A passive electromagnetic situation awareness system based on a digital channelization model

Abstract

The present invention provides a passive electromagnetic situation awareness system based on a digital channel model, which includes a signal receiving module, a channelization module, a parameter measurement module, a signal direction finding module, and a signal sorting module. This system uses channelization to achieve broadband signal parameter measurement and analysis, and forms pulse description words. It completes the sorting of radiation source signals by combining pre-sorting and main sorting, and uses the interferometer method to complete the direction finding of radiation source signals. It can be well applied in radar reconnaissance systems, is simple and efficient, has low time complexity and space complexity, and solves the problems such as poor clustering effect of clustering algorithms and only obtaining local optimality in the current complex and dense environment and the serious overlap of characteristic parameters.

Description

Technical Field

[0001] The present invention belongs to the technical field of electromagnetic environment measurement, and particularly relates to a passive electromagnetic situation awareness system based on a digital channelization model. Background Art

[0002] In modern high-tech wars and information-based wars, the electronic countermeasure technology has developed rapidly, and various new and complex jamming radars have emerged in an endless stream. Especially with the development of technologies such as intra-pulse waveform transformation technology and multi-parameter agile change, as well as the emergence of an extremely complex radar signal environment characterized by the comprehensive application of various working systems and a variety of anti-jamming technologies, it poses a severe challenge to radar signal capture. How to efficiently and accurately counter the complex and changeable electromagnetic interference environment is crucial for the electromagnetic environment situation awareness and analysis system. Summary of the Invention

[0003] The present invention provides a passive electromagnetic situation awareness system based on a digital channel model, which can be an important part of a passive reconnaissance system. The main function of the system is to complete the detection, direction finding, and sorting of radar signals in a complex electromagnetic environment, and display the results through a host computer.

[0004] The passive electromagnetic situation awareness system includes:

[0005] A signal receiving module, configured to receive radar signals of N channels;

[0006] A channelization module, configured to channelize the radar signals of N channels to form N-channelized signals;

[0007] A parameter measurement module, configured to select one of the N-channelized signals as a first channelized signal, form a synchronization pulse according to the first channelized signal, and complete pulse parameter measurement according to the synchronization signal to obtain parameter measurement information;

[0008] A signal direction finding module, configured to extract the instantaneous phase of the other N-1 channelized signals except the first channelized signal to obtain angle information;

[0009] A signal sorting module, configured to analyze the parameter measurement information and the angle information by using an adaptive K-means clustering algorithm combined with an SDIF algorithm, and output a radar signal sorting result.

[0010] Further, the parameter measurement module processes the radar signals of N channels by using a decimation-in-advance uniform polyphase filter bank to obtain N-channelized signals.

[0011] Further, the parameter measurement module jointly processes the instantaneous frequency and the video pulse to make a decision on the channel.

[0012] Further, the signal direction finding module uses the multi-baseline phase interferometer algorithm to implement signal direction finding.

[0013] Further, the signal direction finding module uses the multi-baseline ambiguity resolution algorithm to sequentially resolve the multi-valued ambiguity of the long baseline from the short baseline to the maximum baseline.

[0014] Further, when the signal sorting module analyzes the parameter measurement information and the angle information using the adaptive K-means clustering algorithm combined with the SDIF algorithm, the following steps are completed:

[0015] (1) Initialization, input all data to be sorted; set the number of clustering categories D = 0;

[0016] (2) Use the improved K-means clustering algorithm to cluster all data according to pulse width, carrier frequency, and bandwidth, and the number of clustering results is D;

[0017] (3) Read each clustering data one by one, and use the optimized SDIF algorithm to calculate the sequence difference for the data in each cluster;

[0018] (4) If the number of PRIs obtained is greater than 1, perform discrimination and stagger verification on these PRIs, and go to (3); if the number of PRIs is less than 2, go to (5);

[0019] (5) If there are still unprocessed clusters, go to (3), otherwise go to (6);

[0020] (6) Perform frequency agile radar verification on the PRIs of different clustering results;

[0021] (7) Output the sorting result.

[0022] Compared with the prior art, the present invention has the following advantages:

[0023] (1) During the channelization process, the decimator is set before the polyphase filter bank and the mixer, the data speed input to the polyphase filter bank is reduced, and the operation efficiency is improved. Processing is performed jointly with the instantaneous frequency and the video pulse, and the channel is judged, reducing the influence of false signals on the channel judgment.

[0024] (2) The multi-baseline phase interferometer algorithm is used to implement signal direction finding, solving the contradiction between the angle measurement accuracy and the ambiguity number, and a multi-stage ambiguity resolution method is adopted to reduce the errors in the ambiguity resolution process.

[0025] (3) The adaptive K-means clustering algorithm combined with the SDIF algorithm is used for analysis, making up for the deficiencies of the original algorithm. It is not necessary to preset the number of radar signals, but to discriminate according to the radar signals to be clustered. The sorting result reaches the global optimum, and it is also simple and efficient for large data sets, with low time complexity and space complexity. Brief Description of the Drawings

[0026] Figure 1 It is a system block diagram of an implementation manner of the passive electromagnetic situation awareness system proposed by the present invention;

[0027] Figure 2 It is a schematic diagram of a uniform filter bank channelization structure;

[0028] Figure 3 It is a schematic diagram of a uniform polyphase filter bank channelization structure;

[0029] Figure 4 It is a schematic diagram of a decimation-preceded uniform polyphase filter bank channelization structure;

[0030] Figure 5 It is a schematic diagram of an efficient digital channelization structure in the present invention;

[0031] Figure 6 It is a flowchart of the formation of frequency measurement information of a channelized receiver in the present invention;

[0032] Figure 7 It is a dual-baseline phase interferometer model;

[0033] Figure 8 It is a flowchart of the implementation of the interferometer algorithm in the present invention;

[0034] Figure 9 It is a flowchart of an adaptive k-means clustering algorithm;

[0035] Figure 10 It is a flowchart of the SDIF algorithm;

[0036] Figure 11 It is a flowchart of signal sorting of the joint adaptive K-means clustering algorithm and the SDIF algorithm in the present invention. Detailed Implementation Manner

[0037] To make the technical problems, technical solutions and advantages to be solved by the present invention clearer, the following will be described in detail with reference to the accompanying drawings and specific embodiments.

[0038] The present invention needs to complete digital channelization function, parameter measurement function, signal sorting function, and signal direction finding function. Parameter measurement mainly measures pulse front frequency, pulse width, pulse arrival time, and instantaneous phase.

[0039] For this purpose, the present invention provides a passive electromagnetic situation awareness system based on a digital channel model, including:

[0040] A signal receiving module, configured to receive radar signals of N channels;

[0041] A channelization module, configured to channelize the received radar signals of N channels to form N-channelized signals;

[0042] A parameter measurement module, which is used to select one of the N-channelized signals as the first channelized signal, form a synchronization pulse based on the first channelized signal, complete pulse parameter measurement according to the synchronization signal, and obtain parameter measurement information;

[0043] A signal direction finding module, which is used to extract the instantaneous phase of the other N - 1 channelized signals except the first channelized signal to obtain angle information;

[0044] A signal sorting module, which is used to analyze the parameter measurement information and angle information by using an adaptive K-means clustering algorithm combined with the SDIF algorithm, and output the radar signal sorting result.

[0045] The following takes the input of a broadband intermediate frequency signal including 7 channels as an example to illustrate the passive electromagnetic situation awareness system of the present invention.

[0046] The overall block diagram of the system is as Figure 1 shown. One of the signals is used for signal detection and parameter measurement, and the other six signals are only used for extracting the instantaneous phase of the signal. The signal direction finding function is completed according to the phase difference, and the angle information is added to the pulse description word to complete signal sorting together with the pulse parameter measurement information.

[0047] The system mainly includes the following parts:

[0048] (1) The 7-channel ADC acquisition signals are respectively channelized into 16 channels and then become narrowband low-speed signals.

[0049] (2) Select one of the channelized signals to complete the formation of a synchronization pulse, and complete pulse parameter measurement (such as CF, PW, TOA) according to this synchronization signal.

[0050] (3) Each signal completes the extraction of 4 instantaneous phases at the pulse front edge according to the synchronization pulse, and calculates the average value of the phase differences of the 4 selected channels.

[0051] (4) PDW formation, taking CF, PW, PA, TOA, and phase difference as PDW, and transmitting them to the DSP through the EMIF interface.

[0052] (5) After receiving the start command, the DSP samples the PDW data of the FPGA, first completes signal direction finding, generates angle information and integrates it into the PDW to form a new pulse description word, and completes signal sorting.

[0053] (6) The DSP reports the reconnaissance result.

[0054] The overall design idea of the system is as follows:

[0055] (1) Channelization is used to measure and analyze the parameters of broadband signals and form pulse description words. The signal sorting of radiation sources is completed by combining pre-sorting and main-sorting, and the direction finding of radiation source signals is completed by using the interferometer method.

[0056] (2) Using the band-pass sampling theorem, the 7-channel signals are sampled respectively in the third Nyquist interval, and the high-speed serial data is decelerated to parallel data. The sampling results of the main channel are sent to the channelization module, and the data is converted into 16-channel orthogonal IQ data through channelization processing. The IQ data is sent to the CORDIC module to obtain the instantaneous phase and amplitude of the signal. The amplitude information can be used for channel decision, and the envelope of the pulse is formed accordingly, so as to measure the pulse width and arrival time. The instantaneous phase information can be sent to the instantaneous phase difference module for instantaneous frequency measurement. Only IQ data is obtained for the remaining 6 channels.

[0057] (3) Select the IQ signals of the input signals of 4 channels, calculate the phase differences of the corresponding two signals pairwise, and obtain the phase differences at different multiples of half-wavelength. The low-precision unambiguous result calculated from the phase difference of 1.5 times the half-wavelength is used to resolve the ambiguity of the high-precision ambiguous result in the next step, so as to obtain the angle information.

[0058] (4) The pulse description words containing carrier frequency, pulse width, arrival time, and arrival angle information are sent to the signal sorting module. In modern warfare, with the continuous emergence of new radar systems, the electromagnetic environment has become increasingly complex, which poses new challenges to radar signal sorting. The currently commonly used typical clustering algorithms have the disadvantage of presetting the number of radars. This algorithm is difficult to adapt to the modern complex radar signal environment. The present invention proposes an adaptive K-means clustering algorithm, which makes up for the deficiencies of the original algorithm. It does not require pre-setting the number of radar signals, but discriminates according to the radar signals to be clustered. The sorting result reaches the global optimum, and it is also simple and efficient for large data sets, with low time complexity and space complexity.

[0059] The working principles and implementation methods of each module are specifically introduced below.

[0060] I. Channelization Module

[0061] In order to perform real-time signal processing and analysis, it is necessary to use a channelization structure to decelerate the received radar signals and divide them into multiple channels for processing.

[0062] There are two ways of channel division: uniform and non-uniform. The sub-channels of uniform channel division have the same channel bandwidth and the same interval between the center frequencies of each channel. The sub-channels of non-uniform division have unequal bandwidths and unequal intervals between the center frequencies of the channels. The channel division structure of uniform division is easier to implement. In the present invention, a uniform channel division method is used to design the filter bank. The channelization structure of the uniform filter bank is asFigure 2 as shown

[0063] The filter bank divides the monitoring bandwidth into K channels, that is, the filter consists of K filters, including 1 low-pass filter h 0 [n] and K - 1 band-pass filters h k [n]. The length of the low-pass filter h 0 [n] is N, h 0 [n] = {h[0], …, h[N - 1]}, 0 ≤ n ≤ N - 1, and h[0], …, h[N - 1] are filter coefficients. The band-pass filter h k [n] is obtained by modulating the low-pass filter h 0 [n]. The unit impulse response of the k-th filter is:

[0064]

[0065] where ω k = 2πk / K, k = 0, 1, …, K - 1. The frequency response of the k-th filter is

[0066]

[0067] The analog signal x(t) is sampled by the ADC to output the discrete-time signal x(n) and input into the filter bank. The outputs y k (n) of the k sub-bands of the filter bank are

[0068]

[0069] In order to implement the decision logic, the output y k (n) of the band-pass filter needs to be multiplied by the complex exponential to transform to the baseband. Let the baseband signal be u k (n), then

[0070] u k (n) = y k (n)e -j2πkn / K

[0071] The frequency of the baseband signal is limited in the range of -2π / K ≤ ω ≤ 2π / K. After being decimated by M times, there is

[0072]

[0073] The Fourier transform of the above formula is

[0074]

[0075] After M times of decimation, the output frequency of each channel.. is limited to the range of -2πM / K ≤ ω′ ≤ 2πM / K. In order to avoid aliasing in the transition band of each sub-channel filter, it is necessary to satisfy

[0076] 2πM / K ≤ π

[0077] And K is an integer multiple of M.

[0078] F = K / M ≥ 2

[0079] Let K = FM, and the polyphase representation of the prototype filter h 0 [n] is:

[0080]

[0081] Among them, the polyphase component (rounded up to the nearest integer of N / K). Substitute the above formula into Equation We get:

[0082]

[0083] It can be seen from the above formula that the output of the signal passing through the k-th band-pass filter h k (n) is the result of performing IDFT on the output after the signal passes through the polyphase component filtering. The channelized structure after the polyphase representation of the filter bank is as Figure 3 shown.

[0084] The polyphase representation of the signal transforms the filter bank composed of K filters with length N into K filters with length and the IDFT operation of K points. The output of the IDFT is

[0085]

[0086] Among them, (k = 0, 1,... K - 1), and its Z-transform is

[0087]

[0088] The k-th output y k [n] of the IDFT, after digital quadrature demodulation and M-fold decimation, we have

[0089]

[0090] Figure 3 The decimator is after the low-pass filter bank and the mixer, indicating that only 1 / M of the sampled data after low-pass filtering and quadrature down-conversion is output, wasting a large amount of calculation results. Therefore, it is necessary to modify the operation order to decimate first and then filter digital down-conversion, that is, move the M-fold decimation forward to before the polyphase structure E l (z K ). According to the multi-rate data conversion formula, the output of the polyphase filter bank after moving the decimation forward satisfies

[0091]

[0092] The z-transform of the above formula is

[0093]

[0094] Since K / M = F is an integer, so e -j2πmK / M = 1, we have

[0095]

[0096] After M-fold decimation and shifting, the data rate of the input to the polyphase filter bank is reduced by M times, and the output of the polyphase filter bank is

[0097]

[0098] where N is the length of the filter. The output of the IDFT is

[0099]

[0100] The channelized structure after M-fold decimation and shifting is as Figure 4 shown.

[0101] According to the above conditions, to avoid aliasing between adjacent channels, it is necessary to satisfy F = K / M ≥ 2. In the present invention, F = 2 is taken. Therefore, the polyphase component The polyphase component E l (z 2 ) means inserting zeros twice for the coefficients of each polyphase branch filter. When F = 2, the exponential product term for digital quadrature demodulation And e -jπkn satisfies

[0102]

[0103] Replacing the IDFT operation with the fast algorithm IFFT, an efficient digital channelized receiver structure can be obtained as Figure 5 shown.

[0104] Figure 5 In to are the channelized outputs of K channels. After orthogonal baseband demodulation, the channelized outputs except and are real numbers, and the rest are complex numbers. And and have the same output value, while and (1 ≤ l ≤ K / 2 - 1) have conjugate output values. Therefore, only the first K / 2 channelized outputs need to be considered in subsequent data processing.

[0105] When the adjacent channel frequency responses overlap by 50%, an input signal will fall on two adjacent channels simultaneously, resulting in spurious signals. To reduce the impact of spurious signals on channel decision-making, the present invention combines instantaneous frequency and video pulses for processing and makes channel decisions. The complete block diagram of channel decision-making implementation is as shown in Figure 6 Figure [to be filled in later].

[0106] (1) The amplitude A k (n) of the k-th sub-channel obtained by the CORDIC algorithm is compared with the threshold V th . When it is greater than the threshold, the frequency parameter estimation is triggered.

[0107] (2) When the frequency of the input signal satisfies the condition of in channel k, the channel decision considers that the signal belongs to channel k. The decision condition for the signal to be valid in the k-th sub-channel is

[0108]

[0109] where f c is the signal processing bandwidth of each sub-channel of the channelized receiver, and is the average frequency of channel k.

[0110] (3) The frequency measurement uses the first-order difference method of instantaneous phase, and the formula is where T s is the sampling interval.

[0111] (4) The instantaneous phase φ k [n] is restricted within the range of [-π, π]. To obtain the true phase , phase unwrapping is required. The unwrapping algorithm is based on the original phase. According to the backward difference of the phase, a correction sequence c[n] is added to the instantaneous phase φ k [n], with the initial value c[n] = 0.

[0112]

[0113] When using the CORDIC algorithm to calculate the signal amplitude, the calculation accuracy is related to the iteration period of the CORDIC implementation. The more iterations, the more accurate the calculation, but the processing time is longer. When implemented using FPGA, the JPL algorithm can be used for approximate calculation, and its deviation is less than 3%. Assuming that the real and imaginary parts of the complex signal are I and Q respectively, to calculate Let A = max{|I|, |Q|} and B = min{|I|, |Q|}, then the approximate operation of the modulus can be expressed as

[0114]

[0115] Using the above method to calculate the modulus value of a complex signal can be completed with only a small amount of logic resources.

[0116] II. Channelization Module

[0117] The direction finding of a signal is realized by using the multi-baseline phase interferometer algorithm. The multi-baseline phase interferometer direction finding method is proposed to solve the contradiction between the angle measurement accuracy and the ambiguity number. Multiple single-baseline phase interferometers with different baseline lengths are formed by multiple antennas. The interferometer system formed by combining these baselines with different lengths is called a multi-baseline phase interferometer. The short-baseline interferometer is used to eliminate the phase ambiguity, and the long-baseline interferometer is used to improve the angle measurement accuracy, also known as the long-short baseline method.

[0118] As Figure 7 shown, it is a dual-baseline phase interferometer. K 0 K 1 K 2 constitute a dual-baseline phase interferometer direction finding system on a plane. K 0 K 1 constitute the short baseline d, K 0 K 2 constitute the long baseline D. Then there are:

[0119]

[0120]

[0121] φ 2 = m × φ 1

[0122] D = m × d

[0123] In the above formula, and are both phase differences obtained through the phase discriminator. φ 1 and φ 2 are the true values of the phase differences corresponding to the actual incident angle θ. m is the ratio of the lengths of the long and short baselines. According to the design rules, the short baseline d must be small enough to ensure that it is an unambiguous baseline. In this way, is equal to φ 1 . In this way, the phase difference obtained through the short baseline can be used to solve the ambiguous value n of the long baseline phase difference, and high-precision data measured by the long baseline can be obtained.

[0124] The actual de-ambiguation process is as follows:

[0125] (1) Given that the phase difference obtained from the long baseline is then a set of phase value arrays with a difference of 2π is obtained according to n equal to 0, ±1, ±2...

[0126] (2) Find the one closest to the short baseline phase difference m×φ 1 The value obtained is the actual accurate value φ 2 .

[0127] (3) Based on the accurate value φ obtained in the previous step 2 , this value is the angle measurement phase difference with better accuracy obtained using the long baseline than the short baseline. Finally, the incident angle θ of the signal is calculated.

[0128] During the use of multi-baseline ambiguity resolution, the following issues need attention:

[0129] (1) The shortest baseline of the multi-baseline phase interferometer must ensure no ambiguity for the measured angle, which is the basis for ensuring the correct operation of the entire direction-finding system. The critical value of the shortest baseline to ensure no ambiguity value is proportional to the wavelength of the carrier signal. That is to say, if the system can effectively resolve ambiguity in the case of the shortest wavelength, then it can still effectively resolve ambiguity for signals with longer wavelengths. That is, if the system is applicable at the highest frequency, then it is also applicable to signals with slightly lower frequencies.

[0130] (2) In actual situations, there are many factors that cause errors, resulting in a certain probability that the system wrongly resolves ambiguity and obtains incorrect results. Therefore, what needs attention is how to improve the probability of ambiguity resolution in the presence of problems such as noise interference and channel consistency.

[0131] (3) Using a long baseline can reduce the mutual coupling effect of the antennas and can reduce the direction-finding error and improve the accuracy. Therefore, within the allowable range of the system size, the longer the baseline length, the better.

[0132] (4) To minimize errors during the ambiguity resolution process as much as possible, a multi-level ambiguity resolution method is adopted: screen and sort the distance differences between 4 array elements, and use the non-ambiguous and low-accuracy results of the first and second array elements (array element spacing is half a wavelength) as the initial values to resolve the multi-value ambiguity situation in the secondary direction finding (the distance between the second and third array elements is 2 times half a wavelength), and obtain an intermediate value (phase difference) with slightly higher accuracy. And so on, when performing four-level ambiguity resolution, we use the first and fourth array elements to reach the maximum baseline length to obtain the final value with the highest accuracy, and thus the spatial signal source angle can be obtained.

[0133] In summary, the flow chart for implementing the interferometer algorithm is as Figure 8 shown.

[0134] III. Signal sorting

[0135] Radar signal sorting is based on whether there is correlation between radar signal parameters. In an electronic warfare environment, radar signals received by radar reconnaissance receivers are usually retained in the form of pulse streams, and then the receiver converts the pulse stream into a radar pulse descriptor PDW that can be identified and calculated through a series of processes. Usually, the pulse descriptor PDW contains five parameters: pulse arrival time TOA, pulse amplitude PA, carrier frequency RF, pulse width PW, and pulse arrival angle DOA.

[0136] Traditional sorting algorithms mainly include extended association method, traditional histogram method, cumulative difference histogram method (CDIF), and sequence difference histogram method (SDIF). A single sorting algorithm will be constrained in the face of complex electromagnetic environments. Therefore, it is necessary to combine multiple sorting algorithms for improvement to achieve optimization.

[0137] After testing, the present invention proposes to use an adaptive K-means clustering algorithm combined with the SDIF algorithm to sort the signal, thereby achieving a better sorting effect. The adaptive K-means clustering algorithm is used to cluster the carrier frequency, bandwidth and pulse width of the sorted data, and the SDIF algorithm is used to perform PRI statistics and jitter signal discrimination for each clustering result. After obtaining the PRI result, the data results of the same cluster are checked for discrepancy, and finally the results of different clusters are checked for frequency agility, and finally the complete parameters of the radar are given.

[0138] (I) Adaptive K-means clustering algorithm

[0139] In the face of complex electromagnetic environments, it is impossible to predict how many radar signals there are, and the number of radars will change with the change of reconnaissance positions. Therefore, in signal sorting, presetting the number of cluster types will result in deviations, which will have a great impact on subsequent results. In the adaptive K-means clustering algorithm, cluster types are no longer limited, and are judged based on the data to be clustered. If there are N data to be clustered, 1 to N clusters may appear after clustering.

[0140] Since the initial cluster center has a significant impact on the final clustering result, if the cluster centers are too close, only a local optimal solution can be obtained, not a global optimal solution. Therefore, the data to be sorted is used as the initial cluster center, and data that is not within the range of all cluster centers is determined as a new cluster center.

[0141] Considering that carrier frequency, pulse width and bandwidth measurements have been widely used in hardware in practical applications, and the pulse arrival angle is greatly affected by the direction of the antenna array and the measurable angle range is limited, this paper uses pulse arrival time, carrier frequency, pulse width and bandwidth as pulse descriptors for the detection signal.

[0142] Assuming the detected pulse description word array is Θ, its expression is:

[0143] Θ = [P 1 , P 2 , … P N

[0144] where P is each pulse descriptor, and the expression is:

[0145] P j = [TOA j , PW j , BW j , RF j

[0146] During the clustering process, using PW, BW, and RF as criteria, the data to be sorted is classified. Similar to the traditional K - means clustering algorithm, the Euclidean distance is used to measure the difference between different data domains to be clustered and the clustering centers. The specific form is as follows:

[0147]

[0148] where i represents the i - th clustering center and j represents the j - th data to be classified.

[0149] In the calculation, due to the detection error differences caused by hardware precision and the different tolerances that different radar parameters can tolerate, different parameters need to be weighted so that the obtained Euclidean distance is comparable. The modified Euclidean distance is as follows:

[0150]

[0151] where α, β, and χ represent the weights of pulse width, carrier frequency, and bandwidth respectively.

[0152]

[0153] where μ represents the reliability coefficient, representing the stability of this variable; δ represents the tolerance. Generally, it is determined through multiple measurements, which are experimental parameters; τ represents the measurement precision, which is determined by the detection precision of each parameter by the hardware platform.

[0154] No longer presetting the clustering types, during the clustering process, if all the data is too discrete and the number of clustering centers is huge, each data to be sorted needs to be compared with all the clustering centers, resulting in a large amount of calculation and operation time. Setting termination conditions can well improve this phenomenon.

[0155]

[0156] where PW, BW, and RF represent the maximum tolerable fluctuation ranges within the same clustering range of pulse width, carrier frequency, and bandwidth. In practice, it is determined through multiple experiments according to the form of radar signals to be processed and the external environmental interference.​​

[0157] When d ij <d 0 , it is determined that this data to be clustered belongs to this cluster. The setting of d 0 , essentially limits the clustering radius, can reduce the step of determining whether the current clustering error converges in the traditional K-means clustering algorithm, simplifies the algorithm calculation amount. At the same time, in the face of a large number of radar signals in a complex electromagnetic environment, it can reduce the calculation amount and operation time caused by comparing the Euclidean distance with each clustering center one by one when there are a large number of clustering centers due to the existence of outliers.

[0158] After data is assigned to cluster X i , it is necessary to update the clustering center M i .

[0159]

[0160] The flowchart of the adaptive k-means clustering algorithm is as Figure 9 shown. The specific implementation steps of the adaptive k-means clustering algorithm are as follows:

[0161] (1) Initialize each array and input N data to be clustered;

[0162] (2) Let the data number to be clustered j = 1, the clustering type k = 1, and use the information of the first data to be clustered as the first clustering center;

[0163] (3) The data number to be clustered j = j + 1; i = 1, if j > N, go to (6);

[0164] (4) Calculate the weighted Euclidean distance d ij between the j-th data and the i-th clustering center;

[0165] (5) If d ij < d 0 , then assign this data to be clustered to the current cluster, update the current clustering center, and return to (3); otherwise i = i + 1; if i > k, k = k + 1, use the parameter information of the j-th data to be clustered as the k-th clustering center; return to (3), otherwise return to (4);

[0166] (6) Output the clustering result.

[0167] (II) Principle of SDIF algorithm

[0168] The Sequential Difference Histogram Method (SDIF) is improved based on the Cumulative Difference Histogram Algorithm (CDIF). The flowchart is as Figure 10 shown.

[0169] The SDIF algorithm mainly consists of two steps: PRI measurement and sequence search. Calculate the difference histograms of different orders without accumulating the results, and calculate the threshold. If there is only one threshold crossing in the first-order difference, perform sequence retrieval; otherwise, count the next level. If there are multiple-order differences, perform sequence retrieval on the PRIs that cross the threshold. End when exceeding five-order differences and perform jitter discrimination.

[0170] One of the most important parameters in the SDIF algorithm is the detection threshold. The detection threshold is related to the reliability and accuracy of the PRI value. The peak of the histogram is inversely proportional to the interval between two pulses. Within a finite sampling time, the pulse interval is inversely proportional to the number of pulses. Therefore, the threshold is directly proportional to the total number of pulses E and inversely proportional to the pulse interval τ, as shown in the formula.

[0171]

[0172] In the formula, x is a constant less than 1.

[0173] When detecting pulses from multiple radars within the sampling time, the adjacent pulse intervals are random events and follow the Poisson distribution. Divide the sampling time T into n pulse sub-intervals. In the time interval τ = t 2 -t 1 , the probability of k random Poisson occurrences is:

[0174]

[0175] Among them, represents the number of pulse sub-intervals within the sampling time, and the probability of the adjacent pulse interval τ is approximately P 0 (τ) = e -λτ , which is the approximate form of the first-order difference histogram. The histogram is actually an approximation of the probability distribution function of random events, and the higher-order difference histograms are in the form of exponential distributions. The number of pulse groups that make up the C-th order difference histogram is the occurrence of (E - C) events, and the Poisson flow parameter Then the optimal threshold function is:

[0176] T threshold (τ) = x(E - C)e -τ / kN

[0177] Among them, E is the total number of pulses, C is the order of the difference, k is a constant less than 1, N is the sampling time, and the constant x is determined by experiments. After a large number of experiments, take k = 0.35 and x = 0.5.

[0178] The SDIF algorithm is adopted as the main sorting algorithm. This method does not accumulate the statistical results of the time difference of arrival histograms of different orders, simplifying the operation. After a peak that meets the threshold appears in the current range difference, the next-level difference is no longer statistically analyzed. Whether to perform the next sequence search is judged based on the number of levels of the difference histogram and the possible number of PRIs, reducing the amount of computation.

[0179] (3) Staggered radar calibration

[0180] When the PRI values and jitter amounts of two or three sorted pulse trains are very close, staggered discrimination is required to determine whether they are a radar signal transmitting a staggered pulse signal or several independent radar signals with close PRIs and similar jitter amounts.

[0181] Suppose there is a two-staggered pulse train. Through the above sorting procedure, two independent pulse trains with equal PRIs will be sorted out. Similarly, two independent pulse trains with equal PRIs and similar jitter amounts overlapping together also have the form of a staggered pulse train. Therefore, to distinguish whether two similar pulses are staggered, differences need to be found from their fundamental attributes. If it is a staggered pulse train, the two pulses are generated by the same timer and their changes are correlated, while two independent pulse trains are not correlated. This difference is manifested as different variances. The variance of the staggered pulse train is very small, while the variance of the two independent pulse trains will be very large.

[0182]

[0183]

[0184] In the expression, B i and A i are the arrival times of the i-th pair of pulses.

[0185] For two independent pulse trains, their mean square error δ B-A is close to their jitter amount PRI, while for the staggered pulse train, δ B-A is much smaller than the pulse jitter amount. Therefore, 1 / 2 - 1 / 3ΔPRI is taken as the standard for staggered discrimination.

[0186] Taking a three-staggered radar signal as an example. A group of signals of a staggered radar signal with the same carrier frequency and pulse width, after carrier frequency pre-sorting and pulse width pre-sorting, the staggered radar signal is sorted into the same carrier frequency range and the same pulse width range. After the SDIF algorithm, three signals with the same PRI, the same pulse width, and the same carrier frequency will be sorted out. The TOA data of these three signals satisfy that the TOA of the signal in the front is always before the TOA of the signal in the back, and the interval between them is equal, which is the sub-PRI.

[0187] When entering the staggered signal inspection program, all sequence differences and corresponding data are input. The data includes arrival time, pulse width, carrier frequency, and amplitude. Take the first sequence difference of the PRI sequence difference (mi = 1) and compare it with the remaining sequence differences (mi + nn) one by one to detect whether there is the same PRI. After searching for the same PRI, detect whether the pulse widths of the two are the same. If the pulse widths are the same, it is preliminarily determined as a possible staggered radar signal. Sort the TOA data of the two in ascending order, take the first difference of the TOA, search for the first difference less than the PRI, and record the corresponding position. Find the first difference before the mi-th sequence difference and determine it as the potential sub-PRI. According to the sub-PRI, verify whether the data satisfies that the TOA data of the mi-th is in the front and the data of the mi + nn-th is in the back. If it is satisfied, it is determined as a staggered radar signal. Otherwise, proceed to detect the next group of signals.

[0188] (4) Frequency Agile Radar Calibration

[0189] Frequency agile radar is a common form of radar. In this system, frequency agile radar will be divided into different clusters during the clustering stage. After each clustering ends, frequency agility calibration is performed on the results of different clusters.

[0190] Regardless of inter-pulse agility or pulse-group agility, in different clusters, the carrier frequencies are different, and other parameters such as pulse width, bandwidth, and pulse repetition period tolerance are the same.

[0191] Taking a 5-frequency agile radar signal as an example to elaborate on the pulse repetition period of the final inter-pulse agile radar signal. According to the improved K-Means algorithm, the inter-pulse agile radar signal sequence will be divided into five different clusters. In different clusters, the pulse repetition period, pulse width, and bandwidth parameters are the same within the tolerance range, and the only difference is the different carrier frequencies.

[0192] Search for five PRI sequences with the same pulse repetition period, pulse width, and bandwidth tolerance range and different carrier frequencies in different clusters, and verify the corresponding data of these sequences to obtain the radar signal pulse repetition period corresponding to the inter-pulse frequency agile radar signal. The PRI of the inter-pulse frequency agile radar signal is as follows:

[0193]

[0194] Among them, PRI ture represents the pulse repetition period of the inter-pulse agile radar signal, and PRI represents the PRI obtained from different clusters.

[0195] (5) Jitter Radar Signal Analysis

[0196] The PRI of the jitter signal varies within the range of 10% - 15% of the central value. The carrier frequency and pulse width of the jitter signal are the same. Therefore, after carrier frequency pre-sorting and pulse width pre-sorting, the jitter signals will be sorted into the same pulse width interval within the same carrier frequency interval. Since the PRIs of the jitter signals are different, the sequence difference cannot be obtained through the SDIF algorithm. Therefore, the sorting of the jitter signals is carried out on the remaining data after the SDIF sorts out the sequence difference.

[0197] According to the same pulse width of the jitter signal, organize the TOA data remaining after SDIF for each pulse width pre-sorting interval, calculate the first-order difference, and find the average value of the first-order difference. Search for the maximum / minimum value of the first-order difference. If the maximum / minimum value is not within the 15% jitter range of the average value, delete the corresponding maximum / minimum value, re-calculate the average value of the remaining first-order difference, and continue to compare whether the maximum / minimum value is within the jitter interval of the average value. If the maximum / minimum value is within the jitter interval of the average value and the length of the remaining first-order difference is greater than 5, it is considered that a jitter signal is successfully sorted out, and the central value (PRI), pulse width, carrier frequency, amplitude, and PRI tolerance of the jitter signal are given.

[0198] (7) Signal sorting using the combined adaptive K-means clustering algorithm and the SDIF algorithm

[0199] Through the above analysis, the present invention finally adopts the combined adaptive K-means clustering algorithm and the SDIF algorithm for signal sorting. The specific steps are as Figure 11 shown, including:

[0200] (1) Initialization, input all data to be sorted; set the number of clustering categories D = 0;

[0201] (2) Use the improved K-means clustering algorithm to cluster all data according to pulse width, carrier frequency, and bandwidth, and the number of clustering results is D;

[0202] (3) Read each clustering data one by one, and use the optimized SDIF algorithm to calculate the sequence difference for the data in each clustering;

[0203] (4) If the number of PRIs obtained is greater than 1, perform the same signal discrimination and stagger check on these PRIs, and go to (3); if the number of PRIs is less than 2, go to (5);

[0204] (5) If there are still unprocessed clusters, go to (3), otherwise go to (6);

[0205] (6) Perform frequency agile radar check on the PRIs of different clustering results;

[0206] (7) Output the sorting result.

[0207] The situation awareness system proposed by the present invention can be well applied to the radar reconnaissance system, which is simple, efficient, with low time complexity and space complexity. It solves the problems of poor clustering effect of the clustering algorithm in the current complex and dense environment and serious overlap of characteristic parameters, and can only obtain local optimality, etc. It provides a better solution for signal recognition of the reconnaissance system in the complex electromagnetic environment.

[0208] The above is the preferred embodiment of the present invention. It should be pointed out that for those of ordinary skill in the art, without departing from the principle of the present invention, several improvements and modifications can be made, and these improvements and modifications should also be regarded as the protection scope of the present invention.

Claims

1. A passive electromagnetic situational awareness system based on a digital channel model, comprising: A signal receiving module, used for receiving radar signals of N channels; A channelization module, used for channelizing the radar signals of the N channels to form N channelized signals; A parameter measurement module, configured to select one of the N channelized signals as a first channelized signal, form a synchronization pulse according to the first channelized signal, perform pulse parameter measurement according to the synchronization pulse, and obtain parameter measurement information; A signal direction finding module, used to extract the instantaneous phases of other N-1 channelized signals other than the first channelized signal to obtain angle information; A signal sorting module, used to analyze the parameter measurement information and the angle information by using an adaptive K-means clustering algorithm combined with an SDIF algorithm, and output a radar signal sorting result; When the signal sorting module uses the adaptive K-means clustering algorithm combined with the SDIF algorithm to analyze the parameter measurement information and the angle information, the following steps are completed: (1) Initialization: input all the data to be sorted; set the cluster type D = 0; (2) Use the improved K-means clustering algorithm to cluster all data according to pulse width, carrier frequency, and bandwidth, and the clustering result type is D; (3) Read each cluster data one by one, and use the optimized SDIF algorithm to calculate the sequence difference of each cluster data; (4) If the number of PRIs is greater than 1, perform same signal discrimination and variance check on these PRIs and go to (3); if the number of PRIs is less than 2, go to (5); (5) If there are still unprocessed clusters, go to (3), otherwise go to (6); (6) Perform frequency agile radar calibration on PRIs with different clustering results; (7) Output the sorting results.

2. A passive electromagnetic situational awareness system based on a digital channel model according to claim 1, characterized in that: The parameter measurement module processes radar signals of N channels by using a decimation-forward uniform polyphase filter bank to obtain N channelized signals.

3. A passive electromagnetic situational awareness system based on a digital channel model according to claim 1, characterized in that: The parameter measurement module performs processing in conjunction with the instantaneous frequency and the video pulse to make a decision on the channel.

4. The passive electromagnetic situational awareness system based on a digital channel model according to claim 1, characterized in that: The signal direction finding module uses a multi-baseline phase interferometer algorithm to achieve signal direction finding.

5. The passive electromagnetic situational awareness system based on a digital channel model according to claim 1, characterized in that: The signal direction finding module adopts a multi-baseline defuzzification algorithm to sequentially remove the multi-value fuzziness of the long baseline from the short baseline to the maximum baseline.

Images