A Parallel Adaptive Clustering Method for Radar Pulse Signal Sorting

Abstract

This invention discloses a parallel adaptive clustering method for radar pulse signal sorting. The implementation steps are as follows: gridding of a three-dimensional data space composed of radar pulse descriptor frequency, pulse width, and pulse angle of arrival; adaptive adjustment of clustering algorithm parameters in parallel within the sub-data space; adaptive Birch clustering of the sub-data space; and measurement of differences between clusters and merging of clusters after clustering. The radar pulse signal sorting method proposed in this invention solves the problems of existing methods being unable to quickly complete sorting tasks when the pulse signal data volume is large, and the sorting method's accuracy being limited by parameter selection, making it unsuitable for the measurement accuracy requirements of dense and complex electromagnetic environments. It is applicable to modern electromagnetic environments with severe aliasing and complexity, improves the level of radar signal analysis technology, and can be widely applied to sorting modern radar pulse signals.

Description

Technical Field

[0001] This invention belongs to the field of communication technology, and more specifically relates to a parallel adaptive clustering method for radar pulse signal sorting within the field of radar communication technology. This invention can be used to separate pulse signals belonging to the same radar radiation source, facilitating the analysis of the separated pulse signals. Background Technology

[0002] Radar pulse signal sorting, also known as radar pulse signal deinterleaving, refers to the process of separating individual radar pulse sequences from a randomly interleaved pulse stream. Radar pulse signal sorting primarily distinguishes pulse sequences emitted by different radar sources based on their characteristic differences. By processing the radar pulse signals received by the radar receiver, information that distinguishes different radar sources is extracted, and pulse signals belonging to the same radar source are grouped into the same category, thus achieving radar pulse signal sorting. Due to the complexity of the electromagnetic environment, radar often receives high-density, heavily aliased radar pulse signal sequences, requiring the sorting of a large number of pulses and incurring significant computational demands. Furthermore, the heavily aliased radar pulse sequences received by the radar receiver are difficult to deinterleave, resulting in low sorting accuracy.

[0003] Southwest Jiaotong University disclosed a radar signal sorting method based on a sphere-covering biomimetic pattern recognition algorithm in its patent application document "A Radar Signal Sorting Method Based on a Sphere Covering Bionic Pattern Recognition Algorithm" (application number 202011187495.5, publication number: CN 112014804 A). The implementation steps of this method are as follows: (1) Obtain radar pulse signal data of various types as sample data, clean the false data in the sample data, fill the missing values ​​in the cleaned sample data, and finally normalize the sample data; (2) Perform dimensionality upgrade processing from n-dimensional to n+1-dimensional on the preprocessed sample data so that the Euclidean distance from all sample points to the origin of the coordinate axis is equal in the n+1-dimensional space, that is, map all data onto the sphere S. n+1 On the sphere S, the samples are divided into training and testing sets according to a preset ratio; (3) on the sphere S n+1 Find the coverage center and coverage radius, construct spherical coverages one by one, until all training set sample points are trained to obtain the classification coverage model; (4) Input all test sets into the classification coverage model, and determine the category of the test sample based on the category of the spherical region into which the sample falls. The shortcomings of this method are that there are many processing steps for sample data, and the processing time and sorting time will increase when the amount of data is large. When the amount of pulse signal data is large, the sorting task cannot be completed quickly.

[0004] Southeast University disclosed a radar pulse sequence sorting method based on image features in its patent application document "A radar pulse sequence sorting method based on image features" (application number 202110136843.4, application publication number: CN 112986925 A). The implementation steps of this method are as follows: First, the radar pulse descriptor set is pre-sorted based on frequency and histogram statistics, dividing the original pulse descriptor set into several pulse descriptor subsets. Second, pulse descriptors are constructed using arrival time, frequency, amplitude, phase, and pulse width, combined with the pulse sequence number. Third, for each pulse descriptor subset, image processing is performed based on arrival time and phase, and Hough segments are extracted using the Hough transform method. Then, several Hough segment clusters are obtained using the DBSCAN clustering method. Fourth, the obtained image data and Hough segment clusters are phase-periodicly extended, and a second DBSCAN clustering is performed to obtain Hough segment cluster classification labels. Then, radar radiation source information is obtained based on hierarchical clustering, and pulse sequence classification is performed. Fifth, based on the obtained hierarchical clustering results, signal capture is performed for each original pulse signal point, and each point is assigned to a certain radar category or classified as noise signal that does not belong to any radar category. Sixth, the classification results are analyzed to obtain the radar radiation source pulse repetition period type and optimize the redundancy sorting results. The drawback of this method is that the parameters of the DBSCAN algorithm are fixed, and the algorithm parameters cannot be dynamically adjusted according to the data distribution of the divided pulse descriptor subset. The final sorting accuracy will be limited by the choice of parameters and cannot adapt to the measurement accuracy requirements of dense and complex electromagnetic environments.

[0005] In summary, existing radar pulse signal sorting methods are not ideal. Due to the numerous data processing steps, the processing and sorting times increase with large data volumes, making it difficult to complete the sorting task quickly when the pulse signal data volume is large. At the same time, the sorting accuracy is limited by the fixed sorting algorithm parameters, which cannot adapt to the measurement accuracy requirements of dense and complex electromagnetic environments. Summary of the Invention

[0006] The purpose of this invention is to address the shortcomings of the existing technology by proposing a parallel adaptive clustering radar pulse signal sorting method. This method solves the problems of existing methods being unable to quickly complete the sorting task when the amount of pulse signal data is large, and the sorting accuracy being limited by the selection of parameters, making it unable to adapt to the measurement accuracy requirements of dense and complex electromagnetic environments.

[0007] The technical approach to achieving the objective of this invention is as follows: This invention utilizes a grid partitioning and parallel processing technique for the pulse descriptor data space. This technique enables simultaneous data processing and clustering algorithms for multiple data subspaces, addressing the problem in existing technologies where processing and sorting times increase with large data volumes, hindering rapid sorting when pulse signal data is abundant. Furthermore, this invention employs an adaptive Birch clustering algorithm. This algorithm adaptively adjusts its parameters based on the data distribution of pulse descriptors within the partitioned data subspaces, overcoming the limitation of existing sorting methods' accuracy by parameter selection, which cannot adapt to the measurement accuracy requirements of dense and complex electromagnetic environments.

[0008] To achieve the above objectives, the technical solution adopted by the invention includes the following steps:

[0009] Step 1: Based on the three-dimensional feature space composed of carrier frequency, pulse width, and pulse arrival angle, perform mesh division to divide the original pulse descriptor feature space into multiple pulse descriptor subspaces;

[0010] Step 2: The pulse description word subspace is divided into clusters in parallel, and the range of carrier frequency, pulse width, and pulse angle of arrival values ​​of each cluster in the sub-block is calculated in parallel.

[0011] Step 3: Take the median values ​​of carrier frequency, pulse width, and pulse arrival angle span in each sub-block in parallel, and use them as the feature unit length of the corresponding pulse descriptor in that sub-block;

[0012] Step 4: Adaptively adjust the radius parameter and maximum number of clusters parameter of the Birch clustering algorithm using the comprehensive scaling factor of the impulse descriptor subspace; perform the Birch clustering algorithm in parallel on each impulse descriptor subspace using the adaptive parameters, and generate pseudo-labels for the resulting clusters.

[0013] Step 5: Measure the feature similarity of each result cluster after sorting, unify the pseudo-labels of the clusters that meet the merging conditions to the same label value, realize the merging of result clusters, and complete the sorting of radar pulse signals.

[0014] Furthermore, the steps for mesh generation based on the three-dimensional feature space composed of carrier frequency, pulse width, and pulse angle of arrival are as follows:

[0015] The first step is to receive pulse descriptor data, which consists of five pulse descriptors: pulse arrival time, carrier frequency, pulse width, pulse arrival angle, and pulse amplitude.

[0016] The second step is to follow the preset granularity N = [n 载频 ,n 脉宽 ,n 脉冲到达角The three-dimensional feature space is divided into multiple sub-blocks, each corresponding to a three-dimensional mesh element. Its data set can be represented as follows:

[0017] D ijk ={x(1),x(2),x(3),…x(t),…x(n)}

[0018] Among them, D ijk The data space is divided into pulse descriptor sub-blocks, where i, j, and k represent the indices of the carrier frequency, pulse width, and pulse angle of arrival, respectively; n represents the number of received pulses; and x(t) represents the set of pulse descriptors corresponding to the t-th pulse data point.

[0019] The third step is to calculate the density of each sub-block using the following formula:

[0020]

[0021] Where, ρ ijk |D represents the data density of a sub-block. ijk | represents the sum of the absolute values ​​of all elements within this sub-block, V ijk The pulse description word sub-block data space D ijk Volume;

[0022] The fourth step is to set an upper and lower threshold for density based on prior knowledge. Sub-blocks with a density greater than the set upper threshold are further subdivided, sub-blocks with a density less than the set lower threshold are merged with adjacent sub-blocks, and the granularity of sub-blocks with a density not less than the set lower threshold and not greater than the set upper threshold remains unchanged.

[0023] Furthermore, the steps for calculating the span of carrier frequency, pulse width, and pulse angle of arrival values ​​for each cluster within a sub-block are as follows:

[0024] The first step is to partition the data in parallel within each sub-block, resulting in m clusters, which means dividing the sub-block into D clusters. ijk =[C1,C2,…C e ,…C m ], where C represents a cluster, C e This represents the e-th cluster after data partitioning;

[0025] The second step is to calculate the difference between the maximum and minimum values ​​of the carrier frequency, pulse width, and pulse angle of arrival for each cluster after dividing each sub-block, so as to obtain the span of the carrier frequency, pulse width, and pulse angle of arrival for that cluster within the sub-block.

[0026] Furthermore, the step of selecting the median value of the carrier frequency, pulse width, and pulse angle of arrival span of each sub-block to form the feature unit length of the pulse descriptor corresponding to that sub-block means selecting the median of the span of the carrier frequency, pulse width, and pulse angle of arrival of each of the m clusters in each sub-block as the feature unit length corresponding to the carrier frequency, pulse width, and pulse angle of arrival in that sub-block.

[0027] Furthermore, the comprehensive scaling factor refers to the scaling factor of the three pulse descriptors within a sub-block, obtained by dividing the feature unit length corresponding to the three pulse descriptors of each sub-block by the span of the corresponding pulse descriptor within that sub-block; and taking the cube root of the product of the scaling factors of the three pulse descriptors of each sub-block as the comprehensive scaling factor of the sub-block data.

[0028] Furthermore, the adaptive adjustment of the radius parameter and the maximum number of clusters parameter of the Birch clustering algorithm means that, for each sub-block, the set initial radius parameter is divided by the comprehensive scaling factor of the sub-block to obtain the radius parameter used by the Birch algorithm for that sub-block; and the set initial maximum number of clusters parameter is divided by the comprehensive scaling factor of the sub-block to obtain the maximum number of clusters parameter used by the Birch algorithm for that sub-block.

[0029] Furthermore, the use of adaptive parameters to perform Birch clustering in parallel on each pulse descriptor subspace means that the Birch clustering algorithm for each sub-block is performed in parallel using the adjusted adaptive Birch clustering algorithm radius parameter and the maximum number of clusters parameter corresponding to that sub-block.

[0030] Furthermore, the pseudo-label of the generated cluster refers to adding a non-negative integer label to the pulse data clustered into the same cluster within each sub-block, based on the results of the adaptive Birch clustering algorithm performed on each sub-block.

[0031] Furthermore, the steps for calculating the feature similarity of each result cluster are as follows:

[0032] The first step is to calculate the Pulse Repetition Interval (PRI) for each different cluster after adaptive Birch clustering, according to the following formula:

[0033] PRI r =TOA r+1 -TOA r

[0034] Among them, PRI r This represents the calculated r-th PRI, TOA rThis represents the pulse arrival time value corresponding to the r-th pulse of a cluster obtained after performing the Birch algorithm, where r ranges from [1, s-1] and s represents the number of pulses in the cluster;

[0035] The second step is to measure the similarity of PRI in different clusters by comparing the differences in the histograms of PRI in different clusters. If the difference in the histograms of PRI in two clusters is less than the set PRI difference threshold, it is determined that the radar pulses corresponding to the two clusters belong to the same radar radiation source. The two clusters are then merged, that is, the labels of the pulses in the two clusters are modified to the same value.

[0036] The third step is to calculate the mean values ​​of carrier frequency, pulse width, and pulse angle of arrival (PAA) of the intra-cluster data after PRI difference merging. The difference in the mean values ​​of the corresponding pulse descriptors of different clusters is calculated to measure the differences in carrier frequency, pulse width, and PAA of different clusters. If the difference in the mean values ​​of carrier frequency, pulse width, and PAA of two clusters is less than their respective set merging thresholds (carrier frequency difference threshold, pulse width difference threshold, and PAA difference threshold), then it is determined that the radar pulses corresponding to these two clusters belong to the same radar radiation source, and the two clusters are merged, that is, the labels of the pulses of these two clusters are modified to the same value.

[0037] Furthermore, the merging thresholds PRI difference threshold, carrier frequency difference threshold, pulse width difference threshold, and pulse angle of arrival difference threshold refer to non-negative values ​​that are limited based on prior knowledge of the numerical ranges of PRI, carrier frequency, and pulse width of the detected radar, and the pulse angle of arrival difference threshold is a four-decimal value limited to the range [0°, 4°].

[0038] Compared with the prior art, the present invention has the following advantages:

[0039] First, this invention performs grid partitioning on the pulse descriptor space. By partitioning the three-dimensional data space composed of carrier frequency, pulse width, and pulse angle of arrival into a grid, it can simultaneously process and cluster data within multiple data sub-modules, performing clustering tasks in parallel. This solves the problem in existing technologies where processing and sorting times increase with large amounts of data, making it impossible to quickly complete sorting tasks when there is a large amount of pulse signal data. This invention improves the parallelism and computation speed of the sorting algorithm, supporting the requirement for faster radar pulse signal sorting in complex electromagnetic environments and improving the efficiency of radar pulse signal sorting.

[0040] Secondly, this invention uses an adaptive Birch clustering algorithm. By calculating the scaling factor of the three-dimensional data space composed of carrier frequency, pulse width, and pulse arrival angle for different sub-blocks, the parameters of the Birch clustering algorithm for each sub-block are adjusted. This solves the problem that the accuracy of existing sorting methods is limited by the selection of parameters and cannot adapt to the measurement accuracy requirements of dense and complex electromagnetic environments. As a result, the radar pulse signal sorting accuracy of this invention is improved, which can support the radar pulse signal sorting accuracy requirements in complex electromagnetic environments and has a wider range of applications. Attached Figure Description

[0041] Figure 1 This is a flowchart of the present invention;

[0042] Figure 2 The figure shows the simulation results of this invention. Detailed Implementation

[0043] To more clearly illustrate the present invention, a further detailed description is provided below in conjunction with the embodiments and accompanying drawings. Obviously, the accompanying drawings and embodiments described below are merely some embodiments of the present invention, and not all embodiments. For those skilled in the art, other drawings can be obtained based on these drawings without any inventive effort. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0044] Reference Figure 1 The specific implementation steps of the present invention will be described in further detail below.

[0045] Step 1: Receive pulse description word data from the radar receiver and divide the data space composed of carrier frequency, pulse width, and pulse angle of arrival into a grid.

[0046] Step 1: The received pulse descriptor data consists of five pulse descriptors: pulse arrival time, carrier frequency, pulse width, pulse amplitude, and pulse arrival angle, as shown in Table 1.

[0047] Table 1. Overview of Pulse Descriptor Word Data Composition

[0048] Pulse arrival time carrier frequency Pulse width Pulse Amplitude Pulse arrival angle Pulse arrival time 1 Carrier frequency 1 Pulse width 1 Pulse amplitude 1 Pulse arrival angle 1 Pulse arrival time 2 Carrier frequency 2 Pulse width 2 Pulse amplitude 1 Pulse arrival angle 2 Pulse arrival time 3 Carrier frequency 3 Pulse width 3 Pulse amplitude 1 Pulse arrival angle 3 Pulse arrival time 4 Carrier frequency 4 Pulse width 4 Pulse amplitude 1 Pulse arrival angle 4

[0049] Step 2, according to the preset granularity N = [n 载频 ,n 脉宽 ,n 脉冲到达角 The three-dimensional feature space is divided into multiple sub-blocks, where n is the number of received pulses, and each sub-block corresponds to a three-dimensional mesh element. Its data set can be represented as follows:

[0050] Dijk ={x(1),x(2),x(3),…x(t)…x(n)}

[0051] Where x(t) represents the pulse descriptor set corresponding to the t-th pulse data point, and i, j, k correspond to the indices of the pulse descriptor carrier frequency, pulse width, and pulse angle of arrival, respectively. Then, the density of each sub-block is calculated according to the following formula:

[0052]

[0053] Where, ρ ijk |D represents the data density of a sub-block within the corresponding pulse descriptor word carrier frequency, pulse width, and pulse angle of arrival. ijk | represents the sum of the absolute values ​​of all elements within this sub-block, V ijk This indicates that the sub-block D ijk The volume of the sub-blocks is determined by the density of the sub-blocks. Sub-blocks with a density greater than the set upper limit threshold are further subdivided. Sub-blocks with a density less than the set lower limit threshold are merged with adjacent sub-blocks. Sub-blocks with a density not less than the set lower limit threshold and not greater than the set upper limit threshold have their granularity unchanged.

[0054] Step 2: Calculate the overall scaling factor for each sub-block in parallel.

[0055] Step 1: Perform parallel data partitioning within each sub-block defined in Step 1, resulting in m clusters, i.e., divide the sub-block into D clusters. ijk =[C1,C2,…C e …,C m ], where C represents a cluster, C e This represents the e-th cluster obtained after the initial partitioning.

[0056] Step 2: For each cluster within each sub-block, calculate the difference between the maximum and minimum values ​​of the carrier frequency, pulse width, and pulse angle of arrival for that cluster, and obtain the span of the carrier frequency, pulse width, and pulse angle of arrival for that cluster within that sub-block.

[0057] Step 3: Divide the feature unit length corresponding to the three pulse descriptors of each sub-block by the span of the corresponding pulse descriptor within that sub-block to obtain the scaling factor of the three pulse descriptors within that sub-block; take the cube root of the product of the scaling factors of the three pulse descriptors of each sub-block as the comprehensive scaling factor of the sub-block data.

[0058] Step 3: Perform adaptive Birch clustering algorithm in parallel for different sub-blocks.

[0059] Step 1: For each sub-block, divide the set initial radius parameter by the comprehensive scaling factor of the sub-block to obtain the radius parameter used by the Birch algorithm for the sub-block; divide the set initial maximum number of clusters parameter by the comprehensive scaling factor of the sub-block to obtain the maximum number of clusters parameter used by the Birch algorithm for the sub-block.

[0060] Step 2: Perform the Birch clustering algorithm on each sub-block in parallel using the adjusted adaptive Birch clustering algorithm radius parameter and the maximum number of clusters parameter corresponding to that sub-block. Based on the results of the adaptive Birch clustering algorithm on each sub-block, add non-negative integer labels to the pulse data that are clustered into the same cluster within that sub-block.

[0061] Step 4: Measure the similarity of different clusters and merge them to complete the sorting.

[0062] Step 1: For each cluster after adaptive Birch clustering in Step 3, calculate the PRI of that cluster according to the following formula:

[0063] PRI r =TOA r+1 -TOA r

[0064] Among them, PRI r This represents the calculated r-th PRI, TOA r This represents the pulse arrival time value corresponding to the r-th pulse of a cluster obtained after performing the Birch algorithm, where r ranges from [1, s-1] and s represents the number of pulses in the cluster.

[0065] Step 2: By comparing the differences in the histograms of PRI from different clusters, the similarity of PRI from different clusters is measured. If the difference in the histograms of PRI from two clusters is less than the set PRI difference threshold, it is determined that the radar pulses corresponding to these two clusters belong to the same radar radiation source. The two clusters are then merged, that is, the labels of the pulses from these two clusters are modified to the same value.

[0066] Step 3: Calculate the mean values ​​of carrier frequency, pulse width, and pulse angle of arrival (PAA) of the intra-cluster data after PRI difference merging. Calculate the difference in the mean values ​​of the corresponding pulse descriptors of different clusters to measure the differences in carrier frequency, pulse width, and PAA of different clusters. If the difference in the mean values ​​of carrier frequency, pulse width, and PAA of two clusters is less than their respective set merging thresholds (carrier frequency difference threshold, pulse width difference threshold, and PAA difference threshold), then it is determined that the radar pulses corresponding to these two clusters belong to the same radar radiation source, and the two clusters are merged, that is, the labels of the pulses of these two clusters are modified to the same value.

[0067] Step 4: Write the cluster labels of the sorted pulse data into the data file to complete the parallel adaptive clustering sorting based on pulse descriptors.

[0068] The effects of this invention will be further illustrated below with simulation experiments:

[0069] 1. Simulation experimental conditions.

[0070] The hardware platform for the simulation experiment of this invention is: Intel i5 9300H CPU with a main frequency of 2.4GHz and 16GB of memory.

[0071] The software platform for the simulation experiment of this invention is: Windows 10 operating system and PyCharm 2023.3.5.

[0072] The data files and settings used in the simulation experiment of this invention are as follows:

[0073] The file used in this simulation is a user-provided radar pulse descriptor data file, the contents of which are shown in Table 2.

[0074] Table 2. Parameters of Radar Pulse Descriptor Data File Provided by the User

[0075] Pulse arrival time carrier frequency Pulse width Pulse Amplitude Pulse arrival angle Label Pulse arrival time 1 Carrier frequency 1 Pulse width 1 Pulse amplitude 1 Pulse arrival angle 1 Tag 1 Pulse arrival time 2 Carrier frequency 2 Pulse width 2 Pulse amplitude 1 Pulse arrival angle 2 Tag 2 Pulse arrival time 3 Carrier frequency 3 Pulse width 3 Pulse amplitude 1 Pulse arrival angle 3 Tag 3 Pulse arrival time 4 Carrier frequency 4 Pulse width 4 Pulse amplitude 1 Pulse arrival angle 4 Tag 4

[0076] In the simulation experiment of this invention, the initial radius parameter was set to 0.5, and the initial maximum number of clusters was set to 12. In the comparative experiment, the neighborhood radius parameter of the DBSCAN algorithm was set to 0.6, and the minimum number of points within the neighborhood parameter was set to 150.

[0077] 2. Simulation content and result analysis.

[0078] The simulation experiment of this invention uses the method and DBSCAN algorithm of this invention to conduct radar pulse signal sorting experiments based on the pulse descriptor data file provided by the user. The simulation results are as follows: Figure 2 As shown.

[0079] The following is combined with Figure 2 The resulting figures further illustrate the effects of the present invention.

[0080] Figure 2 (a) A raw, unsorted 3D data map drawn based on partial data of carrier frequency, pulse width, and pulse angle of arrival from a radar pulse descriptor file provided by the user. Figure 2 In (a), "3D Scatter plot of RF,PW,DOAData" means Figure 2(a) is a three-dimensional data plot drawn based on carrier frequency, pulse width, and pulse angle of arrival data. The RF axis represents the numerical range of carrier frequency in the pulse descriptor word, the PW axis represents the numerical range of pulse width in the pulse descriptor word, and the DOA axis represents the numerical range of pulse angle of arrival in the pulse descriptor word.

[0081] Figure 2 (b) A raw, unsorted 3D data plot drawn based on partial data of the carrier frequency, pulse width, and pulse angle of arrival from the radar pulse descriptor file provided by the user, along with their corresponding original labels. Figure 2 (b) "3DScatter Plotby Label" indicates Figure 2 (b) is a three-dimensional plot of raw data clusters drawn based on carrier frequency, pulse width, pulse angle of arrival data and corresponding raw labels. The RF axis represents the numerical range of carrier frequency in the pulse descriptor, the PW axis represents the numerical range of pulse width in the pulse descriptor, and the DOA axis represents the numerical range of pulse angle of arrival in the pulse descriptor. Figure 2 In (b), the different colors and adjacent numbers in the legend correspond to three-dimensional radar pulse descriptor data with the same original label.

[0082] Figure 2 (c) is a three-dimensional data map of the cluster distribution of carrier frequency, pulse width, and pulse angle of arrival after sorting by the method of the present invention based on partial data of carrier frequency, pulse width, and pulse angle of arrival from the radar pulse descriptor file provided by the user. Figure 2 (c) "3D Scatter Plot by pred" means Figure 2 (c) is a three-dimensional diagram of the predicted sorting cluster drawn based on the carrier frequency, pulse width, pulse angle of arrival data and the corresponding predicted tags after sorting using the method of the present invention. The RF axis represents the numerical range of the carrier frequency in the pulse descriptor, the PW axis represents the numerical range of the pulse width in the pulse descriptor, and the DOA axis represents the numerical range of the pulse angle of arrival in the pulse descriptor. Figure 2 In (c), the different colors and the numbers next to them in the legend correspond to three-dimensional radar pulse descriptor data with the same sorting prediction label.

[0083] Figure 2 (d) is a three-dimensional data map of the cluster distribution of carrier frequency, pulse width, and pulse angle of arrival based on partial data of the radar pulse descriptor file provided by the user, after applying the DBSCAN clustering algorithm. Figure 2 (d) "3D Scatter Plot by DBSCAN" means Figure 2(d) is a 3D diagram of DBSCAN predicted sorting clusters drawn based on carrier frequency, pulse width, pulse angle of arrival data and corresponding DBSCAN predicted labels after sorting using the DBSCAN clustering algorithm. The RF axis represents the numerical range of carrier frequency in the pulse descriptor, the PW axis represents the numerical range of pulse width in the pulse descriptor, and the DOA axis represents the numerical range of pulse angle of arrival in the pulse descriptor. Figure 2 In (d), the different colors and the numbers next to them in the legend correspond to the three-dimensional radar pulse descriptor data with the same DBSCAN sorting prediction label.

[0084] Figure 2 (e) is a graph showing the results of data partitioning and corresponding subspace data sorting accuracy based on a complete dataset of pulse descriptors provided by the user, using the method of the present invention, where Accuracy represents the accuracy after sorting the corresponding pulse descriptor data subspace.

[0085] from Figure 2 (a) Figure 2 As can be seen in (b), the radar pulse descriptor dataset provided by the user has severe numerical aliasing, making it easy to classify radar pulse descriptor data belonging to different radars into the same radar data cluster, which is quite difficult. This indicates that the electromagnetic environment is currently severely aliased and complex.

[0086] from Figure 2 (c) Figure 2 As can be seen in (d), under the same radar pulse descriptor carrier frequency, pulse width, and pulse angle of arrival, the method of the present invention can distinguish pulse descriptor data that do not belong to the same type of radar compared with the DBSCAN method of the prior art. This indicates that the method of the present invention has improved the performance of radar pulse descriptor sorting compared with the prior art.

[0087] from Figure 2 As can be seen in (e), the method of the present invention can divide the pulse descriptor data into multiple data subspaces for sorting on a complete dataset of pulse descriptors provided by the user, and the sorting accuracy is high. This indicates that the method of the present invention has improved the performance of radar pulse descriptor sorting compared with existing methods.

[0088] The simulation results above show that, under the condition that the received radar pulse descriptor data is the same, the method of the present invention can determine the pulse descriptor data belonging to different radars, and the sorting accuracy of each pulse descriptor data subspace is also high. It is undoubtedly a radar pulse descriptor sorting method with higher sorting efficiency and accuracy.

Claims

1. A radar pulse signal sorting method of parallel adaptive clustering, characterized by, The three-dimensional data space composed of carrier frequency, pulse width and pulse arrival angle is meshed, and the scaling factor of the three-dimensional data space composed of carrier frequency, pulse width and pulse arrival angle is calculated for different sub-blocks, and the parameters of Birch clustering algorithm for each sub-block are adjusted; the steps of the sorting method include the following: Step 1, meshing based on the three-dimensional feature space composed of carrier frequency, pulse width and pulse arrival angle, dividing the original pulse description word feature space into multiple pulse description word subspaces; Step 2, dividing the pulse description word subspace in parallel to obtain clusters, and calculating the span of the carrier frequency, pulse width and pulse arrival angle values of each cluster in the sub-block in parallel; Step 3, taking the median value of the carrier frequency, pulse width and pulse arrival angle span in each sub-block as the feature unit length of the corresponding pulse description word in the sub-block in parallel; Step 4, using the comprehensive scaling factor of the pulse description word subspace, adaptively adjusting the radius parameter and the maximum cluster number parameter of the Birch clustering algorithm; using the adaptive parameters, performing the Birch clustering algorithm on each pulse description word subspace in parallel, and generating the pseudo-labels of the result clusters; Step 5, measuring the feature similarity of each result cluster after sorting, unifying the corresponding pseudo-labels of the clusters that meet the merging conditions to the same label value, realizing the merging of the result clusters, and completing the sorting of the radar pulse signal.

2. The parallel adaptive clustering radar pulse signal sorting method according to claim 1, wherein, The step of meshing based on the three-dimensional feature space composed of carrier frequency, pulse width and pulse arrival angle in step 1 is as follows: First step, receiving pulse description word data composed of five pulse description words of pulse arrival time, carrier frequency, pulse width, pulse amplitude and pulse arrival angle; Secondly, the three-dimensional feature space is divided into multiple sub-blocks according to a preset granularity N=[n 载频 ,n 脉宽 ,n 脉冲到达角 ], each sub-block corresponds to a three-dimensional grid unit, and the data set of each sub-block can be expressed as follows: D ijk = {x(1), x(2), x(3),... x(t),... x(n)}, where D ijk represents the divided pulse descriptor word sub-block data space, i, j, k represent the indexes of carrier frequency, pulse width, and pulse arrival angle, respectively; n represents the number of received pulses, and x(t) represents the pulse descriptor word set corresponding to the tth pulse data point. Third step, calculating the density of each sub-block according to the following formula: ， where ρ ijk represents the data density of a sub-block, |D ijk | represents the sum of absolute values of each element within the sub-block, V ijk represents the volume of the sub-block data space D ijk of the pulse descriptor word. Fourth step, setting the upper threshold and lower threshold of the density according to prior knowledge, further subdividing the sub-blocks with a density greater than the set upper threshold of the density, merging the sub-blocks with a density less than the set lower threshold of the density with adjacent sub-blocks, and keeping the granularity of the sub-blocks with a density not less than the set lower threshold of the density and not greater than the upper threshold of the density unchanged.

3. The parallel adaptive clustering radar pulse signal sorting method according to claim 1, wherein, The step of calculating the span of the carrier frequency, pulse width and pulse arrival angle values of each cluster in the sub-block in step 2 is as follows: Firstly, data partition is performed in each sub-block in parallel to obtain m clusters, i.e. the sub-block is divided into D ijk = [C1,C2,…C e ,…C m ], wherein C represents a cluster, and C e e represents the e-th cluster after data partition; Second step, for each cluster divided in each sub-block, the difference between the maximum value and the minimum value of the carrier frequency, pulse width and pulse arrival angle of the cluster is calculated respectively, and the span of the carrier frequency, pulse width and pulse arrival angle of the cluster in the sub-block is obtained.

4. The parallel adaptive clustering radar pulse signal sorting method according to claim 1, wherein, In step 3, the median value of the carrier frequency, pulse width and pulse arrival angle span of each sub-block is selected to form the feature unit length of the corresponding pulse description word of the sub-block, which means that according to the span of the carrier frequency, pulse width and pulse arrival angle of the m clusters in each sub-block, the median value of the span of the carrier frequency, pulse width and pulse arrival angle is selected as the feature unit length corresponding to the carrier frequency, pulse width and pulse arrival angle of the sub-block.

5. The parallel adaptive clustering radar pulse signal sorting method according to claim 1, wherein, The comprehensive scaling factor in step 4 refers to dividing the characteristic unit length corresponding to the three pulse description words of each sub-block by the span of the corresponding pulse description word in the sub-block to obtain the scaling factor of the three pulse description words in the sub-block; and taking the cube root of the product of the scaling factors of the three pulse description words of each sub-block as the comprehensive scaling factor of the data of the sub-block.

6. The parallel adaptive clustering radar pulse signal sorting method according to claim 1, wherein, The adaptive adjustment of the radius parameter and the maximum cluster number parameter of the Birch clustering algorithm in step 4 refers to dividing the initial radius parameter by the comprehensive scaling factor of each sub-block to obtain the radius parameter used by the sub-block in the Birch algorithm; and dividing the initial maximum cluster number parameter by the comprehensive scaling factor of the sub-block to obtain the maximum cluster number parameter used by the sub-block in the Birch algorithm.

7. The parallel adaptive clustering radar pulse signal sorting method according to claim 1, wherein, The parallel Birch clustering algorithm using adaptive parameters for each pulse description word subspace in step 4 refers to using the adjusted adaptive Birch clustering algorithm radius parameter and the maximum cluster number parameter corresponding to each sub-block to perform the Birch clustering algorithm for the sub-block.

8. The parallel adaptive clustering radar pulse signal sorting method according to claim 1, wherein, The pseudo-label of the result cluster in step 4 refers to adding a non-negative integer label to the pulse data in the same cluster in each sub-block according to the result of the adaptive Birch clustering algorithm for the sub-block.

9. The parallel adaptive clustering radar pulse signal sorting method according to claim 1, wherein, The step of calculating the feature similarity of each result cluster in step 5 is as follows: First, calculate the pulse repetition interval (PRI) of each adaptive Birch clustering cluster according to the following formula: ， wherein PRI r represents the calculated rthPRI, TOA r represents the pulse arrival time value corresponding to the rthpulse of a cluster obtained after performing the Birch algorithm, r ranges from [1, s-1], and s represents the number of pulses in the cluster; Second, measure the PRI similarity of different clusters by comparing the differences in the PRI histogram of different clusters. If the difference in the PRI histogram of two clusters is less than the set PRI difference threshold, it is determined that the radar pulses corresponding to the two clusters belong to the same radar emitter, and the two clusters are merged, i.e., the labels of the pulses in the two clusters are modified to the same value. Third, calculate the mean values of the carrier frequency, pulse width, and pulse arrival angle of the data in the cluster after PRI difference merging, and calculate the differences between the mean values of the corresponding pulse description words of different clusters to measure the differences in the carrier frequency, pulse width, and pulse arrival angle of different clusters. If the differences between the mean values of the carrier frequency, pulse width, and pulse arrival angle of two clusters are all less than the respective set merging threshold carrier frequency difference threshold, pulse width difference threshold, and pulse arrival angle difference threshold, it is determined that the radar pulses corresponding to the two clusters belong to the same radar emitter, and the two clusters are merged, i.e., the labels of the pulses in the two clusters are modified to the same value.

10. The parallel adaptive clustering radar pulse signal sorting method of claim 9, wherein, The PRI difference threshold, carrier frequency difference threshold, and pulse width difference threshold are non-negative values defined according to the prior knowledge of the value range of the PRI, carrier frequency, and pulse width of the detected radar. The pulse arrival angle difference threshold is a four-digit decimal value limited in the range [0°, 4°].

Images