Radar signal sorting method based on improved OPTICS and PRI sequence correction matching

Abstract

The invention relates to a radar signal sorting method based on improved OPTICS and PRI sequence correction matching, belongs to the technical field of radar signal pattern recognition, and solves the problems of low sorting accuracy and sorting efficiency caused by factors such as serious signal pulse parameter overlapping, more pulse loss and more interference pulses of a radar signal sorter. Comprising the following steps: 1, preprocessing an input signal; 2, pre-sorting the pre-processed signals, and outputting a plurality of clusters; and step 3, main sorting is carried out on each cluster, and an identification result of the radar signal type in the input signal is obtained. For eight common radar signals, the average accuracy of signal sorting reaches 98% under the noise proportion and the pulse loss rate of 50%, the jitter signals with the PRI jitter range of 1-50% can be accurately recognized, and the high efficiency of radar signal sorting is further verified.

Description

Technical Field

[0001] The present invention relates to a radar signal sorting method based on improved OPTICS and PRI sequence correction matching, and belongs to the technical field of radar signal pattern recognition. Background Art

[0002] In the information-based war, it is undoubtedly crucial to correctly intercept radar information. Sorting out various radar radiation sources from the complex electromagnetic environment has become one of the key factors in winning the information-based war. Radar signal sorting can be divided into two stages: pre-sorting and main sorting. Pulse description words are usually used in pre-sorting to preliminarily extract the input complex signals. In the main sorting stage, for the signal clusters extracted by pre-sorting, the pulse repetition interval (PRI) is used for signal pattern recognition.

[0003] Aiming at the problems of low sorting accuracy caused by factors such as serious overlap of signal pulse parameters, many pulse losses, and many interference pulses received by the radar signal sorter, the density clustering algorithm has been used by many scholars in signal sorting. For example, Ping L et al. proposed an adaptive density peak clustering method based on subspace decomposition to solve the problem of low signal sorting accuracy. Liu Lutao, Su Y et al. introduced the dominant set and kernel density estimation methods into the DBSCAN algorithm respectively to achieve adaptive input parameters. However, it does not consider the specific implementation process of the subsequent main sorting stage, and the density clustering algorithm requires high complexity, large amount of computation, and low real-time performance. At the same time, common main sorting methods such as the PRI search method, PRI histogram, and PRI transformation method all have corresponding defects. For example, the PRI search method is only applicable to a simple electromagnetic environment and cannot handle complex changing situations. Typical representatives of the PRI histogram method include the cumulative difference and sequence difference histogram methods. The core idea is to set a threshold function to obtain potential pulse repetition intervals from the histogram. Such methods cannot extract jitter signals. The essence of the PRI transformation method is to utilize the correlation between PRI sequence values. When dealing with jitter signals, it cannot extract the PRI either. Summary of the Invention

[0004] The present invention aims to solve the problems of low sorting accuracy and sorting efficiency caused by factors such as serious overlap of signal pulse parameters, many pulse losses, and many interference pulses faced by the radar signal sorter, and further proposes a radar signal sorting method based on improved OPTICS and PRI sequence correction matching.

[0005] The technical solution adopted by the present invention to solve the above problems is as follows: The present invention includes the following steps:

[0006] Step 1: Preprocess the input signal;

[0007] Step 2: Presort the preprocessed signal and output several clustering clusters;

[0008] Step 3: Perform primary sorting on each clustering cluster to obtain the recognition result of the radar signal type in the input signal.

[0009] Further, step 1 specifically includes: normalizing and standardizing the carrier frequency and pulse width parameters of the input signal.

[0010] Further, step 2 specifically includes:

[0011] Step 2.1: Map the preprocessed input signal to a two-dimensional normalized plane coordinate system and divide the plane into several cells;

[0012] Step 2.2: Denoise the data in each cell according to the set denoising threshold;

[0013] Step 2.3: Extract data from the denoised data set;

[0014] Step 2.4: Perform density clustering according to the data extraction result to obtain several clustering clusters.

[0015] Further, step 2.2 specifically includes:

[0016] Step 2.2.1: Calculate the average value of the data points in the cell and round up as the denoising threshold;

[0017] Step 2.2.2: For each cell, if the number of data points in the cell is less than the denoising threshold, determine that the data points in the cell are noise points; if the number of data points in the cell is not less than the denoising threshold, determine that the data points in the cell are radar signals;

[0018] When the data in the cell is a noise point, the relational expression between the denoising threshold and the data point is:

[0019]

[0020] In formula (1), n is the number of signal types, A i is the number of pulses of the i-th signal, S is the number of noise points, and grid is the total number of grids;

[0021] When the data in the cell is a radar signal, the relational expression between the denoising threshold and the data point is:

[0022]

[0023] In formula (2), A i _grid is the number of unit grids occupied by the i-th signal.

[0024] Further, step 2.3 specifically includes:

[0025] Step 2.3.1: Set the data search range;

[0026] Step 2.3.2: Use the denoised data set as the original data set A. Randomly select a data point from the original data set A and add it to the blank set B as data point d, and delete the selected data point from the original data set A.

[0027] Step 2.3.3: Take out data point c from the original data set A. Determine whether the data point is within the search range of data point d in set B. If it is, represent data point c with data point b and do not update set B. If not, add data point c to set B for update operation.

[0028] Step 2.3.4: Repeat Step 2.3.3 until the original data set A is an empty set, and complete data extraction.

[0029] Further, Step 3 specifically includes:

[0030] Step 3.1: Correct the PRI sequence of the radar signals in the pre-sorted clustering clusters.

[0031] Step 3.2: Determine the type of the corrected radar signals according to the PRI histogram distributions of fixed radar signals, staggered radar signals, pulse group agile radar signals, pulse interval agile radar signals, linear sliding radar signals, triangular sliding radar signals, sinusoidal sliding radar signals, and jitter radar signals.

[0032] Further, Step 3.1 specifically includes:

[0033] Step 3.1.1: Set the original PRI mean of the radar signals in the clustering cluster as X and the pulse loss rate as P. Extract the adjacent pulses with n pulses lost between them in the clustering cluster to form the nth set, where the PRI mean of the constructed set is (n + 1)X and the weight is P. n ;

[0034] Step 3.1.2: Calculate the weighted mean of the nth set to obtain the PRI mean of the signal reception sequence. Convert the calculation formula of the PRI mean of the signal reception sequence according to the method of staggered subtraction of geometric series. Substitute the preset pulse loss rate P into the converted formula to obtain the maximum value of PRI. Take the maximum value of PRI as the correction threshold, and remove the radar signals higher than the correction threshold to complete the correction of the PRI sequence of the radar signals.

[0035] The calculation formula of the PRI mean of the signal reception sequence is:

[0036]

[0037] In formula (3), is the PRI mean of the reception sequence;

[0038] The calculation formula for the average PRI of the converted signal reception sequence is as follows:

[0039]

[0040] Further, step 3.2 specifically includes:

[0041] Search the PRI sequence of the corrected radar signal. If 5 pulses can be continuously searched, it is a successful search; if not, it is a failed search. Match the search result according to the PRI histogram distributions of fixed radar signals, staggered radar signals, pulse group agile radar signals, pulse interval agile radar signals, linear sliding radar signals, triangular sliding radar signals, sine sliding radar signals, and jitter signal radar signals to identify the corresponding 8 types of radar signals.

[0042] Further, the matching of the successfully searched PRI sequence specifically includes:

[0043] Search the PRI sequence of the corrected radar signal. If there are multiple peaks in the obtained PRI histogram, it is a staggered radar signal;

[0044] If there is only one peak, perform PRI analysis. If the PRI distribution is fixed, it is a fixed radar signal. If the PRI distribution performs frequency hopping in the form of "groups" and the number of peaks in the PRI histogram is 1 and the peak sizes are the same, it is a pulse group agile radar signal.

[0045] Further, the matching of the failed searched PRI sequence specifically includes:

[0046] Calculate the PRI correlation of the corrected PRI sequence. If the PRI correlation is greater than 0.95, that is, the hopping frequency points of the clustering clusters differ by less than 5%, it is a pulse interval agile radar signal. Extract the clustering clusters with a correlation greater than 0.95 and merge them according to the pulse arrival time in the clusters to obtain the complete pulse interval agile radar signal;

[0047] If the PRI correlation is not greater than 0.95, calculate the first-order difference histogram of the PRI sequence. If there is only one peak in the first-order difference histogram of the PRI sequence, judge the positive and negative situations of the first-order difference. If all the first-order differences in the first-order difference histogram of the PRI sequence are positive, it is a linear sliding radar signal. If half of the first-order differences in the first-order difference histogram of the PRI sequence are positive and half are negative, it is a triangular sliding signal;

[0048] If there are multiple peaks in the first-order difference histogram of the PRI sequence, the maximum values of the signal PRI sequence are sequentially extracted to form a maximum value sequence, and the deviation values that deviate from three times the standard deviation of the maximum value mean in the maximum value sequence are deleted. When the number of modes in the maximum value sequence is less than 1 / 5 of the total number of the maximum value sequence after deleting the deviation values, or the number of histogram peaks is less than 3, it is judged as a jitter radar signal; otherwise, it is a sine-swept radar signal.

[0049] The beneficial effects of the present invention are as follows:

[0050] 1. First, the present invention combines grid denoising and data extraction technologies to achieve preliminary noise filtering, streamline data, and reduce the computational complexity of the OPTICS algorithm. Then, the PRI sequence of the signal clusters obtained after OPTICS clustering is corrected to reduce the impact caused by pulse loss. Finally, type matching is performed according to the PRI distribution characteristics of the signal to complete signal sorting.

[0051] 2. The present invention conducts simulation experiments on the proposed sorting method. The simulation results show that for 8 common radar signals (including fixed, staggered, jitter, single linear sweep, triangular sweep, sine sweep, inter-pulse agility, and pulse group agility signals), at a 50% noise ratio and pulse loss rate, the average accuracy of signal sorting reaches 98%, and it can accurately identify jitter signals with a PRI jitter range of 1 - 50%, further verifying the high efficiency of the radar signal sorting of the present invention.

[0052] 3. By designing the steps of primary sorting and pre-sorting, the present invention overcomes the problems of the traditional density clustering algorithm, such as high complexity requirements, large computational volume, low real-time performance, and not considering the specific implementation process in the subsequent primary sorting stage. At the same time, in the pre-sorting stage, the present invention uses the OPTICS algorithm with optimized computational volume by grid denoising and data extraction technologies to cluster the carrier frequency and pulse width parameters of the complex received signals to extract signal clusters with different distribution shapes and remove interleaving. Subsequently, in the primary sorting, in response to the pulse loss phenomenon, the PRI sequence of the clustering cluster is corrected, and based on the histogram distribution characteristics of the PRI sequence, signal type matching is judged, overcoming the defects of the primary sorting methods (such as the PRI search method, the PRI histogram method, and the PRI transformation method), such as being unable to handle complex change situations, unable to extract jitter signals, and being unable to extract PRIs when dealing with jitter signals. Description of the Drawings

[0053] Figure 1 It is a schematic flowchart of the radar signal sorting method based on improved OPTICS and PRI sequence correction and matching provided by the present invention;

[0054] Figure 2 It is a schematic diagram of the pulse descriptor provided by the present invention;

[0055] Figure 3 Pre-sorting flowchart provided by the present invention;

[0056] Figure 4 Schematic diagram of the effect of grid denoising provided by the present invention, Figure 4 where (a) is the distribution diagram of initial data points, (b) is the visualization diagram of data point density, and (c) is the distribution diagram of data points after grid denoising;

[0057] Figure 5 Schematic diagram of the effect of data extraction provided by the present invention, Figure 5 where (a) is the distribution diagram of initial data points, and (b) is the distribution diagram of data points after clustering;

[0058] Figure 6 Schematic diagram of the relationship between the correct rate information entropy and the extraction rate provided by the present invention;

[0059] Figure 7 Main sorting flowchart provided by the present invention;

[0060] Figure 8 Schematic diagram of PRI correction for pulse loss of jitter radar signal provided by the present invention;

[0061] Figure 9 PRI correction ratio diagram provided by the present invention;

[0062] Figure 10 PRI schematic diagram of fixed radar signal provided by the present invention;

[0063] Figure 11 PRI schematic diagram of staggered radar signal provided by the present invention;

[0064] Figure 12 PRI schematic diagram of pulse group agile radar signal provided by the present invention;

[0065] Figure 13 Pulse repetition period histogram provided by the present invention, Figure 13 where (a) is the PRI histogram of fixed radar signal, and (b) is the PRI histogram of staggered radar signal;

[0066] Figure 14 PRI schematic diagram of pulse-to-pulse agile radar signal provided by the present invention;

[0067] Figure 15 PRI schematic diagram of linearly sliding radar signal provided by the present invention;

[0068] Figure 16 PRI schematic diagram of triangular sliding radar signal provided by the present invention;

[0069] Figure 17Schematic diagram of PRI of the sine-sweeping radar signal provided by the present invention;

[0070] Figure 18 Histogram of each frequency point of the pulse-to-pulse agile radar signal provided by the present invention, Figure 18 where (a) is the PRI histogram of cluster 1, (b) is the PRI histogram of cluster 2, (c) is the PRI histogram of cluster 3, and (d) is the PRI histogram of cluster 4;

[0071] Figure 19 Schematic diagram of the merged pulse-to-pulse agile radar signal provided by the present invention;

[0072] Figure 20 PRI histograms of the linear-sweeping radar signal and the triangular-sweeping radar signal provided by the present invention, Figure 20 where (a) is the PRI histogram of the linear-sweeping radar signal and (b) is the PRI histogram of the triangular-sweeping radar signal;

[0073] Figure 21 PRI first-difference histograms of the linear-sweeping radar signal and the triangular-sweeping radar signal provided by the present invention, Figure 21 where (a) is the PRI first-difference histogram of the linear-sweeping radar signal and (b) is the PRI first-difference histogram of the triangular-sweeping radar signal;

[0074] Figure 22 PRI histograms of the sine-sweeping radar signal and the jitter radar signal provided by the present invention, Figure 22 where (a) is the PRI histogram of the sine-sweeping radar signal and (b) is the PRI histogram of the jitter radar signal;

[0075] Figure 23 PRI maximum-value histograms of the sine-sweeping radar signal and the jitter radar signal provided by the present invention, Figure 23 where (a) is the PRI maximum-value histogram of the sine-sweeping radar signal and (b) is the PRI maximum-value histogram of the jitter radar signal;

[0076] Figure 24 Schematic diagram of the PRI sequence matching process provided by the present invention;

[0077] Figure 25 Effect diagrams of the improved and unimproved clustering provided by the present invention, Figure 25 where (a) is the schematic diagram of the clustering result after improvement processing and (b) is the schematic diagram of the clustering result after improvement processing;

[0078] Figure 26 Schematic diagram of the relationship between the consumption time correct rate and the number of data points provided by the present invention;

[0079] Figure 27 This is the sorting effect diagram of the radar signal sorting method proposed by the present invention. Detailed implementation manners

[0080] Detailed implementation manner one: Combine Figure 1 To illustrate this implementation manner, as Figure 1 shown, the steps of the radar signal sorting method based on improved OPTICS and PRI sequence correction matching described in this implementation manner include:

[0081] S1: Input the signal to be sorted.

[0082] To digitally describe the characteristics of radar signals, this implementation manner uses pulse description words. The radar pulse description word (Pulse Description Words, PDW) consists of pulse carrier frequency (Carrier Frequency, CF), pulse width (Pulse Width, PW), direction of arrival (Direction of Arrival, DOA), time of arrival (Time of Arrival, TOA), and pulse amplitude (Pulse Amplitude, PA). Figure 2 This is a schematic diagram of the radar signal pulse description word. These five parameters included in the pulse description word can be obtained through measurement, so they are called instantaneous parameters. The instantaneous parameters can be processed twice or multiple times to obtain secondary parameters. The most important of them is the pulse repetition interval (PRI), and the pulse repetition interval is the first-order difference of the signal arrival time sequence.

[0083] S2: Presort the input signal to be sorted.

[0084] S3: Perform main sorting on the presorted signal.

[0085] S4: Output the recognition result of the radar signal type in the signal.

[0086] Detailed implementation manner two: Combine Figure 2-6 To illustrate this implementation manner, this implementation manner is a further illustration of the presorting step in detailed implementation manner one; to digitally describe the characteristics of radar signals, pulse description words are usually used. The radar pulse description word (Pulse Description Words, PDW) consists of pulse carrier frequency (Carrier Frequency, CF), pulse width (Pulse Width, PW), direction of arrival (Direction of Arrival, DOA), time of arrival (Time of Arrival, TOA), and pulse amplitude (Pulse Amplitude, PA). Figure 2It is a schematic diagram of the pulse description word of the radar signal. These five parameters included in the pulse description word can be obtained through measurement, so they are called instantaneous parameters. The instantaneous parameters can be processed twice or multiple times to obtain secondary parameters, and the most important one is the pulse repetition interval (PRI). The pulse repetition interval is the first-order difference of the signal arrival time series.

[0087] As Figure 3 shown, the steps of the pre-sorting described in this embodiment include:

[0088] S201: Perform grid denoising on the input signal to be sorted;

[0089] The OPTICS algorithm itself has a certain anti-interference ability. However, when the number of noise points in the dataset to be clustered is too large, it will seriously consume the clustering time. To address this shortcoming, in this embodiment, the noise is preliminarily filtered in advance.

[0090] Grid denoising uses the density difference between noise and radar signals in the carrier frequency and pulse width parameter plane to remove the noise. The so-called density is defined as the number of data points in a unit cell of the two-dimensional normalized plane coordinate system. Normalization is to eliminate the influence of different dimensions.

[0091] Set the lengths of the X-axis and Y-axis of the two-dimensional graph to 1 unit, and the interval to 0.02, that is, there are 50×50 cells in the plane (the carrier frequency and pulse width have been standardized and normalized). When the number of data points in a cell is less than the preset threshold, the data points in that cell are determined to be noise points; if it is greater, the data points in the grid are considered to belong to the signal.

[0092] The denoising threshold is set to the average value (rounded up) of the data points in each unit grid. To achieve the separation of noise, the following two conditions need to be met:

[0093] Condition 1: The noise density is less than the preset threshold, that is, formula (1) is satisfied:

[0094]

[0095] In formula (1), n is the number of signal types, A i is the number of pulses of the i-th signal, S is the number of noise points, and grid is the total number of grids;

[0096] Condition 2: The density of any signal is greater than the preset threshold, that is, formula (2) is satisfied:

[0097]

[0098] In formula (2), A i _grid is the number of unit grids occupied by the i-th signal.

[0099] After meeting the conditions, the specific implementation effect of grid denoising is as follows Figure 4 shown, and the density visualization result of data points is as follows Figure 4 (b) shown, Figure 4 (a) and Figure 4 (c) By comparison, it can be seen that after grid denoising, the number of noise points is significantly reduced, indicating the effectiveness of introducing grid denoising in this embodiment.

[0100] S202: Extract data from the denoised signal;

[0101] The core of data extraction is that a small cluster of aggregated points is represented by a single data point inside it to streamline the data set and reduce the computational amount required for OPTICS clustering. Using this method can effectively improve the real-time performance of the sorting system. The specific steps include:

[0102] S20201: Set the data search range, which is set by the user himself, generally set to 0.01, and needs to be adjusted according to the actual situation of the radar signal faced.

[0103] S20202: Take the denoised data set as the original data set A, randomly select a data point from the original data set A and add it to the blank set B, and delete this point in A.

[0104] S20203: Take out the data point c (and delete the point c) from A in order, and judge whether there is a point d in B such that the data point c is within the search range of the point d. If it exists, the point c can be represented by the point b, and the set B does not need to be updated; if it does not exist, add the point c to the set B and perform an update operation.

[0105] S20204: Loop S20203 until the data set A is an empty set, and end the data extraction.

[0106] The data extraction effect is as follows Figure 5 shown, where different colors in the figure represent different clusters formed after clustering. Zoom in on the same position of Figure 5 (a) and (b) in, and observe the enlarged image. It can be seen that Figure 5 (b) is sparser than Figure 5 (a), and at the same time Figure 5 (b) The clustering effect corresponds to the original image, and data extraction is realized. Figure 6For the relationship between the accuracy rate, information entropy, and extraction rate. As shown in the figure, the accuracy rate starts to show an obvious downward trend after the extraction rate exceeds 80%. When the extraction rate is 0 (no data extraction operation is performed), the information entropy at this time is 11.5 bit. As the extraction rate increases, the information entropy continuously decreases. When the extraction rate is approximately 80% or more, the decline rate of the information entropy significantly increases. In summary, when performing data extraction operations, to ensure the clustering effect, the extraction rate in this embodiment is controlled within 80%.

[0107] S203: Perform density clustering based on the data extraction results, dividing them into several clustering clusters;

[0108] Specific Embodiment 3: Combining Figure 25 and Figure 26 This embodiment is described. To verify the effect of the pre-sorting step described in Specific Embodiment 2, the following simulation experiment is set up in this embodiment:

[0109] Experimental environment: System Windows 11, processor AMD Ryzen 7 4800H with Radeon Graphics, memory 16GB.

[0110] This embodiment tests the effectiveness of introducing grid denoising and data extraction technology in the radar pre-sorting stage for reducing the computational complexity of the OPTICS algorithm. The input radar signal parameters are shown in Table 1, and the noise ratio is set to 50% of the number of signals. The unit length of grid denoising is set to 0.02, the threshold is set to the average value of grid data points (rounded up), the extraction rate of data extraction is set to 75%, the simulation time is 1 s, and the number of input signal data points is 20,000.

[0111] Table 1

[0112]

[0113]

[0114] The simulation test results are as Figure 25 shown. The gray points in the two figures are the judged signal points, where different colors represent different clustering clusters, and the black points are the noise points judged by the clustering algorithm. It can be found that in both simulation cases, the adopted density clustering algorithm can successfully cluster the mixed signals into corresponding clusters, but obviously, as shown in Figure 25 (a), the number of data points to be processed for clustering after adopting the improved operation is much smaller than that in Figure 25 (a) without the improved treatment. Therefore, the computational complexity of the former is less than that of the latter. After considering the additional running time introduced by introducing grid denoising and data extraction technology, the overall consumption time in the pre-sorting stage after improvement is also less than that of the latter.

[0115] Test multiple groups of data (only change the number of input signal pulses), and plot Figure 26 . Observe Figure 26 From the two blue curves in , it can be seen that when the number of input data points is small, the time consumption of the improved processing and the unimproved processing is not much different. This is because when the number of data points is small, the OPTICS clustering consumes less time, and grid denoising and data extraction also consume time. As the number of input data points increases, the difference in time consumption between the two becomes larger and larger, and the time consumption of the improved density clustering is significantly reduced compared with the unimproved one. And when the number of data points exceeds 25,750 (as shown by the green vertical short dash line in the figure), observe Figure 26 From the two red curves in , it can be seen that the density clustering accuracy rates of the improved processing and the unimproved processing are not much different, and are basically above 98%. The accuracy rate uses the F value (F-measure), which is the harmonic mean of the precision P and the recall rate R. Equations 3-5 are the formulas for the precision P, the recall rate R, and the F value respectively. TP is the number of correctly clustered ones in the clustering cluster, FP is the number of wrongly clustered ones in the clustering cluster, and FN is the number of missed clusters corresponding to the clustering.

[0116]

[0117] The simulation results show that compared with the original algorithm, the improved density clustering algorithm proposed by the present invention can greatly reduce the consumption time, increase the number of data points that can be processed at one time, and at the same time does not affect the clustering accuracy rate when the number of input data points is large.

[0118] Specific Embodiment 4: Combine Figure 7-24 To illustrate this embodiment, this embodiment is a further description of the main sorting step described in the specific embodiment. The main sorting first corrects the PRI sequence of the clustering cluster formed in the preliminary sorting to eliminate the influence caused by pulse loss, and then uses the different characteristics of the pulse repetition periods of different radar signals for type recognition (matching) to improve the sorting accuracy rate.

[0119] As Figure 7 shown, the steps of the main sorting specifically include:

[0120] S301: Correct the PRI sequence of the radar signals in the clustering cluster obtained by preliminary sorting;

[0121] Since there is a phenomenon of pulse loss in the actual electromagnetic environment, the radar signals to be sorted are missing sequences, rather than complete signals. Taking the jitter signal as an example (the center PRI is 250 us and the jitter ratio is 15%), the PRI histogram of the signal under a pulse loss rate of 50% is as Figure 8 shown in (a) of , and it can be seen from the figure that the PRI sequence has multiple high-order components.

[0122] Therefore, if the PRI matching is directly performed on the signal clusters obtained by pre-sorting (the specific matching process is shown in Section 2.3), the signal type cannot be accurately determined. To correctly identify the signal type, the PRI sequence of the signal needs to be corrected and then the matching is performed. The principle and process of PRI sequence correction are as follows:

[0123] S30101: Let the original PRI mean of the signal be X, and the pulse loss rate be P. Extract the n lost pulses between adjacent pulses to form the nth set, then the PRI mean of this set becomes (n + 1)X, and the weight is P n . Calculate the weighted mean of the n sets to calculate the PRI mean of the signal reception sequence, as shown in formula (6):

[0124]

[0125] In formula (6), is the PRI mean of the reception sequence;

[0126] S30102: Applying the idea of staggered subtraction of geometric sequences to formula (6) can be simplified to formula (7):

[0127]

[0128] In this embodiment, it is preferably to set the pulse loss rate P to 50% and substitute it into formula (4). After pulse loss, the PRI average value of the signal becomes 2X. Considering that the jitter range of the jitter signal can reach up to 50%, that is, the maximum PRI value is 1.5X. To ensure the correction effect of the sequence (i.e., only removing the values outside the maximum value range), the correction threshold is set to 3 / 4 times the PRI mean of the signal clusters obtained by pre-sorting, and the removal operation is performed for values higher than the threshold. The histogram of the PRI sequence of the jitter signal after correction is as Figure 8 (b) shown, compared with Figure 8 (a), it can be observed that the high-order components caused by pulse loss are removed, and the correction operation of the sequence is successfully achieved.

[0129] Eight groups of experiments with different PRI values of the input signal are carried out, and the corresponding data are recorded in Table 2, and the graph is drawn based on the data Figure 9 , as Figure 9 shown. It can be seen that the mean of the received PRI sequence and the original mean are approximately in a 2-fold relationship numerically, which is consistent with the theoretical derivation. The deviation between the mean of the corrected sequence and the original mean is within 1 us and can be ignored. These eight groups of experiments further illustrate the effectiveness of the correction idea for correcting the PRI sequence of the signal.

[0130] Table 2

[0131]

[0132] S302: Determine the type of the corrected radar signal based on the PRI histogram distributions of fixed radar signals, staggered radar signals, pulse group radar signals, pulse-to-pulse frequency-agile radar signals, linearly-swept PRI radar signals, triangular-swept PRI radar signals, sinusoidal-swept PRI radar signals, and jitter radar signals.

[0133] The PRI sequence matching realizes signal type determination according to the different characteristics of the PRI histogram distributions of different signals. The implementation process of the PRI sequence matching in this embodiment is an improvement on the basis of traditional technologies and adds the type determination of triangular and sinusoidal-swept signals.

[0134] For a complete PRI sequence of a signal, if 5 pulses can be continuously searched, it is considered that the search is successful. Among the 8 types of radar signals studied in this embodiment, for fixed, staggered, and pulse group frequency-agile signals, since the PRI is fixed, they can be successfully searched, while the other signals cannot be successfully searched. According to whether the signal sequence can be successfully searched, the signals to be matched can be divided into two categories: successful search and failed search. For the common 8 types of radar signals, the overall main sorting stage flow chart of radar signals (including the PRI sequence correction and matching process) designed according to the PRI distribution characteristics of each radar signal is as Figure 24 shown.

[0135] S30201: Successful search:

[0136] The signals belonging to this category are fixed, staggered, and pulse group frequency-agile radar signals. The PRI schematic diagrams of the signals are respectively as Figure 10 , Figure 11 and Figure 12 shown. The PRI of the fixed signal is a constant value. The number of sub-pulse repetition periods of the staggered signal is m. Every time m pulses are generated, the pulse repetition period repeats cyclically, and the m sub-pulse repetition periods form a large frame pulse repetition period. It can be expressed by formula (8):

[0137] PRI i = PRI1 + PRI2 + … + PRI m i = 1, 2, …, m (8);

[0138] The carrier frequency of the pulse group frequency-agile signal jumps in the form of "groups". Generally, the number of pulses in a group is greater than or equal to 6. Figure 12 There are 4 pulse groups in

[0139] and each pulse group consists of 8 adjacent pulses. The carrier frequency randomly jumps among CF1, CF2, and CF3. Figure 13 Since the PRI of the fixed signal is fixed, the number of peaks in the PRI histogram in Figure 13(b) The number of peaks in the PRI histogram is greater than 1; the pulse group agile signal performs frequency hopping in the form of "groups". After clustering, the "batch increase" problem (the same signal is clustered into different clusters) will occur. The number of pulses in a "group" is usually not less than 6. Therefore, the number of peaks in the PRI histogram corresponding to these clusters is 1, and the peak sizes are the same. The method for determining the pulse group agile signal is that when the pulse width parameters, the number of peaks in the PRI histogram of multiple clusters are 1, and the peak sizes are the same, it is considered that these clusters form a pulse group agile signal.

[0140] S30202: Search failed:

[0141] Inter-pulse agile, linear sweep, triangular sweep, sinusoidal sweep, and jitter signals cannot be searched through sequences. For inter-pulse agile signals, the carrier frequencies of adjacent pulses randomly select from several preset parameter values for hopping. For example Figure 14 As shown, the carrier frequency is set to randomly select from 5 values. Linear, triangular, and sinusoidal sweep signals are respectively as Figure 15 、 Figure 16 and Figure 17 shown, and the PRI sweeps according to a certain rule. The PRI of the jitter signal jitters within the jitter range of the central PRI.

[0142] Due to the random hopping of the carrier frequencies of adjacent pulses in the inter-pulse agile signal, it is recognized as multiple clusters after density clustering. However, precisely because of the random hopping, there is a correlation between these clusters. As Figure 18 shown, the PRI histogram distributions of 4 clusters belonging to the same inter-pulse agile signal (hopping frequency points are 4.6, 4.7, 4.8, 4.9 GHz) are similar. Using the correlation of the PRI sequences of these clusters, the inter-pulse agile signal is extracted from the signal types with search failures. After determining the composition of the clustering clusters, they can be merged according to the arrival times of the pulses in the clusters to obtain a complete inter-pulse agile signal. The PRI histogram of the complete signal is as Figure 19 shown.

[0143] The PRI histograms of linear sweep and triangular sweep signals are as Figure 20 shown. Although the PRI change forms of both signals are linear, there are still differences. The linear signal increases from the minimum value to the maximum value, then jumps to the minimum value, and repeats cyclically; the triangular signal increases from the minimum value to the maximum value and then decreases to the minimum value, repeating cyclically

[13] . Since the paths from the maximum value to the minimum value of the linear and triangular signals are different, the first-order difference histograms of their PRI sequences are different. As shown in (a) and (b) of Figure 21 , the first-order differences of the linear signal are all positive values, while the first-order differences of the triangular signal are half positive and half negative. Based on this feature, it can be determined whether the clustering cluster of the signal to be determined is a linear sweep or a triangular sweep signal.

[0144] The adjacent PRIs of the sine-swept signal change according to the sine law, while the jitter signal is randomly selected within the specified range of the central value.

[13] , and the pulse repetition period is generally within ±15% of the average pulse repetition period. The jitter range γ is calculated as shown in formula (9):

[0145]

[0146] The PRI histograms of the sine-swept and jitter signals are as shown in Figure 22 . The PRI variation range is set to 5% for both. Based on the different PRI variation laws of the sine and jitter signals, the signal category can be determined. The specific process is to sequentially extract the maximum values of the PRI sequences of the two signals to form a maximum value sequence, delete the larger deviation values (the determination criterion can be taken as deviating from the mean by three standard deviations), and then calculate the variance of the maximum value sequence. Since the PRI of the jitter signal jitters continuously, the variance of its maximum value must be much larger than that of the sine signal. Therefore, when the number of modes in the maximum value sequence is less than 1 / 5 of the total number of the maximum value sequence after deleting the deviation values, or the number of histogram peaks is less than 3, it is judged as a jitter radar signal; otherwise, it is a sine-swept radar signal.

[0147] Simulation verification: For the signals in Figure 22 , the maximum values of the PRI sequences are found respectively. After removing the deviation, as shown in Figure 23 , a histogram is used to show the fluctuation of the maximum values, and the variance of the maximum values of each signal is marked at the lower right corner of the figure.

[0148] In the present invention, by introducing the OPTICS algorithm improved by grid denoising and data extraction techniques in pre-sorting, and proposing the PRI correction and matching determination algorithm in main sorting, the overall process of radar signal sorting is completely realized. Simulation experiments show that for 8 common radar signals with mixed pulse repetition frequencies, the sorting model has good sorting performance and can identify jitter signals with a small or large jitter ratio. Through experimental simulations from different angles, the effectiveness and feasibility of the signal sorting model are verified.

[0149] Specific implementation mode five: To verify the technical effect of the main sorting step proposed in specific implementation mode four, this implementation mode conducts a jitter ratio sorting simulation, specifically as follows:

[0150] Experimental environment: System Windows 11, processor AMD Ryzen 7 4800H with Radeon Graphics, memory 16GB.

[0151] Experimental conditions: The noise ratio and pulse loss rate are both 50%. The input signal is a mixed superposition of a sine-swept signal, a jitter signal, and noise. Among them, the central PRI of the sine signal is 200 μs, the central PRI of the jitter signal is 130 μs, and the simulation time is set to 1 s.

[0152] Procedure: Keep the fixed and sine signal parameters unchanged, and change the jitter ratio of the jitter signal PRI. The variation range is set to 1 - 50% of the central PRI. Table 5 shows the corresponding parameters of the jitter signal with different PRI jitter ratios and the sorting results in the radar signal sorting model (the corresponding jitter ratio is shown in the parentheses after the PRI data). Table 3 lists the measured PRI values of the signals and the sorting accuracy rates.

[0153] Table 3

[0154]

[0155] It can be seen from Table 3 that when the PRI parameter of the jitter signal jitters within the range of 1 - 50% of the center, the sorting accuracy rate of the sine sweep signal pulses is stable at 99%, and the accuracy rate of the jitter signal can reach 99.7%. At the same time, the error between the measured value and the set value of PRI is within 2%. These data demonstrate that the radar signal sorting model established by the present invention can achieve correct sorting for the jitter range of the jitter signal between 1 - 50%. In addition, when the variation amplitude ratio of the input sine sweep signal PRI is the same as the jitter ratio of the jitter signal PRI, the sorting model can also accurately determine the signal type, verifying the theory in the fourth specific implementation manner regarding the distinction between sine sweep and jitter signals.

[0156] Specific implementation manner six: To verify the technical effect of the radar signal sorting method based on improved OPTICS and PRI sequence correction matching proposed in the first specific implementation manner, the following simulation experiments were conducted in this implementation manner:

[0157] Experimental environment: System Windows 11, processor AMD Ryzen 7 4800H with Radeon Graphics, memory 16GB.

[0158] Input 8 common radar signals with different repetition frequencies to test the sorting effectiveness of the signal sorting model. The signal parameter settings refer to the Patriot system, and its core uses a certain multi-functional phased array radar. The radar bandwidth range is 4 - 8 GHz (or 4 - 6 GHz), and the frequency agility bandwidth is 640 MHz. The PRI value range is 80 - 1050 us, with a 50% noise ratio and pulse loss rate. The corresponding set parameters for each radar signal are shown in Table 4. The simulation time is 1 s, and the number of pulses flowing into the sorting model per second is 60,000 (including noise). Table 5 shows the final sorting results, listing the recognized types and accuracy rates (the calculation formula is shown in formula (5)). Figure 27 It is the sorting effect diagram of the radar signal. For easy observation, in Figure 27 the subgraphs all use the same color to represent the same radar signal, and different colors represent different radar signals. To better observe the signals, Figure 27(c) is Figure 27 (b) The effect diagram after noise filtering.

[0159] Table 4

[0160]

[0161] Table 5

[0162]

[0163] It can be observed from Table 5 that the measured values of the pulse repetition period of each signal in the input mixed signal and the signal type are accurately identified, and the sorting correct rate is above 98%. Table 4 and Figure 27 illustrate the effectiveness of the establishment of the signal sorting model for the pulse sorting of the mixed PRI signals.

[0164] In summary, the present invention overcomes the problems of high complexity, large computational amount, low real-time performance required by the traditional density clustering algorithm and the lack of consideration of the specific implementation process in the subsequent main sorting stage by designing the steps of main sorting and pre-sorting. At the same time, in the pre-sorting stage, the present invention uses the OPTICS algorithm optimized for computational amount by grid denoising and data extraction techniques to cluster the carrier frequency and pulse width parameters of the complex received signals to extract signal clusters with different distribution shapes and remove interleaving. Subsequently, for the main sorting, in response to the pulse loss phenomenon, the PRI sequence of the clustering cluster is corrected, and the signal type matching judgment is made based on the histogram distribution characteristics of the PRI sequence, overcoming the defects that the main sorting methods (such as PRI search method, PRI histogram, PRI transformation method) cannot handle complex change situations, cannot extract jitter signals, and cannot extract PRI when dealing with jitter signals.

[0165] The above is only a preferred embodiment of the present invention, and it does not impose any form of limitation on the present invention. Although the present invention has been disclosed above with a preferred embodiment, it is not intended to limit the present invention. Any person skilled in the art can make some changes or modifications to the above-disclosed technical content to form equivalent embodiments with equivalent changes within the scope of the technical solution of the present invention. However, as long as it does not depart from the content of the technical solution of the present invention, any simple modification, equivalent replacement, and improvement made to the above embodiments within the spirit and principle of the present invention still fall within the protection scope of the technical solution of the present invention.

Claims

1. A radar signal sorting method based on improved OPTICS and PRI sequence correction matching, characterized in that The steps of the radar signal sorting method based on improved OPTICS and PRI sequence correction matching include: Step 1: Preprocess the input signal; Step 2: Perform pre-sorting on the preprocessed signal and output several clustering clusters; Step 3: Perform main sorting on each clustering cluster to obtain the recognition result of the radar signal type in the input signal.

2. The radar signal sorting method based on improved OPTICS and PRI sequence correction matching according to claim 1, wherein Step 1 specifically includes: Standardize and normalize the carrier frequency and pulse width parameters of the input signal.

3. The radar signal sorting method based on improved OPTICS and PRI sequence correction matching according to claim 1, characterized in that, Step 2 specifically includes: Step 2.1: Map the preprocessed input signal to a two-dimensional normalized plane coordinate system and divide the plane into several cells; Step 2.2: Denoise the data in each cell according to the set denoising threshold; Step 2.3: Extract data from the denoised data set; Step 2.4: Perform density clustering according to the data extraction result to obtain several clustering clusters.

4. The radar signal sorting method based on improved OPTICS and PRI sequence correction matching according to claim 3, characterized in that Step 2.2 specifically includes: Step 2.2.1: Calculate the average value of the data points in the cell and round up to obtain the denoising threshold; Step 2.2.2: For each cell, if the number of data points in the cell is less than the denoising threshold, determine that the data points in the cell are noise points. If the number of data points in the cell is not less than the denoising threshold, determine that the data points in the cell are radar signals; When the data in the cell is a noise point, the relational expression between the denoising threshold and the data point is: In formula (1), n is the number of signal types, A i is the number of pulses of the i-th signal, S is the number of noise points, and grid is the total number of grids; When the data in the cell is a radar signal, the relational expression between the denoising threshold and the data point is: In formula (2), A i _grid is the number of unit grids occupied by the i-th signal.

5. The radar signal sorting method based on improved OPTICS and PRI sequence correction matching according to claim 3, characterized in that, Step 2.3 specifically includes: Step 2.3.1: Set the data search range; Step 2.3.2: Use the denoised data set as the original data set A, randomly select a data point from the original data set A and add it to the blank set B as the data point d, and delete the selected data point from the original data set A; Step 2.3.3: Take out the data point c from the original data set A and determine whether the data point is within the search range of the data point d in the set B. If it is, represent the data point c with the data point d and the set B does not need to be updated. If not, add the data point c to the set B for update operation; Step 2.3.4: Repeat step 2.3.3 until the original data set A is an empty set, and complete the data extraction.

6. The radar signal sorting method based on improved OPTICS and PRI sequence correction matching according to claim 1, wherein Step 3 specifically includes: Step 3.1: Correct the PRI sequence of the radar signals in the clustering clusters obtained by pre-sorting; Step 3.2: Determine the type of the corrected radar signal according to the PRI histogram distributions of fixed radar signals, staggered radar signals, pulse group agile radar signals, pulse interval agile radar signals, linear sliding radar signals, triangular sliding radar signals, sinusoidal sliding radar signals, and jitter signal radar signals.

7. The radar signal sorting method based on improved OPTICS and PRI sequence correction matching according to claim 6, wherein Step 3.1 specifically includes: Step 3.1.1: Set the original PRI mean of the radar signals in the clustering cluster as X, and the pulse loss rate as P. Extract the cases where n pulses are lost between adjacent pulses in the clustering cluster to form the nth set. The PRI mean of the constructed set is (n + 1)X, and the weight is P n ; Step 3.1.2: Calculate the weighted mean of the nth set to obtain the PRI mean of the signal reception sequence. Convert the calculation formula of the PRI mean of the signal reception sequence by the method of staggered subtraction of the geometric sequence. Substitute the preset pulse loss rate P into the converted formula to obtain the maximum value of PRI. Use the maximum value of PRI as the correction threshold and remove the radar signals higher than the correction threshold to complete the correction of the PRI sequence of the radar signals; The calculation formula for the mean PRI of the signal reception sequence is as follows: In formula (3), is the mean PRI of the received sequence; The calculation formula for the mean PRI of the converted signal reception sequence is as follows:

8. The radar signal sorting method based on improved OPTICS and PRI sequence correction matching according to claim 6, wherein, Step 3.2 specifically includes: Search the PRI sequence of the corrected radar signal. If 5 pulses can be continuously searched, it is a successful search; otherwise, it is a failed search. Match the search result according to the PRI histogram distributions of fixed radar signals, staggered radar signals, pulse group agile radar signals, pulse interval agile radar signals, linear sliding radar signals, triangular sliding radar signals, sinusoidal sliding radar signals, and jitter radar signals to identify the corresponding 8 types of radar signals.

9. The radar signal sorting method based on improved OPTICS and PRI sequence correction matching according to claim 8, wherein The matching of the successfully searched PRI sequence specifically includes: Search the PRI sequence of the corrected radar signal. If there are multiple peaks in the obtained PRI histogram, it is a staggered radar signal; If there is only one peak, perform PRI analysis. If the PRI distribution is fixed, it is a fixed radar signal. If the PRI distribution performs frequency hopping in the form of "groups" and the number of peaks in the PRI histogram is 1 and the peak sizes are the same, it is a pulse group agile radar signal.

10. The radar signal sorting method based on improved OPTICS and PRI sequence correction matching according to claim 8, wherein The matching of the failed searched PRI sequence specifically includes: Calculate the PRI correlation of the corrected PRI sequence. If the PRI correlation is greater than 0.95, that is, the frequency hopping points of the clustering clusters differ by less than 5%, it is a pulse interval agile radar signal. Extract the clustering clusters with a correlation greater than 0.95 and merge them according to the pulse arrival time in the cluster to obtain the complete pulse interval agile radar signal; If the PRI correlation is not greater than 0.95, calculate the first-order difference histogram of the PRI sequence. If there is only one peak in the first-order difference histogram of the PRI sequence, judge the positive and negative conditions of the first-order difference. If all the first-order differences in the first-order difference histogram of the PRI sequence are positive, it is a linear sliding radar signal. If half of the first-order differences in the first-order difference histogram of the PRI sequence are positive and half are negative, it is a triangular sliding signal; If there are multiple peaks in the first-order difference histogram of the PRI sequence, sequentially extract the maximum values of the signal PRI sequence to form a maximum value sequence, and delete the deviation values that deviate from the mean of the maximum value by three standard deviations in the maximum value sequence. When the number of modes in the maximum value sequence after deleting the deviation values is less than 1 / 5 of the total number of the maximum value sequence, or the number of histogram peaks is less than 3, it is judged as a jitter radar signal; otherwise, it is a sinusoidal sliding radar signal.

Images