CN104809447A - Radiation source individual feature extraction method - Google Patents

Abstract

The invention discloses a radiation source individual feature extraction method. The method includes the following steps: setting two-dimensional position coordinates corresponding to each cloud droplet as S (i)=(x(i), y(i)), and solving average values of all cloud droplets; solving an entropy value of each cloud droplet according to an entropy value formula; calculating an average entropy value of n corresponding cloud droplets; calculating mean square error of the entropy values, and utilizing the above result to acquire super entropy according to a super entropy calculation formula. On the basis of conventional feature extraction, a cloud cluster model is built for conventional fuzzy features acquired at a low signal-to-noise ratio, and distribution characteristics of a feature cloud cluster are depicted by utilizing basic digital features of a cloud model theory so as to realize finer signal feature extraction.

Description

A kind of radiation source personal feature extracting method

Technical field

The present invention relates to a kind of feature extracting method, particularly relate to a kind of radiation source personal feature extracting method.

Background technology

Specific emitter identification technology has great importance in signal transacting field.Its needs in the noise circumstance of complexity, identify the individual signal of radiation source, thus provides foundation for further treatment and analysis signal.In Modern Communication System, communication environment is complicated and changeable, due to the existence of various interference in the measuring error of communication facilities itself and transmission environment, the signal characteristic parameter received is compared with true value and has certain randomness and ambiguity, this just makes the key link in specific emitter identification---and feature extraction becomes a difficult problem.Therefore, how in electromagnetic environment complicated and changeable with less computation complexity, less computing time, under low signal-to-noise ratio, reach higher discrimination to radiation source individuality, be the key of Modern Communication System research.

In existing technical method, the characteristic parameter extraction algorithm based on time domain calculates relatively simple, and easily realize, but the characteristic parameter extracted is easily affected by noise, is suitable for the environment that signal to noise ratio (S/N ratio) is higher, thus its range of application is subject to certain restrictions.In frequency domain or in time-frequency domain, spectrum analysis is carried out to different signals, and then signal is identified, relative to the Signal analysis under simple time domain, there is better noiseproof feature, apply comparatively extensive, the method of feature extraction is varied especially, but its computation complexity is comparatively complicated relative to temporal signatures extraction algorithm.As can be seen from the document that Chinese scholars was delivered in recent years, the Feature extraction and recognition theory of signal is more and more subject to the attention about scholar, various modern signal processing technology, comprising wavelet theory, fractal theory, artificial neural network, high-order statistic, spectral correlation theory etc., all start or be applied in the research to this theory.But, how in the feature compared with extraction signal effective under Low SNR, and reduce the complexity of computing method as far as possible, still well do not solved, become the emphasis of current research.

Summary of the invention

For overcoming the deficiency that above-mentioned prior art exists, the object of the present invention is to provide a kind of radiation source personal feature extracting method, on the basis that traditional characteristic extracts, to the traditional fuzzy feature construction cloud cluster model extracted under low signal-to-noise ratio, the basic digital features of recycling cloud models theory, portrays the distribution character of feature cloud cluster, thus realize meticulousr signal characteristic abstraction, the present invention calculates simply, and by Further Feature Extraction, realizes the specific emitter identification under more low signal-to-noise ratio.

For reaching above-mentioned and other object, the present invention proposes a kind of radiation source personal feature extracting method, comprises the steps:

Step one, if the two-dimensional position coordinate corresponding to each water dust (S (i)) is S (i)=(x (i), y (i)), obtains the average (Ex) of all water dusts;

Step 2, obtains entropy (En (i)) to each water dust according to entropy formula;

Step 3, calculates the average entropy (En) of a corresponding n water dust;

Step 4, calculates the mean square deviation of entropy (En (i)), utilizes the above results to obtain super entropy (He) according to super entropy computing formula.

Further, the two dimensional character of described each water dust for utilizing traditional algorithm to extract.

Further, in step one, obtain the average (Ex) of all water dusts according to following formula:

$\mathrm{Ex}=\frac{1}{n}\underset{i=1}{\overset{n}{\mathrm{\Σ}}}x\left(i\right)$

Wherein, x (i) represents the numerical values recited of i-th water dust horizontal ordinate, i=1,2 ..., n, n are the number of water dust.

Further, in step 2, this entropy formula is:

$\mathrm{En}\left(i\right)=\sqrt{-\frac{{(S\left(i\right)-\mathrm{Ex})}^2}{2\ln y\left(i\right)}}.$

Further, in step 3, calculate average entropy according to following formula:

$\mathrm{En}=\frac{1}{n}\underset{i=1}{\overset{n}{\mathrm{\Σ}}}\mathrm{En}\left(i\right).$

Further, in step 4, this super entropy computing formula is:

$\mathrm{He}=\sqrt{\frac{1}{n-1}\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{(\mathrm{En}\left(i\right)-\mathrm{En})}^2}.$

Further, described sample to be tested (S (i)) and the distance of the center-of-gravity value of water dust group, determine that this sample belongs to the size of this classification possibility.

Further, described entropy (En (i)) reflection water dust group energy the scope that accepts by sample of signal feature.

Further, the expression of the slackness that described super entropy (He) is entropy, has measured the uncertainty of entropy.

Compared with prior art, a kind of Interval Grey associative classification device method for designing based on adaptive entropy power of the present invention, can classify for the signal with overlapping features distribution extracted under low signal-to-noise ratio, and utilize entropy power algorithm to improve interval associative classification device, by selecting the significance level of different characteristic, improve the adaptive ability of interval associative classification device algorithm.

Accompanying drawing explanation

Fig. 1 is cloud model distribution plan;

Fig. 2 is the flow chart of steps of a kind of radiation source personal feature of the present invention extracting method;

Fig. 3 is a tradition feature extraction result figure;

Fig. 4 is that secondary cloud model numerical characteristic of the present invention extracts result.

Embodiment

Below by way of specific instantiation and accompanying drawings embodiments of the present invention, those skilled in the art can understand other advantage of the present invention and effect easily by content disclosed in the present specification.The present invention is also implemented by other different instantiation or is applied, and the every details in this instructions also can based on different viewpoints and application, carries out various modification and change not deviating under spirit of the present invention.

Cloud model can integrate the ambiguity of object to be described, and randomness, form one qualitative (concept in domain) and mapping that quantitatively degree of membership of property value (in the domain) is mutual, this is just for the individual signal characteristic abstraction of radiation source provides good theoretical foundation.Because under low signal-to-noise ratio, noise ratio is larger, have impact on the precision of signal characteristic abstraction, to such an extent as to the signal characteristic extracted distribution presents certain ambiguity, and unique point distributes and is similar to the distribution characteristics of cloud model in certain interval, utilize this characteristic, extract the feature of signal characteristic distribution with three numerical characteristics of cloud model further, thus realize more describing more accurately signal under low signal-to-noise ratio.

The basic definition of cloud model is:

For a common set X, definition domain X={x}.Suppose fuzzy set be a set belonging in domain X, now, for arbitrary element x, definition has the random number of steady tendency be called x couple degree of membership.If each element is all simply orderly, so, the variable based on X becomes, the distribution of its degree of membership on X is referred to as membership clouds; If each element is not orderly, but according to correspondence rule f, X is mapped on an orderly domain X', satisfy condition simultaneously, have in X' and only have an x' and x corresponding, now, X' can be counted as basic underlying variables, claims degree of membership to be distributed as membership clouds on X'.

The most frequently used to the expression formula of normal cloud model can be expressed as:

$\mathrm{\μ}=\exp [-\frac{{(x_i-\mathrm{Ex})}^2}{2{\mathrm{En}}^2}]$

Wherein, x ⁱbe any number in domain, Ex is the central value of domain; En is an entropy of domain ranging concept.Draw the state scattergram of cloud model as shown in Figure 1, the meaning of each numerical characteristic also invests in figure.Wherein, horizontal ordinate represents the distribution of object value to be described, and ordinate represents each degree of membership for object to be described.

Cloud model is that characterize overall distribution character, the meaning representated by them can be expressed as follows by expecting Ex, entropy En, super entropy He:

The numerical value of water dust point distribution is expected to represent with expectation Ex, and it reflects the centre of gravity place of water dust group, is the point that can represent qualitativing concept.

" entropy " theory has been introduced into information theory, statistical physics scheduling theory, is used to the state parameter describing thermodynamic argument at first, characterizes the uncertainty degree of object to be described.In the definition of cloud model concept, qualitativing concept measurable degree entropy En represents, its value is larger, and the concept of representative is more macroscopical.

And probabilistic measure definitions of entropy is super entropy He, it reflects the cohesion degree of water dust, namely in domain space the coherency of uncertainty a little.Its size illustrates dispersion degree and the thickness of cloud indirectly, is jointly determined by the randomness of entropy and ambiguity.

As can be seen from the definition of cloud models theory, the distribution of cloud model mid point meets normal distribution, and the signal characteristic parameter changed under signal to noise ratio (S/N ratio), along with the change of noise, eigenwert fluctuates near some stable values, its distribution meets Normal Distribution Characteristics, therefore, the numerical characteristic parameter of cloud model can be utilized to expect Ex, entropy En, super entropy He reflects the traditional characteristic distribution of the instability under change signal to noise ratio (S/N ratio), the more meticulous parameter attribute extracting different radiation source individuality.

The present invention utilizes the numerical characteristic of cloud model, by precise information, the i.e. distribution of water dust in cloud model, effectively be converted to its numerical characteristic average Ex, entropy En, with the concept represented by super entropy He, and utilize numerical characteristic to represent the entirety of the water dust that precise information reflects, and then the accurate description of the ambiguity achieved signal characteristic distribution and randomness.Fig. 2 is the flow chart of steps of a kind of radiation source personal feature of the present invention extracting method.As shown in Figure 2, a kind of radiation source personal feature of the present invention extracting method, comprises the steps:

Step 201. establishes each water dust S (i), i.e. the two dimensional character that extracts of traditional algorithm, and corresponding two-dimensional position coordinate is: S (i)=(x (i), y (i)), by formula obtain the average Ex of all water dusts.Wherein, x (i) represents the numerical values recited of i-th water dust horizontal ordinate, i=1,2 ..., n, n are the number of water dust.

Step 202, to each water dust: S (i)=(x (i), y (i)), by entropy formula:

$\mathrm{En}\left(i\right)=\sqrt{-\frac{{(S\left(i\right)-\mathrm{Ex})}^2}{2\ln y\left(i\right)}},$ Obtain entropy En (i);

Step 203, according to formula obtain the average entropy En of a corresponding n water dust;

Step 204, finally, asks the mean square deviation of En (i), utilizes super entropy computing formula substitute into above-mentioned correlation values of trying to achieve, namely can obtain super entropy He.

Wherein, sample to be tested S (i) and the distance of the center-of-gravity value Ex of water dust group, determine the size that this sample belongs to this classification possibility;

In number field space, entropy En (i) reflect water dust group energy the scope that accepts by sample of signal feature;

Super entropy He is the expression of the slackness of entropy, has measured the uncertainty of entropy.

The traditional characteristic extracted correspond to the water dust (x (i) in cloud feature, y (i)), according to above computing method, by cloud model feature second extraction, second extraction is carried out to the distribution situation of traditional characteristic, thus realize compared with under low signal-to-noise ratio to the object that signal identifies.

For the feature extraction of conventional two-dimensional entropy, calculate the Shannon entropy of 6 kinds of unlike signals under different signal to noise ratio (S/N ratio)s and Exponential Entropy feature, form two-dimensional feature vector.Due to the existence of noise, the two-dimensional entropy eigenwert of each signal is not a fixed value, but fluctuates at some characteristic intervals along with the change of signal to noise ratio (S/N ratio), simultaneously, along with the increase of noise, the waving interval of entropy eigenwert is also along with change is large, and simulation result as shown in Figure 3.

Wherein, horizontal ordinate H1 represents the Shannon entropy feature of signal, and ordinate H2 represents the Exponential Entropy feature of signal.Figure (a), (b), (c), (d) represents that signal to noise ratio (S/N ratio) is the simulation result under 4dB, 0dB ,-4dB ,-8dB environment respectively.

As can be seen from the simulation result of accompanying drawing 3, when signal to noise ratio (S/N ratio) is 4dB, the aggregation of signal characteristic is better, and the eigenwert of different modulating type signal has very little overlapping, and the feature of signal is relatively easily classified; Along with the reduction of signal to noise ratio (S/N ratio), when signal to noise ratio (S/N ratio) is-8dB, the eigenwert of signal has certain discreteness, and the eigenwert overlapping interval of different modulating type signal becomes large.

In order to improve the entropy feature extracted class in degree of separation between aggregation and class, the numerical characteristic of the cloud model utilizing the present invention to propose, average Ex, entropy En, super entropy He, the distribution character of the entropy feature of second extraction 6 kinds of unlike signals.Simulation result as shown in Figure 4.

Wherein, x coordinate axis Ex represents the average that signal characteristic distributes, i.e. the central value of signal characteristic distribution; Y coordinate axis En represents the entropy that signal characteristic distributes, i.e. the discrete feature of signal characteristic distribution; Z coordinate axle He represents the super entropy that signal characteristic distributes, i.e. the dispersion degree of the entropy feature of signal characteristic distribution.Same master drawing (a), (b), (c), (d) represents the simulation result under 4dB, 0dB ,-4dB ,-8dB signal to noise ratio (S/N ratio) environment respectively.

As can be seen from the result of accompanying drawing 4, the distribution character of the entropy feature of second extraction signal, the Two dimensional Distribution entropy feature of signal is converted into three-dimensional cloud model feature, makes the three dimensional signal feature extracted relative to two-dimensional entropy feature, there is in better class degree of separation between concentration class and class.Utilize grey correlation sorter, the communication signal recognition rate under each signal to noise ratio (S/N ratio) environment in computer sim-ulation figure, meanwhile, with the discrimination Comparative result of two-dimensional entropy value tag only extracting signal, recognition result is as shown in table 1.

Based on the signal identification rate of entropy characteristic sum entropy cloud feature under the different signal to noise ratio (S/N ratio) of table 1

As can be seen from the simulation result of table 1, when signal to noise ratio (S/N ratio) is 4dB, signal identification rate based on two kinds of feature extraction algorithms can reach 100%, along with the reduction of signal to noise ratio (S/N ratio), when signal to noise ratio (S/N ratio) is 0dB, during-4dB, feature extraction algorithm based on entropy cloud feature still can reach the discrimination of 100%, but the discrimination based on the feature extraction algorithm of entropy feature decreases, when signal to noise ratio (S/N ratio) is reduced to-8dB, the discrimination of the algorithm that this invention proposes is significantly improved than the tool of primal algorithm.

In sum, a kind of radiation source personal feature of the present invention extracting method utilizes the numerical characteristic of cloud model, by precise information, the i.e. distribution of water dust in cloud model, effectively be converted to its numerical characteristic average Ex, entropy En, and the concept represented by super entropy He, and utilize numerical characteristic to represent the entirety of the water dust that precise information reflects, and then the accurate description of the ambiguity achieved signal characteristic distribution and randomness.

Above-described embodiment is illustrative principle of the present invention and effect thereof only, but not for limiting the present invention.Any those skilled in the art all without prejudice under spirit of the present invention and category, can carry out modifying to above-described embodiment and change.Therefore, the scope of the present invention, should listed by claims.

Claims (9)

1. a radiation source personal feature extracting method, comprises the steps:

Step one, if the two-dimensional position coordinate corresponding to each water dust (S (i)) is S (i)=(x (i), y (i)), obtains the average (Ex) of all water dusts;

Step 2, obtains entropy (En (i)) to each water dust according to entropy formula;

Step 3, calculates the average entropy (En) of a corresponding n water dust;

Step 4, calculates the mean square deviation of entropy (En (i)), utilizes the above results to obtain super entropy (He) according to super entropy computing formula.

2. a kind of radiation source personal feature extracting method as claimed in claim 1, is characterized in that: the two dimensional character of described each water dust for utilizing traditional algorithm to extract.

3. a kind of radiation source personal feature extracting method as claimed in claim 2, is characterized in that: in step one, obtains the average (Ex) of all water dusts according to following formula:

$\mathrm{Ex}=\frac{1}{n}\underset{i=1}{\overset{n}{\mathrm{\Σ}}}x\left(i\right)$

Wherein, x (i) represents the numerical values recited of i-th water dust horizontal ordinate, i=1,2 ..., n, n are the number of water dust.

4. a kind of radiation source personal feature extracting method as claimed in claim 3, is characterized in that, in step 2, this entropy formula is:

$\mathrm{En}\left(i\right)=\sqrt{-\frac{{(S\left(i\right)-\mathrm{Ex})}^2}{2\ln y\left(i\right)}}.$

5. a kind of radiation source personal feature extracting method as claimed in claim 4, is characterized in that, in step 3, calculate average entropy according to following formula:

$\mathrm{En}=\frac{1}{n}\underset{i=1}{\overset{n}{\mathrm{\Σ}}}\mathrm{En}\left(i\right).$

6. a kind of radiation source personal feature extracting method as claimed in claim 5, is characterized in that, in step 4, this super entropy computing formula is:

$\mathrm{He}=\sqrt{\frac{1}{n-1}\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{(\mathrm{En}\left(i\right)-\mathrm{En})}^2}.$

7. a kind of radiation source personal feature extracting method as claimed in claim 6, is characterized in that: described sample to be tested (S (i)) and the distance of the center-of-gravity value of water dust group, determine that this sample belongs to the size of this classification possibility.

8. a kind of radiation source personal feature extracting method as claimed in claim 6, is characterized in that: described entropy (En (i)) reflection water dust group energy the scope that accepts by sample of signal feature.

9. a kind of radiation source personal feature extracting method as claimed in claim 6, is characterized in that: the expression of the slackness that described super entropy (He) is entropy, has measured the uncertainty of entropy.