CN107238817B - Radar radiation source signal sorting method with parameter self-adaptive setting and automatic adjustment - Google Patents

Abstract

The invention discloses a radar radiation source signal sorting method with adaptive parameter setting and automatic adjustment. The method includes firstly, combining pulse arrival angles in selection of feature vectors and simultaneously introducing intra-pulse features, namely symmetrical Holder coefficients, for sorting, then setting expected number of classes by utilizing a method for searching histogram peak value statistics based on an iterative self-organizing data analysis algorithm, wherein the algorithm is abbreviated as ISODATA, the upper limit of standard deviation of each data distribution in ISODATA algorithm classes and the lower limit of minimum distance between various centers are preliminarily set according to the features of the feature vectors, and finally, the upper limit of the standard deviation of each data distribution in the classes and the lower limit of the minimum distance between various centers are controlled and set with an adaptive criterion when the ISODATA algorithm runs, so that the purpose of accurate final sorting is achieved, and the problem of signal sorting is solved through the method.

Description

A radar radiation source signal sorting method with parameter adaptive setting and automatic adjustment

technical field

The invention relates to a method for sorting radar radiation source signals.

Background technique

With the rapid development of modern radar technology, especially the wide application of various new systems of radar, the electromagnetic environment of the battlefield is increasingly complex and changeable. The boundary of the feature space described by the basic parameters is seriously overlapped, and the radiation source signal sorting based on the traditional five basic parameters basically cannot meet the actual needs.

The intrapulse feature analysis method opens up a new research idea for reducing the overlap of feature spaces, and it makes the idea of extracting new signal parameters to further fully characterize the signal characteristics of radar radiation sources a reality.

The advantage of the ISODATA algorithm is that through the process of gradual differentiation, it can recognize the essential attributes that most of the objects to be classified are unlikely to show at the beginning of the classification. It determines the final number of clusters according to the differences of the samples, which makes the classification more scientific. . However, the ISODATA algorithm also has its obvious shortcomings: the preset parameters of the traditional ISODATA clustering algorithm are often set based on experience, while in the electromagnetic environment of the modern battlefield, there are many types of radar radiation source signals, which are unpredictable, and it is impossible to obtain a priori information. The parameters of the algorithm are accurately preset, resulting in unsatisfactory clustering results.

Therefore, the present invention supplements the intra-pulse characteristic parameter, the symmetrical Holder coefficient, into the sorting parameters, and reasonably realizes the sorting of radar radiation source signals by adaptively adjusting the ISODATA algorithm.

SUMMARY OF THE INVENTION

In order to solve the problem of accurate sorting of radar radiation source signals, the invention discloses a radar radiation source signal sorting method with parameter adaptive setting and automatic adjustment. The eigenvectors used in this method are the pulse arrival angle and the symmetrical Holder coefficient. It is based on an iterative self-organizing data analysis algorithm, which is referred to as ISODATA for short. First, according to the characteristics of the eigenvectors themselves, the method of searching the peak statistics of the histogram is used to set the expectation. The upper limit of the standard deviation of each data distribution within the class and the lower limit of the minimum distance between various centers are preliminarily set. When the algorithm runs, the upper limit of the standard deviation of each data distribution within the class and the minimum distance between various centers are set. The lower limit control of the distance sets the adaptive criterion, in this way the signal sorting problem is solved.

The detailed technical measures steps of the present invention are as follows:

The first step: the selection of feature vectors

Select the radar radiation source signal pulse arrival angle and symmetrical Holder coefficient to form the eigenvector.

(1) Extraction of pulse arrival angle of radar radiation source

Extract the pulse of the radar signal pulse stream, and obtain the pulse arrival angle in the pulse description word of the radar signal pulse stream.

(2) Extraction of symmetrical Holder coefficients

The pulses of the radar signal pulse stream are extracted, and the extracted signals are preprocessed, mainly including the conversion of the frequency domain of the signal, and the normalization of bandwidth and energy.

The spectrum of the pulse and the symmetrical Holder coefficients _HU and _HT of the rectangular signal and the triangular signal are calculated according to the following equations.

(1) Symmetric Holder coefficient with rectangular signal U(k)

(2) Symmetric Holder coefficient with triangular signal T(k)

In the formula, F(i) is the frequency spectrum of the signal, and p=5 and q=5 are selected at the same time.

Construct feature vectors B=(DOA, H _T ) and D=(DOA, H _T , H _U ), where B is used for searching histogram peak statistics and D is used for final radiation source signal sorting.

Step 2: Search for a method of histogram peak statistics to estimate the approximate number of clusters

After obtaining the feature vector B, set the expected number of classes of the ISODATA algorithm by searching the peak statistics of the histogram.

Denote the feature vector B={B _i }={B ₁ ,B ₂ ,...,B _N }(i=1,2,...,N), and the dimension of the feature vector is 2 dimensions.

The division interval is determined by the sample size. Let the number of division intervals in each dimension be q,

表示不超过·的最大整数。In the formula,

Represents the largest integer not exceeding ·.

Denote DOA _max , DOA _min , _HTmax , and _HTmin as the maximum and minimum values of DOA and _HT in the feature vector, respectively.

Then the vector DOA dimension interval width DOA _Δ and the vector H _T dimension interval width H _TΔ are respectively

So far, the plane area where the feature vector B is located is divided into Q=q ² small intervals, and each small interval can be expressed as

{b _ij =(DOA _i ,H _Tj )|DOA _i ∈[DOA _m ,DOA _m+1 ],H _Tj ∈[H _Tn ,H _Tn+1 ],m,n∈[1,q]} (8 )

In the formula, DOA _m =DOA _min +(m-1)DOA _Δ , and H _Tn =H _Tmin +(m-1) H _TΔ .

Then separately count the number of vector elements that fall in different cell intervals to the bar height of the interval statistic value. Compare the statistic value in each well-divided interval with the adjacent interval, and if the statistic value of the interval is greater than the statistic value of all its adjacent intervals, record the interval as a peak interval. The final number of statistical peak intervals is the number of peaks S.

The preset number of classes is set to the number of histogram peaks, i.e.

表示超过·的最小整数。In the formula,

Represents the smallest integer that exceeds ·.

Step 3: Pre-setting the upper limit of the standard deviation of each data distribution within the class and the lower limit of the minimum distance between the centers of each category

The feature vector D is obtained, and the upper limit θ _S of the standard deviation of each data distribution within the class and the lower limit θ _D of the minimum distance between various centers are preset.

First perform attribute normalization for each dimension in D:

为第k维中最大的元素，

为第k维中最小的元素；In the formula: x _ij ' is the sample data after normalization, x _ij is the sample data before normalization,

is the largest element in the kth dimension,

is the smallest element in the kth dimension;

In the formula: u _j is the center of the jth dimension of the attribute normalized sample data _xi ' _j .

Continue to find the distribution sparsity of the jth dimension of the sample data:

λj is the normalized deviation of the _jth dimension in the sample data after normalization.

Set the initial value from this:

where S is the expected number of classes;

Then directly find the variance of the jth dimension of the sample data itself

Set the initial value from this

Step 4: Run the ISODATA algorithm for sorting according to the adaptive criteria

For the operation of the whole algorithm, the self-adaptive principle is adopted, that is, by setting the upper limit θ _S of the standard deviation of each data distribution within the class and the lower limit θ _D of the minimum distance between various centers, the optimal clustering effect can be automatically approached. Purpose.

First calculate the intra-class distance when all sample data are assumed to be of the same class

Run the ISODATA algorithm after setting the parameters.

After each clustering algorithm is finished, the intra-class distance Θ _S ' and the inter-class distance Θ _D ' of the extraction results are calculated respectively.

In the formula: c' is the number of classifications generated by the algorithm operation, C _p is the set of classified data, vi is the center of the _i -th class, x _n is the data belonging to the C _p class, and v _j is the class different from vi _. center of.

After the first clustering, only θ _S is changed, and the adaptive principle is set as follows

Run the ISODATA algorithm again

And set the adaptive criteria as follows:

The conditions for exiting the adaptive iteration are:

Δ _s = |θ _s '-θ _s |=0 (21)

Δ _D =|θ _D '-θ _D |=0 (22)

In the formula: _ΔS and _ΔD are the difference between the intra-class distance and the inter-class distance between the results of two consecutive runs before and after the algorithm.

After exiting the adaptive iteration, the sorting is completed, and the sorting results are counted.

So far, the whole process of a radar radiation source signal sorting method for parameter self-adaptive setting and automatic adjustment ends.

The present invention has the following beneficial effects:

①The sorting accuracy is high. In the present invention, the relatively accurate setting of the expected number of classes and the setting of the self-adaptive criterion can improve the accuracy rate. In addition, the intra-pulse feature of the symmetrical Holder coefficient is introduced, which greatly increases the separability of the feature vector, so the sorting effect is improved. relatively ideal.

Description of drawings

Fig. 1 is the flow chart of the radar radiation source signal sorting method based on adaptive adjustment;

Figure 2 is a distribution diagram of radar sample data;

Fig. 3 is the statistical effect diagram of the peak value of the search histogram;

Figure 4 is an ideal classification diagram;

Figure 5 shows the effect of algorithm sorting.

Detailed ways

The present invention will be described in further detail below with reference to the accompanying drawings and embodiments.

Using the method shown in Figure 1, five radar radiation source signals with different modulation types and relatively close parameters are selected for simulation experiments. The variation types and value ranges of working parameters are shown in Table 1.

Table 1 Radar radiation source signal characteristic parameter table

(1) Selection of eigenvectors

Feature vector extraction is performed on the radar radiation source signal, and 500 sample data are generated for each type of radar.

(2) The method of searching histogram peak statistics to estimate the approximate number of clusters

Determine the number of divisions based on the number of sample data

The sample data is distributed on the plane as shown in Figure 2, and then through the search for the peak value of the histogram, the peak value of the histogram is obtained as shown in Figure 3, and the number of peaks after analysis is

S=5

the default number of categories

(3) Presetting the upper limit of the standard deviation of each data distribution within the class and the lower limit of the minimum distance between the centers of each class

Continue to find the distribution sparsity of the jth dimension of the sample data:

get

λ ₁ =0.2330, λ ₂ =0.2248, λ ₃ =0.2596

Set the initial value from this:

Then directly find the variance of the jth dimension of the sample data itself

get

σ ₁ =1.7908, σ ₂ =1.9398, σ ₃ =2.6530

(4) According to the self-adaptive principle, run the ISODATA algorithm to classify

1) Calculate the intra-class distance when all sample data are of the same class

2) Set basic parameters

Minimum number of data allowed in each category: θ _n = 4;

Maximum number of logarithms of classes that can be merged while iterating: L=2

The maximum number of iterations allowed: I=70;

Expected number of classes: C=8;

The upper limit of the standard deviation of each data distribution within the class: θ _s =0.35816;

Lower bound of the minimum distance between different class centers: θ _D =1.124.

2) Run the ISODATA algorithm

Calculate intra-class distance and inter-class distance after getting the first result

_Θs = 1067.1

Θ _D = 2.4872

Run the algorithm again after changing the upper limit of the standard deviation of each data distribution within the class

_Θs = 1067.1

Θ _D = 2.4872

Run the algorithm again after changing the parameters

_Θs = 1067.1

Θ _D = 2.4872

at this time

Δ _s = |θ _s '-θ _s |=0

Δ _D =|θ _D '-θ _D |=0

Exit the algorithm and get the sorting results as shown in Table 2. The actual distribution of sample data is shown in Figure 4, and the method classification is shown in Figure 5. Different labels in the figure represent different classes.

Table 2 Simulation results

Claims (4)

1. a radar radiation source signal sorting method of parameter self-adaptive setting and automatic adjustment, it is characterized in that the steps of the method are as follows: ①Select the radar radiation source signal pulse arrival angle and the symmetrical Holder coefficient to form the eigenvector, in which the pulse arrival angle and the signal spectrum and the symmetrical Holder coefficient of the triangular signal form the eigenvector of the peak statistics of the search histogram; the pulse arrival angle, the signal spectrum and the rectangular signal , The symmetrical Holder coefficient of the triangular signal constitutes the sorting eigenvector; ② The expected number of classes determined by the method of searching histogram peak statistics, the expected number of classes is 1.5 times the number of peaks; ③ Pre-set the upper limit θ _S of the standard deviation of each data distribution within the class and the lower limit θ _D of the minimum distance between various centers in the ISODATA algorithm, specifically: First normalize the attributes of each dimension in the sample data:

In the formula: x _ij ' is the sample data after normalization, x _ij is the sample data before normalization,

is the largest element in the kth dimension,

is the smallest element in the kth dimension;

In the formula: u _j is the center of the jth dimension of the normalized sample data x _ij '; Continue to find the distribution sparsity of the jth dimension of the sample data:

λj is the normalized deviation of the _jth dimension in the sample data; This sets the initial value of the lower limit θ _D of the minimum distance between various centers: where S is the expected number of classes; Then directly find the variance of the jth dimension of the sample data itself This sets the initial value of the upper limit θ _S of the standard deviation of each data distribution within the class:

④Run the ISODATA algorithm for sorting according to the adaptive criteria; The self-adaptive criterion is to achieve the purpose of automatically approximating the best clustering effect by setting the upper limit of the standard deviation of each data distribution within the class and the lower limit of the minimum distance between various centers. First calculate the intra-class distance when all sample data are assumed to be of the same class as

After setting the parameters, run the ISODATA algorithm, After each clustering algorithm is finished, calculate the intra-class distance Θ _S ' and the inter-class distance Θ _D ' of the extraction algorithm results, respectively,

In the formula: c' is the number of classifications generated by the algorithm operation, C _p is the set of classified data, vi is the center of the _i -th class, x _n is the data belonging to the C _p class, and v _j is the class different from vi _. center of After the first clustering, only θ _S is changed, and the adaptive principle is set as follows

Run the ISODATA algorithm again after changing the parameters And set the adaptive criteria as follows:

The conditions for exiting the adaptive iteration are: Δ _S =|θ _s '-θ _s |=0 Δ _D =|θ _D '-θ _D |=0 In the formula: _ΔS and _ΔD are the difference between the intra-class distance and the inter-class distance between the results of two consecutive runs before and after the algorithm. 2 . The method for sorting radar radiation source signals according to claim 1 , wherein the signal spectrum in step ① is specifically: a spectral line diagram generated after fast Fourier transform of the received signal. 3 . 3. radar radiation source signal sorting method according to claim 1, is characterized in that, the triangular signal and rectangular signal in described step 1. are specifically: choose the length of triangular signal and rectangular signal and should match the length of signal spectrum . 4. The method for sorting radar radiation source signals according to claim 1, wherein the search histogram peak statistics in the step 2. are specifically: each element in a vector matrix X _{N × M} is shown in the figure. It is described in the form of a bar. If the number of columns of the vector matrix is 1, the obtained histogram is a two-dimensional graph; if the number of columns is 2, the obtained histogram is a three-dimensional graph, and then the histogram is extended to the columns of the vector matrix. When the number is M (M≥3), the obtained histogram is a graph of M+1, and then the bar height of the interval statistical value can be obtained, and the statistical value in each divided interval is compared with the adjacent interval, If the statistic value of this interval is greater than the statistic value of all its adjacent intervals, the interval is recorded as a peak interval; the number of the last statistical peak interval is the number of peaks.

Images