Unknown radar radiation source sorting method based on time sequence feature clustering

Abstract

The invention discloses an unknown radar radiation source sorting method based on time sequence feature clustering. The unknown radar radiation source sorting method comprises the following steps: acquiring radar signal pulse description words of pulse signals; according to the radar signal pulse description word, performing spatial clustering on the pulse signal to obtain a spatial clustering result; for each spatial clustering result, according to the radar signal pulse description word of the pulse signal in each spatial clustering result, performing time feature clustering on the pulse signal in each spatial clustering result to obtain a time sequence clustering result; and analyzing a time parallel relationship and a time continuous relationship of the time sequence clustering results, and performing pulse group sequence blending on the time sequence clustering results to obtain a sorting result. On the basis of the existing clustering algorithm, the information of the pulse signal in the time dimension is introduced, the time sequence feature clustering of the pulse signal is realized by using the ST-DBSCAN algorithm thought and introducing the TOA-PA constraint interval, and the method has strong robustness for the complex electromagnetic environment.

Description

Technical Field

[0001] The present invention relates to the technical field of radar radiation source signal sorting, and more particularly to a method for sorting unknown radar radiation sources based on time series feature clustering. Background Art

[0002] Electronic warfare is primarily divided into three parts: electronic reconnaissance, electronic defense, and electronic jamming. The primary task of electronic reconnaissance is to detect and intercept electromagnetic radiation signals from enemy electronic equipment, then process and analyze them to determine information such as the type, purpose, and threat posed by the enemy's radiation source. This not only provides intelligence for operational planning but also supports electronic jamming and defense efforts. Therefore, electronic reconnaissance is fundamental and crucial to electronic warfare. Radar emitter sorting and identification are key components of the reconnaissance process. Using the radar signal pulse descriptor word (PDW) obtained by the parameter measurement module, the interleaved data stream of continuously arriving radar emitter signals is decomposed into multiple sequences of individual radar emitter signals. Radar emitter identification, including model identification, individual identification, and operational status identification, is then performed, providing critical intelligence support for subsequent situational assessment and strategic deployment.

[0003] During reconnaissance system analysis, radar signals are primarily characterized by PDW, including time of arrival (TOA), angle of arrival (DOA), carrier frequency (RF), pulse width (PW), pulse amplitude (PA), and pulse repetition interval (PRI). Consequently, domestic and international scholars have begun conducting in-depth research on radar emitter sorting methods based on PDW. Classic radar emitter sorting methods based on PDW can be divided into two main categories: PRI-based and multi-parameter-based.

[0004] Radar emitter sorting based on the PRI (Primary Pricing) (PRI) is the most widely used sorting method in radar reconnaissance systems. The first-order difference (DTOA) of the pulse time-of-arrival (TOA) sequence emitted by a single radar typically exhibits regularity. Possible PRI values ​​are identified from the DTOA sequence within the interleaved pulse sequence, and pulse sequence searches are performed using these PRI values ​​to sort radar emitters. These methods primarily include the PRI histogram method and the PRI transformation method. The traditional PRI histogram method calculates multiple TOA differences of the interleaved radar emitter signals, accumulates and statistically analyzes them, and uses PRIs exceeding a threshold as the target radar's PRI. Pulses are then searched for and sorted.

[0005] Due to the increasing complexity of the electromagnetic environment and the advancement of parameter measurement technology, radar emitter sorting is no longer limited to a single PRI parameter; multi-parameter sorting methods are beginning to gain traction. Because pulse signals emitted by the same radar emitter are naturally similar, while the similarity between pulse signals from different radar emitters is generally lower than that between pulse signals from the same emitter, sorting methods based on unsupervised multi-parameter clustering have garnered widespread attention and research.

[0006] PRI-based methods are prone to missing target pulse sequences when faced with high loss rates or noisy pulses, resulting in decreased pulse utilization and "over-batch" phenomena. Classic multi-parameter clustering and sorting methods rely primarily on parameters such as DOA, RF, and PW in the PDW, sorting pulses based on the similarity of their parameters in the spatial dimension. However, pulse signals are received by reconnaissance systems as pulse streams and also contain rich feature information in the temporal dimension. Existing methods are unable to effectively utilize the temporal information of the pulse sequence. Summary of the Invention

[0007] The present invention provides a method for sorting unknown radar radiation sources based on time series feature clustering, which solves the problem that the existing classical sorting method based on multi-parameter clustering cannot effectively utilize the time information of pulse sequences.

[0008] In order to solve the above technical problems, the technical solutions of the present invention are as follows:

[0009] The present invention provides a method for sorting unknown radar radiation sources based on time series feature clustering, comprising the following steps:

[0010] Obtaining the radar signal pulse description word of the pulse signal;

[0011] performing spatial clustering on the pulse signal according to the radar signal pulse description word to obtain a spatial clustering result;

[0012] For each spatial clustering result, according to the radar signal pulse description word of the pulse signal in each spatial clustering result, the pulse signal in each spatial clustering result is clustered by time characteristics to obtain a time series clustering result;

[0013] The time parallel relationship and time continuous relationship of the time series clustering results are analyzed, and the pulse group sequences of the time series clustering results are batched to obtain the sorting results.

[0014] Among the above technical means, combined with the idea of ​​spatiotemporal clustering (ST-DBSCAN), the time series feature clustering algorithm is divided into two stages: spatial feature clustering and time feature clustering. The pulse time dimension information such as TOA and PA is combined with the classic multi-parameter clustering algorithm to retain the temporal nature of the pulse characteristic parameters, realize the time segment clustering of the pulse data, and complete the "batching" of the sorting results by analyzing the temporal relationship of the pulse clustering results and combining the directed graph method, thereby improving the sorting accuracy in the actual electromagnetic environment.

[0015] Furthermore, the radar signal pulse description words include arrival time TOA, arrival angle DOA, carrier frequency RF, pulse width PW, pulse amplitude PA and pulse repetition interval PRI.

[0016] Furthermore, spatial clustering is performed on the pulse signal according to the radar signal pulse description word to obtain a spatial clustering result, including:

[0017] According to the pulse signal, setting a DOA-RF-PW three-dimensional grid of a first preset scale, performing grid density clustering on the pulse signal to obtain a first clustering result and noise pulses, merging the first clustering results, and rearranging the pulses in chronological order to obtain a denoised interleaved pulse sequence;

[0018] Setting a DOA-RF-PW three-dimensional grid of a second preset scale according to the denoised interleaved pulse sequence, and calculating a grid density threshold according to the number of grids and grid density of the DOA-RF-PW three-dimensional grid of the second preset scale;

[0019] Obtaining a high-density grid and a low-density grid according to a relationship between each grid density of the DOA-RF-PW three-dimensional grid of the second preset scale and the grid density threshold;

[0020] The adjacent high-density grids are merged and subjected to boundary optimization processing, and the low-density grids adjacent to the high-density grids are merged at the same time to obtain a spatial clustering result.

[0021] Furthermore, calculating a grid density threshold according to the number of grids and the grid density of the DOA-RF-PW three-dimensional grid of the second preset scale includes:

[0022]

[0023] In the formula, MinPts grid is the grid density threshold, D={den1,den2,…,den i ,…,den N} is a sequence of all grid densities, den iis the i-th grid density, N is the number of grids, 1≤i≤N.

[0024] Furthermore, according to the radar signal pulse description word of the pulse signal in each spatial clustering result, the pulse signal in each spatial clustering result is clustered by time characteristics to obtain a time series clustering result, including:

[0025] Calculate the first-order difference of the arrival time TOA of the pulse signal in the current spatial clustering result to obtain the DTOA sequence;

[0026] Calculating the TOA constraint interval of the current spatial clustering result according to the DTOA sequence;

[0027] According to the two pulse signals with the largest and smallest pulse amplitudes PA in the current spatial clustering result, the arrival time TOA difference and the pulse amplitude PA difference of the two pulse signals are obtained;

[0028] Calculating the PA constraint interval of the current spatial clustering result according to the arrival time TOA difference and the pulse amplitude PA difference;

[0029] According to the TOA constraint interval and the PA constraint interval, time feature clustering is performed on the current spatial clustering result to obtain a time clustering result.

[0030] Furthermore, the TOA constraint interval of the current spatial clustering result is calculated according to the DTOA sequence, including:

[0031] Perform histogram statistics on the DTOA sequence and set the statistical threshold to a preset percentage of the total number of pulse signals in the current spatial clustering result. If there is a DTOA value whose statistical value exceeds the statistical threshold, the maximum value among the DTOA values ​​exceeding the statistical threshold is selected and recorded as T s If there is no DTOA value whose statistical value exceeds the statistical threshold, the maximum value among the DTOA values ​​is selected and recorded as T s ;

[0032] Then the TOA constraint interval T of the current spatial clustering result is r for:

[0033] T r =βT s

[0034] Where β is a hyperparameter, which is an empirical value set according to the pulse loss rate in the real environment.

[0035] Furthermore, the PA constraint interval of the current spatial clustering result is calculated based on the arrival time TOA difference and the pulse amplitude PA difference, including:

[0036]

[0037] pa r =Δpa / N T

[0038] Where N T is an intermediate parameter, Δdtoa represents the absolute value of the arrival time TOA difference, Indicates taking the upper limit, pa r It represents the PA constraint interval of the current spatial clustering result, and Δpa is the absolute value of the pulse amplitude PA difference.

[0039] Furthermore, the time parallel relationship and time continuous relationship of the time series clustering results are analyzed, and the pulse group sequences of the time series clustering results are batched, including batching the pulse group sequences of the time series clustering results according to the parallel relationship between the pulse groups:

[0040] Record the arrival time TOA of the first pulse and the arrival time TOA + pulse width PW of the last pulse of each pulse group in the time clustering result as the start time and end time of the pulse group, respectively, and arrange the pulse groups in ascending order according to the start time;

[0041] Analyze and judge the parallel relationship between pulse groups:

[0042] The maximum DTOA value in each pulse group is counted in turn, and the average value is taken as the time difference threshold Treshold1. The start time difference and end time difference between each pulse group and other pulse groups are calculated in turn. If a pulse group pair with a start time difference and an end time difference lower than Treshold1 is found, the next step of judgment is carried out;

[0043] Set the DOA threshold Tresholddoa, record the DOA range and DOA average of the two pulse groups respectively. If the DOA range of one pulse group is completely covered by the DOA range of the other pulse group, or the difference between the DOA averages of the two pulse groups is lower than Tresholddoa, proceed to the next step of judgment.

[0044] Set the PA threshold Tresholdpa and record the PA ranges of the two pulse groups respectively. If the difference between the center values ​​of the PA ranges of the two pulse groups is lower than Tresholdpa, proceed to the next step of judgment.

[0045] Calculate the PRI entropy or duty cycle entropy of the two pulse groups separately. After merging the two pulse groups, rearrange them according to the pulse time sequence, and calculate the PRI entropy or duty cycle entropy of the merged pulse group. If the PRI entropy or duty cycle entropy of the merged pulse group is not greater than the sum of the PRI entropy or duty cycle entropy of the two pulse groups before merging, the two pulse groups meet the parallel relationship.

[0046] Merge the pulse groups that meet the parallel relationship.

[0047] Furthermore, the time parallel relationship and time continuous relationship of the time series clustering results are analyzed, and the pulse group sequences of the time series clustering results are batched, including batching the pulse groups after merging the pulse groups that meet the parallel relationship according to the time continuous relationship between the pulse groups, including the first batching process and the second batching process:

[0048] Recalculate the start time and end time of each pulse group, and sort the pulse groups in ascending order according to the start time;

[0049] The first batching process based on the continuous relationship includes:

[0050] 1) Count the duration of each pulse group in turn, take the average value as the time difference threshold Treshold2, and calculate the difference between the end time of each pulse group and the start time of other subsequent pulse groups in chronological order. If a pulse group pair with a time difference less than Treshold2 is found, proceed to the next step of judgment;

[0051] 2) According to the starting time sequence of the two pulse groups, they are recorded as the front pulse group and the back pulse group respectively. The last 32 pulses of the front pulse group are taken as pulse group A; at the same time, the first 32 pulses of the back pulse group are taken as pulse group B. If the number of pulses in the pulse group is less than 32, all pulses of the pulse group are extracted; the DOA average values ​​of pulse group A and pulse group B are calculated respectively. If the difference between the DOA average values ​​of the two pulse groups is lower than Tresholddoa, the next step of judgment is carried out;

[0052] 3) Calculate the PA range of pulse group A and pulse group B respectively when one of the following conditions is met:

[0053] (a) The PA ranges of the two pulse groups do not overlap, and the PA difference between the last pulse of pulse group A and the first pulse of pulse group B is less than Tresholdpa;

[0054] (b) The PA range of one pulse group is covered by the PA range of another pulse group;

[0055] (c) The difference between the center values ​​of the PA range of the two pulse groups is less than Tresholdpa;

[0056] It is considered that the preceding pulse group and the following pulse group corresponding to pulse group A and pulse group B have a continuous relationship, and the next step of judgment is carried out;

[0057] 4) After determining the continuity relationship between all pulse groups, each pulse group is regarded as a node, and the pulse groups with which it has a continuity relationship are connected backward in chronological order to form a continuous relationship directed graph. The longest path in the directed graph is found, the pulse groups on the path are batched, the path is deleted from the directed graph, and the longest path in the directed graph is found again until the length of the longest path found is lower than the length threshold;

[0058] The second batching process based on the continuous relationship includes:

[0059] Perform continuous relationship analysis again on the pulse group obtained from the first batch process. The processes of steps 1), 2), and 4) remain the same. In step 3), change the judgment condition to:

[0060] The PA ranges of the two pulse groups do not overlap, and the average values ​​of the first-order differences of the PA sequences of the two pulse groups are both less than 0.5dBm;

[0061] Get the sorting results.

[0062] Furthermore, when performing time feature clustering, pulse signals that do not form a clustering result are regarded as isolated points, and the isolated points are judged and classified, including:

[0063] Calculate the start and end time of all sorting results, select an isolated point, and if the isolated point is within the time interval of a clustering result, proceed to the next step of judgment;

[0064] According to the TOA value of the isolated point, the last 50 pulses with TOA values ​​less than the isolated point TOA and the first 50 pulses with TOA values ​​greater than the isolated point TOA are selected from the eligible sorting results. If the number of pulses before or after the isolated point is less than 50, the actual number of pulses is extracted. These pulses are called reference points.

[0065] Calculate the confidence of each characteristic parameter of isolated points and reference points respectively:

[0066]

[0067] Where x k is the kth characteristic parameter value of the isolated point, g k , Δg k , σ k are the mean value, allowable deviation and mean square error of the kth characteristic parameter of the reference point, m is the number of characteristic parameters involved in the calculation, w k is the calculation weight of the mth feature parameter, and F(x) is the confidence that the isolated point belongs to the current clustering result;

[0068] Set the confidence threshold Thresholddoc, calculate the confidence of the isolated point and the reference point of the sorting result that meets the time relationship, and select the maximum value. If the maximum confidence is greater than Thresholddoc, the isolated point is considered to be assigned to the corresponding sorting result; if there are multiple maximum confidences, the sorting result with the largest number of pulses is selected for classification; if the maximum confidence does not exceed the threshold, the isolated point is considered to be a noise point;

[0069] The above steps are performed on all isolated points in sequence until all isolated points are assigned to corresponding sorting results or are regarded as noise points.

[0070] Compared with the prior art, the technical solution of the present invention has the following beneficial effects:

[0071] 1. Based on the existing clustering algorithm, the present invention introduces the information of pulse signals in the time dimension, uses the ST-DBSCAN algorithm idea, and introduces the TOA-PA constraint interval to realize the clustering of the timing characteristics of pulse signals, which has strong robustness to complex electromagnetic environments.

[0072] 2. Secondly, in order to solve the problem of too high batch rate of time series feature clustering sorting results, two basic pulse group time series relationships, parallel relationship and continuous relationship, are introduced. A method for batching time series feature clustering results based on time series relationship is proposed to effectively solve the problem of too high batch rate of sorting results.

[0073] 3. Finally, confidence calculation is used to determine and classify isolated points, effectively improving pulse utilization and sorting accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] Figure 1 A schematic diagram of a flow chart of a method for sorting unknown radar emitters based on time series feature clustering provided by an embodiment of the present invention;

[0075] Figure 2 A schematic diagram of a framework of a method for sorting unknown radar emitters based on time series feature clustering provided by an embodiment of the present invention;

[0076] Figure 3 A schematic diagram of the process of spatial feature clustering provided by an embodiment of the present invention;

[0077] Figure 4 A schematic diagram of the process of spatial feature clustering + temporal feature clustering provided by an embodiment of the present invention;

[0078] Figure 5 A flow chart of a batching method based on the timing relationship of pulse groups provided in an embodiment of the present invention;

[0079] Figure 6A directed graph constructed by a pulse group provided in an embodiment of the present invention;

[0080] Figure 7 A schematic diagram of the process of processing isolated points provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0081] The accompanying drawings are for illustrative purposes only and are not to be construed as limiting this patent;

[0082] In order to better illustrate this embodiment, some parts in the drawings may be omitted, enlarged, or reduced, and do not represent the actual product size;

[0083] It is understandable to those skilled in the art that some well-known structures and descriptions thereof may be omitted in the drawings.

[0084] The technical solution of the present invention is further described below with reference to the accompanying drawings and embodiments.

[0085] Example 1

[0086] The embodiment of the present invention provides a method for sorting unknown radar radiation sources based on time series feature clustering, such as Figure 1 and Figure 2 As shown, the following steps are included:

[0087] Obtaining the radar signal pulse description word of the pulse signal;

[0088] performing spatial clustering on the pulse signal according to the radar signal pulse description word to obtain a spatial clustering result;

[0089] For each spatial clustering result, according to the radar signal pulse description word of the pulse signal in each spatial clustering result, the pulse signal in each spatial clustering result is clustered by time characteristics to obtain a time series clustering result;

[0090] The time parallel relationship and time continuous relationship of the time series clustering results are analyzed, and the pulse group sequences of the time series clustering results are batched to obtain the sorting results.

[0091] In this embodiment, in order to address the problem that the classical sorting method based on multi-parameter clustering cannot effectively utilize the time information of the pulse sequence, the idea of ​​spatiotemporal clustering is combined to divide the time series feature clustering algorithm into two stages: spatial feature clustering and time feature clustering. The time series relationship of the pulse clustering results is analyzed to complete the "batching" of the sorting results, thereby significantly improving the accuracy of radar radiation source sorting.

[0092] Example 2

[0093] This embodiment further illustrates spatial feature clustering based on the first embodiment.

[0094] In a further embodiment, the radar signal pulse description words include arrival time TOA, arrival angle DOA, carrier frequency RF, pulse width PW, pulse amplitude PA and pulse repetition interval PRI.

[0095] In a further embodiment, the pulse signal is spatially clustered according to the radar signal pulse description word to obtain a spatial clustering result, such as Figure 3 Shown, including:

[0096] The first stage is the denoising phase. Since the received unknown signal may contain noise pulses, the parameters of these noise pulses are usually widely distributed and relatively dispersed. Therefore, a DOA-RF-PW three-dimensional grid of a first preset scale can be set based on the pulse signal, and the pulse signal can be clustered by grid density to obtain a first clustering result and noise pulses. Pulses determined to be noise are discarded to separate most of the noise pulses. The first clustering results are merged and the pulses are rearranged in chronological order to form a denoised interleaved pulse sequence. In this embodiment, the first preset scales of the DOA, RF, and PW grids are set to 5°, 200MHz, and 5μs, respectively.

[0097] In the grid density clustering algorithm, it is necessary to divide the space for each parameter dimension to form a grid, set the grid density threshold to determine the grid type, divide each dimension into equal lengths to form a grid space, and map the pulse signal data to the grid space. Then, it is necessary to set the grid density threshold MinPts grid Determine the type of grid and set MinPts based on the number of grids and the average grid density. grid Therefore, based on the de-noised interleaved pulse sequence, a DOA-RF-PW three-dimensional grid of a second preset scale is set, and a grid density threshold is calculated based on the number of grids and grid density of the DOA-RF-PW three-dimensional grid of the second preset scale. In this embodiment, the second preset scales of DOA, RF, and PW are 2°, 20 MHz, and 2 μs.

[0098] According to the relationship between the grid densities of the DOA-RF-PW three-dimensional grid of the second preset scale and the grid density threshold, a grid with a density greater than the set grid density threshold is recorded as a high-density grid, and a grid with a density less than the set grid density threshold is recorded as a low-density grid;

[0099] The adjacent high-density grids are merged and subjected to boundary optimization processing. At the same time, the low-density grids adjacent to the high-density grids are merged to obtain spatial clustering results. Pulses that do not form clustering results are considered to be isolated points.

[0100] In a further embodiment, calculating the grid density threshold according to the number of grids and the grid density of the DOA-RF-PW three-dimensional grid of the second preset scale includes:

[0101]

[0102] In the formula, MinPts grid is the grid density threshold, D={den1,den2,…,den i ,…,den N} is a sequence of all grid densities, den i is the i-th grid density, N is the number of grids, 1≤i≤N.

[0103] Example 3

[0104] This embodiment further illustrates time feature clustering based on Embodiment 1 and Embodiment 2.

[0105] In this embodiment, after performing grid density clustering on the spatial features of the pulse signal, a preliminary clustering result will be obtained. In the time feature clustering process, the TOA and PA of the pulse signal are used to set the corresponding constraint interval, and the time feature clustering is performed on the results of the spatial feature clustering to achieve time segmentation of the clustering results. The specific method is to perform time feature clustering on the pulse signal in each spatial clustering result according to the radar signal pulse description word of the pulse signal in each spatial clustering result, and obtain the time series clustering result, such as Figure 4 Shown, including:

[0106] Calculate the first-order difference of the arrival time TOA of the pulse signal in the current spatial clustering result to obtain the DTOA sequence;

[0107] Calculating the TOA constraint interval of the current spatial clustering result according to the DTOA sequence;

[0108] According to the two pulse signals with the largest and smallest pulse amplitudes PA in the current spatial clustering result, the arrival time TOA difference and the pulse amplitude PA difference of the two pulse signals are obtained;

[0109] Calculating the PA constraint interval of the current spatial clustering result according to the arrival time TOA difference and the pulse amplitude PA difference;

[0110] According to the TOA constraint interval and the PA constraint interval, time feature clustering is performed on the current spatial clustering result to obtain a time clustering result.

[0111] In a further embodiment, calculating the TOA constraint interval of the current spatial clustering result according to the DTOA sequence includes:

[0112] Perform histogram statistics on the DTOA sequence and set the statistical threshold to a preset percentage of the total number of pulse signals in the current spatial clustering result. If there is a DTOA value whose statistical value exceeds the statistical threshold, the maximum value among the DTOA values ​​exceeding the statistical threshold is selected and recorded as T s If there is no DTOA value whose statistical value exceeds the statistical threshold, the maximum value among the DTOA values ​​is selected and recorded as T s ;

[0113] Then the TOA constraint interval T of the current spatial clustering result is r for:

[0114] T r =βT s

[0115] Where β is a hyperparameter, which is an empirical value set according to the pulse loss rate in the real environment. In this embodiment, in order to prevent the TOA constraint interval from being too small, T is set r The minimum value is 1000μs.

[0116] In a further embodiment, if there are multiple pulses with maximum or minimum PA values, the pulse pair with the smallest absolute value of the TOA difference is selected.

[0117] In a further embodiment, calculating the PA constraint interval of the current spatial clustering result according to the arrival time TOA difference and the pulse amplitude PA difference includes:

[0118]

[0119] pa r =Δpa / N T

[0120] Where N T is an intermediate parameter, Δdtoa represents the absolute value of the arrival time TOA difference, Indicates taking the upper limit, pa r Indicates the PA constraint interval of the current spatial clustering result, Δpa is the absolute value of the pulse amplitude PA difference. In this embodiment, in order to prevent the PA constraint interval from being too small, the pa is set r The minimum value is 1dBm.

[0121] Finally, drawing on the principles of the ST-DBSCAN algorithm, we perform temporal feature clustering on each clustering result based on the TOA and PA constraint intervals determined for each preliminary clustering result. After performing temporal feature clustering on each spatial feature clustering result, we obtain the final clustering result. By integrating the spatial feature clustering process with the temporal feature clustering process, we construct a temporal feature clustering algorithm.

[0122] Example 4

[0123] This embodiment further illustrates the batching of pulse group sequences based on embodiments 1 to 3.

[0124] In this embodiment, the time parallel relationship and time continuous relationship of the time series clustering results are analyzed, and the pulse group sequences of the time series clustering results are batched, including batching the pulse group sequences of the time series clustering results according to the parallel relationship between the pulse groups: the pulse sequence emitted by the same radar radiation source may be composed of multiple pulse groups, and there is a certain time series relationship between the pulse groups. If two pulse groups belong to the same radar radiation source, there are generally two types of relationships between the pulse groups on the time axis: one is that the start and end times of the two pulse groups are similar, which is called a parallel relationship; the other is that the two pulse groups are connected end to end in time, which is called a continuous relationship. The TOA of the first pulse and the (TOA+PW) of the last pulse of each pulse group in the clustering result are recorded as the start time and end time of the pulse group, respectively, and the pulse groups are arranged in ascending order according to the start time. According to the time series relationship between the pulse groups, information support can be provided for the subsequent batching of clustering results.

[0125] like Figure 5 As shown in FIG, in the pulse combination batch process, the pulse groups that have been arranged in the order of the start time are batched, which can be mainly divided into two stages: parallel relationship batching and continuous relationship batching.

[0126] First, the parallel relationship between the pulse groups is analyzed and judged, which can be divided into the following four judgment steps:

[0127] Perform time difference determination. Count the maximum DTOA values ​​in each pulse group in turn, and take the average value as the time difference threshold Treshold1. To prevent the time difference threshold from being too small, set the minimum value of Treshold1 to 1000μs. Calculate the start and end time differences between each pulse group and other pulse groups in turn. If a pulse group pair is found whose start and end time differences are both lower than Treshold1, proceed to the next step of determination.

[0128] Perform DOA judgment and set the DOA threshold Tresholddoa, which is set to 2° in this embodiment; record the DOA range and DOA average of the two pulse groups respectively. If the DOA range of one pulse group is fully covered by the DOA range of the other pulse group, or the difference between the DOA averages of the two pulse groups is lower than Tresholddoa, proceed to the next step of judgment;

[0129] Perform PA determination and set the PA threshold Tresholdpa, which is set to 2dBm in this embodiment. Record the PA ranges of the two pulse groups respectively. If the difference between the center values ​​of the PA ranges of the two pulse groups is lower than Tresholdpa, proceed to the next step of determination.

[0130] PRI entropy and duty cycle entropy are determined. The calculation methods of PRI entropy and duty cycle entropy are the same. Take PRI entropy as an example to illustrate: select a pulse group and perform statistics on the DTOA of the pulse group through histogram statistics. Assume that there are J different DTOA values ​​in the pulse group, and the number of pulses corresponding to the jth DTOA value is n j , then the PRI entropy of the pulse group can be expressed as:

[0131]

[0132] Where N is the total number of pulses in the pulse group. Calculate the PRI entropy H of the two pulse groups separately pri1 and H pri2 After merging the two pulse groups, rearrange them according to the pulse time sequence and calculate the PRI entropy H of the merged pulse group pri3 , if the following conditions are met:

[0133] H pri3 ≤(H pri1 +H pri2 )

[0134] The change pattern of the DTOA of the merged pulse group is considered to be more stable, and the two pulse groups meet the merging conditions of PRI entropy. The determination process of duty cycle entropy merging conditions is the same as that of PRI entropy. When two pulse groups meet the merging conditions of one of the entropy, they are considered to meet the parallel relationship and can be batched.

[0135] In a further embodiment, after all pulse groups that meet the parallel relationship are batched, the start and end times of each pulse group are recalculated, and the pulse groups are sorted in ascending order according to the start time. The continuity relationship between the pulse groups is analyzed, and the pulse groups are batched. The batching process for the continuity relationship generally needs to be performed twice.

[0136] The first batching process based on the continuous relationship includes:

[0137] 1) Perform time difference determination. Count the duration of each pulse group in turn and take the average value as the time difference threshold Treshold2. To prevent the time difference threshold from being too small, set the minimum value of Treshold2 to 5000 μs. Calculate the difference between the end time of each pulse group and the start time of subsequent pulse groups in chronological order. If a pulse group pair with a time difference less than Treshold2 is found, proceed to the next step of determination.

[0138] 2) Perform DOA judgment. According to the starting time sequence of the two pulse groups, they are recorded as the front pulse group and the back pulse group respectively. The last 32 pulses of the front pulse group are taken as pulse group A; at the same time, the first 32 pulses of the back pulse group are taken as pulse group B. If the number of pulses in the pulse group is less than 32, all pulses of the pulse group are extracted; the DOA average values ​​of pulse group A and pulse group B are calculated respectively. If the difference between the DOA average values ​​of the two pulse groups is lower than Tresholddoa, the next step of judgment is carried out;

[0139] 3) Perform PA determination and calculate the PA range of pulse group A and pulse group B respectively when one of the following conditions is met:

[0140] (a) The PA ranges of the two pulse groups do not overlap, and the PA difference between the last pulse of pulse group A and the first pulse of pulse group B is less than Tresholdpa;

[0141] (b) The PA range of one pulse group is covered by the PA range of another pulse group;

[0142] (c) The difference between the center values ​​of the PA range of the two pulse groups is less than Tresholdpa;

[0143] It is considered that the preceding pulse group and the following pulse group corresponding to pulse group A and pulse group B have a continuous relationship, and the next step of judgment is carried out;

[0144] 4) Perform directed graph search. After determining the continuous relationship between all pulse groups, treat each pulse group as a node and connect the pulse groups with continuous relationship with itself in chronological order to form a continuous relationship directed graph, such as Figure 6 As shown, find the longest path in the directed graph, batch the pulse groups on the path, delete the path from the directed graph, and find the longest path in the directed graph again until the length of the longest path found is less than the length threshold;

[0145] The second batching process based on the continuous relationship includes:

[0146] The pulse groups obtained from the first batching process were analyzed again for their continuous relationship. The processes for time difference determination, DOA determination, and directed graph search remained consistent. However, the PA determination criteria were changed to "the PA ranges of the two pulse groups do not overlap, and the average first-order difference of the PA sequences of the two pulse groups is less than 0.5 dBm." This addresses situations where the PA value of the electronic scanning radar signal jumps due to changes in the wave position.

[0147] Get the sorting results.

[0148] Example 5

[0149] Based on Examples 1 to 4, this embodiment further judges and classifies isolated points to further improve pulse utilization and sorting accuracy.

[0150] During the time series feature clustering algorithm, pulses that are independent of any clustering results will appear. These pulses are called isolated points. Among these isolated points, some are noise pulses, and some may be valid pulses that have not been correctly sorted. In order to improve the pulse utilization rate of the sorting algorithm, it is necessary to judge and classify these isolated points, such as Figure 7 Shown, including:

[0151] Calculate the start and end time of all sorting results, select an isolated point, and if the isolated point is within the time interval of a clustering result, proceed to the next step of judgment;

[0152] According to the TOA value of the isolated point, the last 50 pulses with TOA values ​​less than the isolated point TOA and the first 50 pulses with TOA values ​​greater than the isolated point TOA are selected from the eligible sorting results. If the number of pulses before or after the isolated point is less than 50, the actual number of pulses is extracted. These pulses are called reference points.

[0153] Calculate the confidence of each characteristic parameter of isolated points and reference points respectively:

[0154]

[0155] Where x k is the kth characteristic parameter value of the isolated point, g k , Δg k , σ k are the mean value, allowable deviation and mean square error of the kth characteristic parameter of the reference point, m is the number of characteristic parameters involved in the calculation, w k is the calculation weight of the mth feature parameter, and F(x) is the confidence that the isolated point belongs to the current clustering result;

[0156] Set the confidence threshold Thresholddoc, calculate the confidence of the isolated point and the reference point of the sorting result that meets the time relationship, and select the maximum value. If the maximum confidence is greater than Thresholddoc, the isolated point is considered to be assigned to the corresponding sorting result; if there are multiple maximum confidences, the sorting result with the largest number of pulses is selected for classification; if the maximum confidence does not exceed the threshold, the isolated point is considered to be a noise point;

[0157] The above steps are performed on all isolated points in sequence until all isolated points are assigned to corresponding sorting results or are regarded as noise points.

[0158] Ideally, after completing the processing of isolated points, the final unknown radar emitter sorting results will be obtained, and each sorting result corresponds to a radar emitter of unknown model.

[0159] Example 6

[0160] This example uses radar signal parameters from a database to construct a radar emitter dataset for demonstration purposes. A total of eight different radar emitters are configured, and interleaved pulse PDW sequences are constructed. The parameter settings for each radar are shown in Table 1. A pulse PDW sequence is generated for each radar, with the Gaussian noise standard deviations for RF, PW, PRI, DOA, and PA being 2 MHz, 0.1 μs, 1 μs, 1°, and 0.05 dBm, respectively. The starting TOA of each radar emitter pulse sequence is randomly selected between [0, 5000] μs, forming an interleaved pulse PDW sequence with a duration of 1 s. The pulse loss rate and noise pulse rate in the interleaved pulse sequence are 0.1, and 0.1, respectively. A total of 27,737 pulse data sets are generated. The RF, PW, DOA, and PA ranges for the noise pulses are [1000, 12000] MHz, [1, 50] μs, [0, 360]°, and [-50, 5] dBm, respectively, and follow a uniform distribution.

[0161] Table 1. Parameter settings of radar emitter

[0162]

[0163]

[0164] In the experiments, algorithms such as grid density clustering, density peak clustering, DBSCAN, DBSCAN+SDIF, and ST-DBSCAN were used to sort interleaved pulse trains using different pulse feature parameters and compared with the proposed method. When it comes to DOA-RF-PW meshing, the meshing process used in these sorting methods is consistent with that used in the proposed spatial feature clustering algorithm. The experimental results are shown in Table 2.

[0165] When clustering using only the spatial characteristics of pulses, the accuracy of the resulting sorting results is not ideal, as the parameters of radar emitter pulse signals are not always fixed. When adding the time series characteristics of TOA and PA to the clustering, the accuracy of the clustering results is significantly improved, but this also significantly increases the batch increase rate. Using the proposed batching algorithm to batch the sorting results after temporal feature clustering can effectively reduce the batch increase rate. Based on the evaluation metrics of various algorithms, our proposed method achieves superior sorting evaluation metrics.

[0166] Table 2. Detailed information on the sorting results of each radar emitter

[0167]

[0168]

[0169] The same or similar reference numerals correspond to the same or similar components;

[0170] The terms used in the drawings to describe positional relationships are for illustrative purposes only and should not be construed as limiting this patent;

[0171] Obviously, the above embodiments of the present invention are merely examples for the purpose of clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention. Those skilled in the art will appreciate that other variations or modifications can be made based on the above description. It is not necessary and impossible to enumerate all embodiments here. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention are intended to be included within the scope of protection of the claims of the present invention.

Claims

1. A method for sorting unknown radar emitters based on temporal feature clustering, characterized in that: The following steps are involved: Obtaining the radar signal pulse description word of the pulse signal; performing spatial clustering on the pulse signal according to the radar signal pulse description word to obtain a spatial clustering result; For each spatial clustering result, according to the radar signal pulse description word of the pulse signal in each spatial clustering result, the pulse signal in each spatial clustering result is clustered by time characteristics to obtain a time series clustering result; The time parallel relationship and time continuous relationship of the time series clustering results are analyzed, and the pulse group sequences of the time series clustering results are batched to obtain the sorting results.

2. The method for sorting unknown radar radiation sources based on temporal feature clustering according to claim 1, characterized in that: The radar signal pulse description word includes arrival time TOA, arrival angle DOA, carrier frequency RF, pulse width PW, pulse amplitude PA and pulse repetition interval PRI.

3. The method for sorting unknown radar radiation sources based on temporal feature clustering according to claim 2, characterized in that: Performing spatial clustering on the pulse signal according to the radar signal pulse description word to obtain a spatial clustering result includes: According to the pulse signal, setting a DOA-RF-PW three-dimensional grid of a first preset scale, performing grid density clustering on the pulse signal to obtain a first clustering result and noise pulses, merging the first clustering results, and rearranging the pulses in chronological order to obtain a denoised interleaved pulse sequence; Setting a DOA-RF-PW three-dimensional grid of a second preset scale according to the denoised interleaved pulse sequence, and calculating a grid density threshold according to the number of grids and grid density of the DOA-RF-PW three-dimensional grid of the second preset scale; Obtaining a high-density grid and a low-density grid according to a relationship between each grid density of the DOA-RF-PW three-dimensional grid of the second preset scale and the grid density threshold; The adjacent high-density grids are merged and subjected to boundary optimization processing, and the low-density grids adjacent to the high-density grids are merged at the same time to obtain a spatial clustering result.

4. The method for sorting unknown radar radiation sources based on temporal feature clustering according to claim 3, characterized in that: Calculating a grid density threshold according to the number of grids and the grid density of the DOA-RF-PW three-dimensional grid of the second preset scale includes: In the formula, MinPts grid is the grid density threshold, D={den1,den2,…,den i ,…,den N } is a sequence of all grid densities, den i is the i-th grid density, N is the number of grids, 1≤i≤N.

5. The method for sorting unknown radar radiation sources based on temporal feature clustering according to claim 3, characterized in that: According to the radar signal pulse description word of the pulse signal in each spatial clustering result, the pulse signal in each spatial clustering result is clustered by time characteristics to obtain the time series clustering results, including: Calculate the first-order difference of the arrival time TOA of the pulse signal in the current spatial clustering result to obtain the DTOA sequence; Calculating the TOA constraint interval of the current spatial clustering result according to the DTOA sequence; According to the two pulse signals with the largest and smallest pulse amplitudes PA in the current spatial clustering result, the arrival time TOA difference and the pulse amplitude PA difference of the two pulse signals are obtained; Calculating the PA constraint interval of the current spatial clustering result according to the arrival time TOA difference and the pulse amplitude PA difference; According to the TOA constraint interval and the PA constraint interval, time feature clustering is performed on the current spatial clustering result to obtain a time clustering result.

6. The method for sorting unknown radar radiation sources based on temporal feature clustering according to claim 5, characterized in that: Calculating the TOA constraint interval of the current spatial clustering result according to the DTOA sequence includes: Perform histogram statistics on the DTOA sequence and set the statistical threshold to a preset percentage of the total number of pulse signals in the current spatial clustering result. If there is a DTOA value whose statistical value exceeds the statistical threshold, the maximum value among the DTOA values ​​exceeding the statistical threshold is selected and recorded as T s If there is no DTOA value whose statistical value exceeds the statistical threshold, the maximum value among the DTOA values ​​is selected and recorded as T s ; Then the TOA constraint interval T of the current spatial clustering result is r for: T r =βT s Where β is a hyperparameter, which is an empirical value set according to the pulse loss rate in the real environment.

7. The method for sorting unknown radar emitters based on temporal feature clustering according to claim 6, characterized in that: Calculating the PA constraint interval of the current spatial clustering result according to the arrival time TOA difference and the pulse amplitude PA difference includes: pa r =Δpa / N T Where N T is an intermediate parameter, Δdtoa represents the absolute value of the arrival time TOA difference, Indicates taking the upper limit, pa r It represents the PA constraint interval of the current spatial clustering result, and Δpa is the absolute value of the pulse amplitude PA difference.

8. The method for sorting unknown radar radiation sources based on temporal feature clustering according to claim 5, characterized in that: Analyze the time parallel relationship and time continuous relationship of the time series clustering results, and batch the pulse group sequences of the time series clustering results, including batching the pulse group sequences of the time series clustering results according to the parallel relationship between the pulse groups: Record the arrival time TOA of the first pulse and the arrival time TOA + pulse width PW of the last pulse of each pulse group in the time clustering result as the start time and end time of the pulse group, respectively, and arrange the pulse groups in ascending order according to the start time; Analyze and judge the parallel relationship between pulse groups: The maximum DTOA value in each pulse group is counted in turn, and the average value is taken as the time difference threshold Treshold1. The start time difference and end time difference between each pulse group and other pulse groups are calculated in turn. If a pulse group pair with a start time difference and an end time difference lower than Treshold1 is found, the next step of judgment is carried out; Set the DOA threshold Tresholddoa, record the DOA range and DOA average of the two pulse groups respectively. If the DOA range of one pulse group is completely covered by the DOA range of the other pulse group, or the difference between the DOA averages of the two pulse groups is lower than Tresholddoa, proceed to the next step of judgment. Set the PA threshold Tresholdpa and record the PA ranges of the two pulse groups respectively. If the difference between the center values ​​of the PA ranges of the two pulse groups is lower than Tresholdpa, proceed to the next step of judgment. Calculate the PRI entropy or duty cycle entropy of the two pulse groups separately. After merging the two pulse groups, rearrange them according to the pulse time sequence, and calculate the PRI entropy or duty cycle entropy of the merged pulse group. If the PRI entropy or duty cycle entropy of the merged pulse group is not greater than the sum of the PRI entropy or duty cycle entropy of the two pulse groups before merging, the two pulse groups meet the parallel relationship. Merge the pulse groups that meet the parallel relationship.

9. The method for sorting unknown radar radiation sources based on temporal feature clustering according to claim 8, characterized in that: Analyze the time parallel relationship and time continuous relationship of the time series clustering results, and batch the pulse group sequences of the time series clustering results, including batching the pulse groups after merging the pulse groups that meet the parallel relationship according to the time continuous relationship between the pulse groups, including the first batching process and the second batching process: Recalculate the start time and end time of each pulse group, and arrange the pulse groups in ascending order according to the start time; The first batching process based on the continuous relationship includes: 1) Count the duration of each pulse group in turn, take the average value as the time difference threshold Treshold2, and calculate the difference between the end time of each pulse group and the start time of other subsequent pulse groups in chronological order. If a pulse group pair with a time difference less than Treshold2 is found, proceed to the next step of judgment; 2) According to the starting time sequence of the two pulse groups, they are recorded as the front pulse group and the back pulse group respectively. The last 32 pulses of the front pulse group are taken as pulse group A; at the same time, the first 32 pulses of the back pulse group are taken as pulse group B. If the number of pulses in the pulse group is less than 32, all pulses of the pulse group are extracted; the DOA average values ​​of pulse group A and pulse group B are calculated respectively. If the difference between the DOA average values ​​of the two pulse groups is lower than Tresholddoa, the next step of judgment is carried out; 3) Calculate the PA range of pulse group A and pulse group B respectively when one of the following conditions is met: (a) The PA ranges of the two pulse groups do not overlap, and the PA difference between the last pulse of pulse group A and the first pulse of pulse group B is less than Tresholdpa; (b) The PA range of one pulse group is covered by the PA range of another pulse group; (c) The difference between the center values ​​of the PA range of the two pulse groups is less than Tresholdpa; It is considered that the preceding pulse group and the following pulse group corresponding to pulse group A and pulse group B have a continuous relationship, and the next step of judgment is carried out; 4) After determining the continuity relationship between all pulse groups, each pulse group is regarded as a node, and the pulse groups with which it has a continuity relationship are connected backward in chronological order to form a continuous relationship directed graph. The longest path in the directed graph is found, the pulse groups on the path are batched, the path is deleted from the directed graph, and the longest path in the directed graph is found again until the length of the longest path found is lower than the length threshold; The second batching process based on the continuous relationship includes: Perform continuous relationship analysis again on the pulse group obtained from the first batch process. The processes of steps 1), 2), and 4) remain the same. In step 3), change the judgment condition to: The PA ranges of the two pulse groups do not overlap, and the average values ​​of the first-order differences of the PA sequences of the two pulse groups are both less than 0.5dBm; Get the sorting results.

10. The method for sorting unknown radar radiation sources based on time series feature clustering according to any one of claims 3 to 9, characterized in that: When performing time feature clustering, pulse signals that do not form a clustering result are regarded as isolated points, and the isolated points are judged and classified, including: Calculate the start and end time of all sorting results, select an isolated point, and if the isolated point is within the time interval of a clustering result, proceed to the next step of judgment; According to the TOA value of the isolated point, the last 50 pulses with TOA values ​​less than the isolated point TOA and the first 50 pulses with TOA values ​​greater than the isolated point TOA are selected from the eligible sorting results. If the number of pulses before or after the isolated point is less than 50, the actual number of pulses is extracted. These pulses are called reference points. Calculate the confidence of each characteristic parameter of isolated points and reference points respectively: Where x k is the kth characteristic parameter value of the isolated point, g k , Δg k , σ k are the mean value, allowable deviation and mean square error of the kth characteristic parameter of the reference point, m is the number of characteristic parameters involved in the calculation, w k is the calculation weight of the mth feature parameter, and F(x) is the confidence that the isolated point belongs to the current clustering result; Set the confidence threshold Thresholddoc, calculate the confidence of the isolated point and the reference point of the sorting result that meets the time relationship, and select the maximum value. If the maximum confidence is greater than Thresholddoc, the isolated point is considered to be assigned to the corresponding sorting result; if there are multiple maximum confidences, the sorting result with the largest number of pulses is selected for classification; if the maximum confidence does not exceed the threshold, the isolated point is considered to be a noise point; The above steps are performed on all isolated points in sequence until all isolated points are assigned to corresponding sorting results or are regarded as noise points.

Images