CN110188647A

CN110188647A - One kind being based on the feature extraction of variation mode decomposition Radar emitter and its classification method

Info

Publication number: CN110188647A
Application number: CN201910433811.3A
Authority: CN
Inventors: 罗明; 付亮; 夏伟; 斯海飞
Original assignee: Xidian University
Current assignee: Xidian University
Priority date: 2019-05-23
Filing date: 2019-05-23
Publication date: 2019-08-30

Abstract

The invention discloses one kind to be based on the feature extraction of variation mode decomposition Radar emitter and its classification method, is firstly received the high-frequency pulse signal of radar, pre-processes to high-frequency pulse signal, obtains preprocessed signal；The pulse descriptive word parameter for generating preprocessed signal is handled using digital channelizing, and extracts angle of arrival and pulse width in pulse descriptive word parameter；And the characteristic parameter of Radar emitter is formed as a kind of supplement of conventional parameter by the centre frequency intrapulse feature of the method preprocessed signal of variation mode decomposition, finally Radar emitter is sorted using the method for fuzzy-C mean value.This method can classify to frequency modulation(PFM) class Radar emitter, have good noise inhibiting ability, can effectively improve the classifying quality of Radar emitter under low signal-to-noise ratio.

Description

Radar radiation source feature extraction and classification method based on variational modal decomposition

Technical Field

The invention relates to the technical field of radars, in particular to a method for extracting and classifying radar radiation source features based on variational modal decomposition, which is applicable to classification of radar radiation sources under the condition of low signal-to-noise ratio.

Background

The radar radiation source identification is an important component in an electronic reconnaissance system, battlefield information and information are provided for the military operations of the own party by intercepting, sorting, identifying and positioning the radar radiation source of an enemy, and the number and quality of enemy information obtained by radar reconnaissance determine the trend of a war office.

In order to ensure that radar reconnaissance equipment has higher interception probability of enemy radiation source signals, parameters of a radar Pulse sequence are obtained in a receiving mode with wide Frequency domain in the sorting process, wherein the parameters mainly comprise Carrier Frequency (CF), Pulse Width (PW), Pulse Amplitude (PA), Time of Arrival (TOA) and Angle of Arrival (AOA), and Pulse Description Words (PDW) are formed by the parameters. However, in the present electromagnetic environment, the electromagnetic signal density is large, and the complex system is variable, and the defects of the method become more obvious.

The intra-pulse characteristics are important pulse parameters of a radar radiation source, the aliasing degree of a parameter space is reduced, a richer characteristic space can be obtained to improve the recognition rate, intra-pulse characteristic analysis of radar signals is a very important research direction at present, and a plurality of functions based on intra-pulse characteristic analysis are added to a new generation of electronic reconnaissance system on the basis of reserving the original PDW function so as to meet the battlefield requirements.

Disclosure of Invention

Aiming at the problems in the prior art, the invention aims to provide a method for extracting and classifying radar radiation source features based on variational modal decomposition, which can classify frequency modulation radar radiation sources, has good noise suppression capability and can effectively improve the classification effect of the radar radiation sources under low signal-to-noise ratio.

In order to achieve the purpose, the invention is realized by adopting the following technical scheme.

The method for extracting the radar radiation source features based on the variational modal decomposition comprises the following steps:

step 1, a reconnaissance receiver receives a high-frequency pulse signal of a radar in real time and preprocesses the high-frequency pulse signal to obtain a preprocessed signal s (t);

step 2, generating pulse description word parameters of the preprocessed signals s (t) by utilizing digital channelization processing, and extracting the arrival angle theta in the pulse description word parameters_AOAAnd pulse width τ_PWAnd is ready for use;

step 3, carrying out mirror extension on the preprocessed signal s (t), carrying out Fourier transform on the preprocessed signal after mirror extension to obtain a frequency spectrum f (omega) after Fourier transform, and reserving a positive half shaft of the frequency spectrum after Fourier transform to obtain a positive half shaft radar signal frequency spectrum

Step 4, carrying out frequency spectrum on the positive half shaft radar signalCarrying out variation modal decomposition to obtain the central frequency omega of the kth modal of the radar signal_k。

(II) a radar radiation source feature classification method based on variational modal decomposition, which comprises the following steps:

step 1, extracting the central frequency omega of the kth mode of a radar radiation source_kAngle of arrival theta_AOAAnd pulse width τ_PWThe classification characteristic V ═ ω of the constituent radar radiation sources₁,…,ω_k,…,ω_K,θ_AOA,τ_PW]；

Each radar radiation source generates A parameter samples, and Z radar radiation sources generate M samples as a sample set; m ═ AZ;

step 2, setting iteration times L, an iteration termination factor epsilon, a clustering number c, a weighting index m and a maximum iteration time L of fuzzy clustering; initializing a center matrix V of cluster centers⁽⁰⁾＝{v₁,v₂,…,v_i,…,v_c}，i＝1,2,…,c；

And 3, fusing the characteristics X of each sample in the sample set to obtain a data set X to be sorted { P ═ P }₁,P₂,…,P_j,…,P_M}，j＝1,2,…,M；P_j＝[ω₁,…,ω_k,…,ω_K,θ_AOA,τ_PW]^T；

And 4, starting iteration, and initializing a membership matrix U-U by using a random number under the condition of meeting constraint conditions_ij}; wherein the element U in the matrix U_ij(0<u_ij<1) Denotes the jth P_jBelongs to the ith radar radiation source clustering center v_iDegree of membership of;

step 5, iteratively updating the membership matrix U, and calculating and updating a clustering center;

step 6, calculating a target function according to the membership matrix U and the clustering center, stopping iteration when the target function is smaller than epsilon or the iteration frequency L reaches the maximum value L, and returning to the step 4 if not;

and 7, iterating to obtain the Z class centers and the membership values of the M samples to the Z class centers.

Compared with the prior art, the invention has the beneficial effects that:

the invention relates to a radar radiation source characteristic extraction and classification method based on variational modal decomposition, which comprises the steps of firstly receiving a high-frequency pulse signal of a radar, and preprocessing the high-frequency pulse signal to obtain a preprocessed signal; generating pulse description word parameters of the preprocessed signals by utilizing digital channelization processing, and extracting an arrival angle and a pulse width in the pulse description word parameters; and preprocessing the characteristics in the central frequency pulse of the signal by a variational modal decomposition method, using the characteristics as a supplement of conventional parameters to form characteristic parameters of the radar radiation source, and finally sorting the radar radiation source by adopting a fuzzy-C mean value method. The method can classify the frequency modulation radar radiation sources, and simulation experiments show that the accuracy of the classification of radar radiation source signals reaches more than 97 percent.

Drawings

The invention is described in further detail below with reference to the figures and specific embodiments.

FIG. 1 is a flow chart of a method for extracting and classifying radar radiation source features based on variational modal decomposition according to the present invention;

FIG. 2 is a time domain diagram of a signal and a modal diagram after a variation modal decomposition; wherein, fig. 2(a) is a time domain waveform of a radar pulse signal received by a receiver; FIG. 2(b) is a comparison graph of two modes after radar signal variation mode decomposition and an original signal;

fig. 3 is a central frequency distribution diagram extracted after signal variation modal decomposition.

Detailed Description

Embodiments of the present invention will be described in detail below with reference to examples, but it will be understood by those skilled in the art that the following examples are only illustrative of the present invention and should not be construed as limiting the scope of the present invention.

step 1, a reconnaissance receiver receives a high-frequency pulse signal of a radar in real time and preprocesses the high-frequency pulse signal to obtain a preprocessed signal s (t).

Specifically, the pretreatment method in step 1 comprises the following steps: converting the high-frequency pulse signal into an intermediate-frequency pulse signal through a low-pass filter, and then amplifying and matching filtering the intermediate-frequency pulse signal by a multi-stage intermediate-frequency amplifier to obtain an amplified intermediate-frequency signal; and carrying out A/D sampling on the amplified intermediate frequency signal to obtain a preprocessed signal s (t).

Step 2, generating pulse description word parameters of the preprocessed signals s (t) by utilizing digital channelization processing, and extracting the arrival angle theta in the pulse description word parameters_AOAAnd pulse width τ_PWAnd then standby.

Step 3, carrying out image continuation on the preprocessed signal s (t), and suppressing the harmonic problem of the signal through the image continuation; fourier transform is carried out on the preprocessed signal after the mirror image extension, the signal is converted into a frequency domain, and a frequency spectrum f (omega) after the Fourier transform is obtained; and clearing the negative half shaft after Fourier transform, and reserving the positive half shaft of the frequency spectrum after Fourier transform to obtain the positive half shaft radar signal frequency spectrumThe formula of the spectrum f (ω) after fourier transform is as follows:

wherein j is an imaginary number, ω is the frequency of the radar signal, and t is the signal time domain.

Step 4, carrying out frequency spectrum on the positive half shaft radar signalCarrying out variation modal decomposition to obtain the central frequency omega of the kth modal of the radar signal_k(ii) a The goal of the variational modal decomposition is to decompose the signal into Intrinsic Mode Functions (IMFs), which are redefined as an am-fm signal.

Specifically, in step 4, the variational modal decomposition includes the following substeps:

in substep 4.1, the eigenmode functions of the variational modal decomposition are as follows:

u_k＝A_kcos(φ_k)；ω_k＝φ′；

wherein u is_kRepresenting an amplitude of A_kFrequency of omega_kA harmonic signal of (a), which has a certain sparseness property reproducing the original signal; a. the_kRepresents u_kInstantaneous amplitude of (e), omega_kIs u_kThe instantaneous frequency of (c).

Substep 4.2, setting a parameter penalty factor α of variational modal decomposition, a decomposition layer number K, a noise tolerance zeta, an iteration number N and a termination condition epsilon;

substep 4.3, where n denotes the nth iteration, k denotes the kth mode, the number of iterations n is set to 1 and the number of modes k is set to 1, and the first mode is initializedSecond modeCenter frequency of first modeCenter frequency of the second modeInitializing lagrange multiplications

(a) The first modality is updated according to the following formula:

wherein,for the first modality after the first iteration,the method is characterized in that the frequency spectrum of the radar signal is a positive half-axis, α is a penalty factor, omega is the frequency of the radar signal, the essence of the method is wiener filtering, the method has a good noise reduction effect on the signal, and h is an intermediate variable used for eliminating other modes of the current mode to sum.

(b) The center frequency of the first mode is updated according to the following formula:

wherein,a center frequency of a first iteration for a first mode;

substep 4.4, increment k by 1, get k 2, update the second mode and its center frequency, and the formula is as follows:

wherein,for the modality after the first iteration of the second modality,in the second mode

The center frequency of one iteration.

Substep 4.5, analogizing in turn, increasing k-1 by 1 to obtain the kth mode and the center frequency thereof, wherein the formula is as follows:

wherein,for the mode after the first iteration of the kth mode, ω_k ²The center frequency of the first iteration for the kth mode; obtaining the mode after the first iteration of the Kth mode until K reaches the preset maximum decomposition layer number KAnd the center frequency ω of the first iteration of the Kth mode_K ²。

Substep 4.6, updating the lagrangian operator according to the following formula:

wherein,ζ is the noise tolerance for the lagrangian updated after the first iteration.

Substep 4.7, increasing the iteration number n by 1 to obtain n equal to 2, and repeating substeps 4.3-4.6 to obtain the mode and the center frequency of the mode after the second iteration of the kth mode, which are specifically as follows:

wherein,for the mode after the second iteration of the kth mode, ω_k ³The center frequency of the second iteration for the kth mode.

And substep 4.8, analogizing in turn, increasing n-1 by 1 to obtain the lagrangian operator, the mode and the central frequency thereof after the nth iteration of the kth mode, which are specifically as follows:

wherein,for the lagrangian updated after the nth iteration,for the mode after the nth iteration of the kth mode, ω_k ⁿ⁺¹Is the center frequency of the kth mode after the nth iteration, i.e. the center frequency omega of the kth mode of the radar signal_k；

And a substep 4.9, if the iteration times N are sequentially increased by 1, judging whether the iteration times N reach the maximum value N or whether the error is smaller than the set termination condition epsilon.

If the iteration number N reaches the maximum value N or the error is smaller than the set termination condition epsilon, stopping the iteration; otherwise, substeps 4.3-4.8 are continued.

Specifically, the formula for the error less than the set termination condition epsilon is as follows:

wherein | | xi | purple₂Is a 2-norm.

Substep 5.0 of centering the center frequency ω of the kth mode of the radar signal_kAs an intra-pulse feature of the signal.

And 5, extracting characteristic parameters of the radar radiation source, specifically as follows:

extracting the center frequency omega of the kth mode of a radar radiation source_kAngle of arrival theta_AOAAnd pulse width τ_PWThe classification characteristic V ═ ω of the constituent radar radiation sources₁,…,ω_k,…,ω_K,θ_AOA,τ_PW](ii) a Each radar source produces a parametric samples and Z radar sources produce M samples as a sample set, where M is AZ.

And 6, fuzzy clustering is to introduce fuzzy mathematics into clustering analysis, and the established fuzzy relation can be more objectively and accurately clustered.

Substep 6.1, firstly, key parameter initialization is carried out: iteration times L, an iteration termination factor epsilon, the number c of clusters, a weighting index m and the maximum iteration times L. Initializing a center matrix V of cluster centers⁽⁰⁾＝{v₁,v₂,…,v_i,…,v_c}，i＝1,2,…,c。

Substep 6.2, performing fusion processing on the characteristics X of each sample in the sample set to obtain a data set to be sortedX＝{P₁,P₂,…,P_j,…,P_M}，j＝1,2,…,M；

Wherein, P_j＝[ω₁,…,ω_k,…,ω_K,θ_AOA,τ_PW]^T，。

Substep 6.3, the iteration is started, the iteration number l is set to 1, and the membership matrix U is initialized by random numbers under the condition that the constraint condition is met_ij}; wherein the element U in the matrix U_ij(0<u_ij<1) Denotes the jth P_jBelongs to the ith radar radiation source clustering center v_iDegree of membership of; the constraint is shown as follows:

wherein u is_ij∈[0,1]Represents each P_jDifferent degrees of membership for different radar radiation sources.Represents each P_jThe sum of the membership degrees for the different radar radiation sources equals 1. Each iteration is unrelated to the last iteration and the iteration is restarted with l.

And substep 6.4, updating the membership matrix U after the first iteration according to the following formula:

wherein d is_ij＝||P_j-v_iI represents P_jAnd a clustering center v_iEuclidean distance of, m (m)>1) The weighting index m is related to the degree of ambiguity between clusters, and r is an intermediate variable.

If j is present, r satisfies d_ij ⁽¹⁾0, then u_ij ⁽¹⁾1 is ═ 1; and when i ≠ r, u_ij ⁽¹⁾＝0。

Calculating and updating the clustering center after the first iteration according to the following formula:

substep 6.5, increasing l-1 by 1, and obtaining a membership matrix with iteration number l as follows:

the clustering center with iteration number l is as follows:

substep 6.6, the objective function is calculated according to the following formula:

wherein v is_iIs the cluster center, j is the number of samples, i is the number of clusters, and d is the Euclidean distance from the sample center to the cluster center. Stopping iteration when the objective function is less than epsilon or the iteration times L reach the maximum value L, otherwise returning to substep 6.3.

Step 7, obtaining the Z class centers and the membership values of the M samples to the Z class centers after iteration; and outputting the final sorting result according to the membership value.

The effects of the present invention can be further demonstrated by the following simulation experiments.

1) Simulation conditions are as follows:

the simulation platform provided by the invention uses Inter (R) core (TM) i5-8250U 1.6GHz, the internal memory is 8GB, a PC computer of Windows10 professional version is operated, and the simulation software is Matlab R2018 a.

2) Simulation content and result analysis:

there are a total of 4 radars in the simulation, each producing 100 samples; namely, Z is 4, a is 100, M is AZ is 100 × 4 is 400; the specific parameters of the radar are shown in table 1, each radar generates 100 pulses under the condition of 0dB respectively according to the parameters in table 1, the characteristics are extracted by using the method of the invention, and then the sorting is carried out, and table 2 is the final sorting result.

TABLE 1

Radar serial number

Type (B)

RF/MHz

PW/us

DOA/。

Type of space

PRI/us

Radar1

Linear frequency modulation

290～350

40

46～48

Difference of transmission

100 125 150

Radar2

Linear frequency modulation

380～440

50

34～56

Fixing

200

Radar3

Linear frequency modulation

410～470

60

52～54

Difference of transmission

250 275 300

Radar4

Linear frequency modulation

320～380

50

49～51

Shaking by 20%

300

TABLE 2

Radar serial number	Radar1	Radar2	Radar3	Radar4
					Number of original data	100	100	100	100
Correct sorting rate	98.3％	97.4％	97.5％	98.6％

As can be seen from Table 2, the sorting accuracy of 4 radar radiation source signals reaches more than 97%, and therefore, the features extracted by the method have better inter-class separation capability and can effectively sort the linearly frequency-modulated radar radiation sources.

Fig. 2 is a time domain diagram of the signal and a mode diagram after the variation mode decomposition, and it can be known from fig. 2 that the radar signal is decomposed into two modes, and the two modes are two modes with the highest energy, which just reflect the intra-pulse characteristics of the signal.

Fig. 3 is a central frequency distribution graph extracted after signal variation modal decomposition, and as can be seen from fig. 3, radar signals with different parameters have better distinctiveness and are distinguished very obviously, which indicates that the central frequency can be used as a better sorting feature.

Although the present invention has been described in detail in this specification with reference to specific embodiments and illustrative embodiments, it will be apparent to those skilled in the art that modifications and improvements can be made thereto based on the present invention. Accordingly, such modifications and improvements are intended to be within the scope of the invention as claimed.

Claims

1. A radar radiation source feature extraction method based on variational modal decomposition is characterized by comprising the following steps:

step (ii) of3, carrying out mirror extension on the preprocessed signal s (t), carrying out Fourier transform on the preprocessed signal after mirror extension to obtain a frequency spectrum f (omega) after Fourier transform, and reserving a positive half shaft of the frequency spectrum after Fourier transform to obtain a positive half shaft radar signal frequency spectrum

2. The method for extracting features of radar radiation source based on variational modal decomposition according to claim 1, wherein in step 1, the preprocessing is: converting the high-frequency pulse signal into an intermediate-frequency pulse signal through a low-pass filter, and then amplifying and matching filtering the intermediate-frequency pulse signal by a multi-stage intermediate-frequency amplifier to obtain an amplified intermediate-frequency signal; and carrying out A/D sampling on the amplified intermediate frequency signal to obtain a preprocessed signal s (t).

3. The method for extracting features of radar radiation source based on variational modal decomposition according to claim 1, wherein the formula of the frequency spectrum f (ω) after fourier transform is as follows:

4. The method for extracting radar radiation source features based on variational modal decomposition according to claim 1, wherein in the step 4, the variational modal decomposition comprises the following sub-steps:

u_k＝A_kcos(φ_k)；ω_k＝φ′；

wherein u is_kRepresenting an amplitude of A_kFrequency of omega_kHarmonic signal of A_kRepresents u_kInstantaneous amplitude of (e), omega_kIs u_kThe instantaneous frequency of (d);

(a) The first modality is updated according to the following formula:

wherein,for the first modality after the first iteration,the method comprises the steps of taking a positive half-axis radar signal frequency spectrum, taking α as a penalty factor, taking omega as the frequency of a radar signal, and taking h as an intermediate variable, wherein the h is used for eliminating other modes of a current mode for summation;

wherein,a center frequency of a first iteration for a first mode;

wherein,for the modality after the first iteration of the second modality,a center frequency of a first iteration for a second mode;

wherein,for the mode after the first iteration of the kth mode, ω_k ²The center frequency of the first iteration for the kth mode;

obtaining the mode after the first iteration of the Kth mode until K reaches the preset maximum decomposition layer number KAnd the center frequency ω of the first iteration of the Kth mode_K ²；

wherein,zeta is the noise tolerance for the Lagrangian updated after the first iteration;

wherein,for the k-th mode for the second timeMode after iteration, omega_k ³The center frequency of the second iteration for the kth mode;

wherein,for the lagrangian updated after the nth iteration,the mode after the nth iteration of the kth mode is selected; omega_k ⁿ⁺¹Is the center frequency of the kth mode after the nth iteration, i.e. the center frequency omega of the kth mode of the radar signal_k；

Substep 4.9, if the iteration times N are sequentially increased by 1, judging whether the iteration times N reach the maximum value N or whether the error is smaller than the set termination condition epsilon;

5. The method for extracting features of radar radiation source based on variational modal decomposition according to claim 4, wherein the formula of the error smaller than the set termination condition epsilon is as follows:

wherein | | xi | purple₂Is a 2-norm.

6. A radar radiation source feature classification method based on variational modal decomposition is characterized by comprising the following steps:

step 2, setting iteration times L, an iteration termination factor epsilon, a cluster number c, a weighting index m and a maximum iteration time L of fuzzy clustering; initializing a center matrix V (of cluster centers)⁰⁾＝{v₁,v₂,…,v_i,…,v_c}，i＝1,2,…,c；

7. The method for classifying radar radiation source features based on variational modal decomposition according to claim 6, wherein the constraint conditions in step 4 are as follows:

wherein,represents each P_jThe sum of the membership degrees for the different radar radiation sources equals 1.

8. The method for classifying radar radiation source features based on variational modal decomposition according to claim 6, wherein the step 5 comprises the following sub-steps:

and a substep 5.1, setting the iteration number l to be 1, and updating the membership matrix U after the first iteration according to the following formula:

wherein d is_ij＝||P_j-v_iI represents P_jAnd a clustering center v_iM is a weighting index, m is related to the fuzzy degree between clusters, r is an intermediate variable;

if j is present, r satisfies d_ij ⁽¹⁾0, then u_ij ⁽¹⁾1 is ═ 1; and when i ≠ r, u_ij ⁽¹⁾＝0；

and substep 5.2, in turn, increasing l-1 by 1 to obtain a membership matrix with iteration number l as follows:

the clustering center with iteration number l is as follows:

9. the method for classifying radar radiation source features based on variational modal decomposition according to claim 6, wherein in step 6, the objective function is calculated according to the following formula:

wherein v is_iIs the cluster center, j is the number of samples, i is the number of clusters, and d is the Euclidean distance from the sample center to the cluster center.