisfy the cross-channel merging relationship are merged, and the frequency range, pulse width, and frequency-time relationship slope are recalculated to generate complete pulse descriptors; A two-dimensional histogram of arrival angle is constructed based on the angle of arrival of the complete pulse descriptor and spatial clustering is performed to group the complete pulse descriptors into the corresponding spatial clusters. Within the spatial clusters, a pulse width histogram is constructed based on the pulse width and subdivision clustering is performed to obtain the pulse width clusters. A multi-level difference histogram is constructed for the pulse width cluster, and threshold determination, harmonic detection, sequence search and iterative elimination are performed to output the repetition frequency sequence. The repetition frequency type is determined based on the repetition frequency sequence, which includes fixed repetition frequency, staggered repetition frequency and jitter repetition frequency.

2. The radio reconnaissance method based on multi-channel parallel processing and intelligent clustering as described in claim 1, characterized in that, The steps of performing polyphase filtering and digital channelization processing on broadband radio signals to obtain channel sequences include: constructing a polyphase filter bank based on a prototype filter and performing parallel filtering and extraction on the broadband radio signals to obtain channel sequences, and associating each channel sequence with a channel frequency identifier, which is used to determine the continuity of the frequency range.

3. The radio reconnaissance method based on multi-channel parallel processing and intelligent clustering as described in claim 1, characterized in that, The steps for establishing candidate correspondences of initial pulse descriptors in adjacent channels based on arrival time proximity constraints and noise consistency constraints corresponding to threshold statistical parameters include: constructing a sliding time window in the set of initial pulse descriptors in adjacent channels; establishing candidate correspondences for initial pulse descriptors that satisfy arrival time proximity constraints within the sliding time window; and filtering based on noise consistency constraints corresponding to threshold statistical parameters, wherein the noise consistency constraints include the threshold statistical parameters satisfying a preset tolerance condition.

4. The radio reconnaissance method based on multi-channel parallel processing and intelligent clustering as described in claim 1, characterized in that, The steps for determining cross-channel merging relationships within candidate correspondences based on frequency range continuity and frequency-time relationship slope consistency include: Within the candidate correspondence, calculate the frequency range endpoint difference between each initial pulse descriptor word, calculate the frequency range overlap, and determine the frequency range continuity. Within the candidate correspondence, a frequency-time relationship is constructed based on the arrival time and midpoint frequency of the frequency range of each initial pulse descriptor. The frequency-time relationship is linearly fitted and the slope residual of the frequency-time relationship is calculated. The consistency of the slope of the frequency-time relationship is determined based on the slope residual of the frequency-time relationship. When the frequency range continuity determination result meets the continuity determination condition, and the frequency-time relationship slope consistency determination result meets the consistency determination condition, the cross-channel merging relationship is determined.

5. The radio reconnaissance method based on multi-channel parallel processing and intelligent clustering as described in claim 4, characterized in that, When there are multiple candidate correspondences that can be used to determine the cross-channel merging relationship, a candidate correspondence is selected as the cross-channel merging relationship according to the priority rule of the frequency-time relationship slope residual. Consistency verification is performed on the candidate correspondences that are not selected, and the candidate correspondences are updated.

6. The radio reconnaissance method based on multi-channel parallel processing and intelligent clustering as described in claim 4, characterized in that, The steps for recalculating the frequency range, pulse width, and frequency-time relationship slope include: Perform a union operation on the frequency ranges of each initial pulse descriptor that satisfies the cross-channel combining relationship to obtain the recalculated frequency range; For each initial pulse descriptor that satisfies the cross-channel combining relationship, an interval operation is performed on the arrival time to obtain the recalculated pulse width; Within the recalculated frequency range, a linear fit is performed based on the arrival time of each initial pulse descriptor that satisfies the cross-channel combining relationship and the midpoint frequency of the frequency range to obtain the slope of the recalculated frequency-time relationship.

7. The radio reconnaissance method based on multi-channel parallel processing and intelligent clustering as described in claim 1, characterized in that, The steps of constructing a two-dimensional histogram of arrival angles based on the complete pulse descriptor and performing spatial clustering include: Within the two-dimensional space of the angle of arrival, the azimuth direction bin width and the pitch direction bin width are set according to the angle of arrival measurement error range corresponding to the complete pulse description word, and the two-dimensional bin of the angle of arrival is constructed. The complete pulse description words falling into each two-dimensional bin at the angle of arrival are counted to form a two-dimensional histogram of the angle of arrival. Based on the adjacency relationship between the two-dimensional bins of the angle of arrival, spatial clustering is performed, and connected component merging is performed on the two-dimensional bins of the angle of arrival with non-zero counts to generate spatial clusters; For two-dimensional bins with arrival angles located at the boundaries of spatial clusters, a re-determination is performed based on the angular distance between the arrival angles corresponding to the bin centers to adjust the affiliation of spatial clusters.

8. The radio reconnaissance method based on multi-channel parallel processing and intelligent clustering as described in claim 1, characterized in that, The steps of constructing a multi-level difference histogram for pulse width clusters and performing threshold determination, harmonic detection, sequence search, and iterative elimination include: When the differential histogram of the pulse width cluster exhibits multi-peak competition, the differential level is switched and a higher-order differential histogram is constructed. Threshold determination is performed based on the higher-order differential histogram to obtain candidate repetition frequencies. Here, multi-peak competition means that the ratio of the peak value of the main peak to the peak value of the secondary peak in the differential histogram meets the preset ratio threshold condition. Harmonic detection based on integer multiple relationship consistency index is performed on candidate repetition frequencies, and the repetition frequencies are divided into a retention set and a rejection set; For the retained set, a sequence search is performed within the pulse width cluster according to the arrival time interval corresponding to the candidate repetition frequency to obtain the pulse set that is consistent with the candidate repetition frequency; Pulses matching the candidate repetition frequency are removed from the pulse width cluster and the difference histogram is updated. The process of difference histogram construction, threshold determination, harmonic detection, and sequence search is repeated until the repetition frequency separation within the pulse width cluster is completed.

9. A radio reconnaissance system based on multi-channel parallel processing and intelligent clustering, characterized in that, include: The multi-channelization module is used to perform polyphase filtering and digital channelization processing on broadband radio signals to obtain adjacent and overlapping channel sequences. The detection generation module is used to perform square law detection and constant false alarm rate (CFAR) detection on each channel sequence and generate an initial pulse descriptor. CFAR detection includes generating threshold statistical parameters and associating them with the initial pulse descriptor. The initial pulse descriptor includes arrival time, pulse width, amplitude, and frequency range. The candidate correspondence module is used to establish candidate correspondences of initial pulse descriptors in adjacent channels based on the arrival time proximity constraint and the noise consistency constraint corresponding to the threshold statistical parameters. The merging determination module is used to determine the cross-channel merging relationship within the candidate correspondence based on the continuity of the frequency range and the consistency of the frequency-time relationship slope. The parameter recalculation module is used to merge initial pulse descriptors that satisfy the cross-channel merging relationship, and recalculate the frequency range, pulse width and frequency-time relationship slope to generate a complete pulse descriptor; The spatial clustering module is used to construct a two-dimensional histogram of arrival angles based on the complete pulse descriptor and perform spatial clustering to group the complete pulse descriptors into the corresponding spatial clusters; The subdivision clustering module is used to construct pulse width histograms based on pulse width within spatial clusters and perform subdivision clustering to obtain pulse width clusters; The frequency repetition separation module is used to construct a multi-level differential histogram for pulse width clusters and perform threshold determination, harmonic detection, sequence search and iterative elimination, output the frequency repetition sequence, and determine the frequency repetition type based on the frequency repetition sequence. The frequency repetition types include fixed frequency repetition, staggered frequency repetition and jittery frequency repetition.

10. The radio reconnaissance system based on multi-channel parallel processing and intelligent clustering as described in claim 9, characterized in that, The frequency repetition separation module includes: The level switching submodule is used to switch the difference level and construct a higher-order difference histogram when the difference histogram of the pulse width cluster presents multi-peak competition. Multi-peak competition is defined as the ratio of the peak value of the main peak to the peak value of the secondary peak in the difference histogram satisfying a preset ratio threshold condition. The threshold determination submodule is used to perform threshold determination based on the high-order difference histogram to obtain candidate repetition frequencies; The harmonic detection submodule is used to perform harmonic detection on candidate repetition frequencies based on the consistency index of integer multiple relationship, and divide them into a retention set and a rejection set; The sequence search submodule is used to perform a sequence search within the pulse width cluster for the retained set according to the arrival time interval corresponding to the candidate repetition frequency, so as to obtain a pulse set that is consistent with the candidate repetition frequency. The iterative elimination submodule is used to remove pulses that match the candidate repetition frequency from the pulse width cluster and update the difference histogram. It then repeats the difference histogram construction, threshold determination, harmonic detection, and sequence search until the repetition frequency separation within the pulse width cluster is completed.

Images