CN105119862A - Signal modulation type identification method and signal modulation type identification system - Google Patents

Abstract

The invention discloses a signal modulation type identification method and a signal modulation type identification system. The method comprises the following steps: preprocessing a signal to be identified, extracting a predetermined number of feature parameters, and using a feature vector formed by the predetermined number of feature parameters to represent the signal to be identified; using an optimal projection matrix to extract the features of the signal to be identified, and projecting the signal to be identified into a low-dimensional feature subspace, wherein the optimal projection matrix is obtained through a locality preserving projection algorithm; and calculating the Euclidean distance between the signal to be identified in the low-dimensional feature subspace and a training signal of which the signal modulation type is known, and determining the signal modulation type of the signal to be identified based on the nearest neighbor algorithm of Euclidean distance. According to the technical scheme, the signal to be identified is projected into the low-dimensional feature subspace by using the optimal projection matrix, which reduces the amount of calculation; and the optimal projection matrix obtained through the locality preserving projection algorithm reduces the deviation of the signal in the process of projection, and has better robustness and higher identification rate.

Description

A kind of identification of signal modulation method and system

Technical field

The present invention relates to blipology field, be specifically related to a kind of identification of signal modulation method and system.

Background technology

The identification of signal modulation style is the intermediate steps of signal interception and demodulation, has important researching value in dual-use field, is widely used in the fields such as signal confirmation, disturbance ecology, spectrum management, electronic countermeasures.Under the electromagnetic environment of complexity, realize detecing receipts to signal, form valuable communication intelligence, thus provide according to being the task of having challenge for means such as follow-up discriminating, interference.If the modulation system of detecing and to collect mail number can be identified, just can employ less resource and cost is disturbed specific signal parameter, reach the effect of getting twice the result with half the effort.Traditional identification of signal modulation be mostly analyst by various instrument, by means of practical experience make analyze judge, efficiency is lower, and the signal type of identification is also limited.

In recent years, the automatic identification of signal modulation style causes the extensive concern of domestic and international researcher, achieves good achievement in research.At present, the automatic identification of modulation type mainly contains recognition methods based on decision theory and statistical pattern recognition method.Regard Modulation recognition as Hypothesis Testing Problem based on the recognition methods of decision theory, this class methods amount of calculation is comparatively large, needs the correct threshold value of setting and to noise-sensitive.The method of Corpus--based Method pattern recognition is mainly divided into feature extraction and two stages of Classification and Identification, realizes simple and good stability.The one of statistical pattern recognition method attempts being based on support vector machines (SupportVectorMachine) by the multi classifier based on SVM, for signal modulate, tests the recognition performance under different IPs function.Another kind of trial is the spectral correlation analyzing different modulating mode, and adopts principal component analysis PCA method (PrincipalComponentsAnalysis) to compress dimension, ideally can obtain good effect in signal to noise ratio.

But, all only otherness between training signal sample is paid close attention to based on support vector machines or based on the trial of principal component analysis PCA, result in existing signal modulate scheme in non-cooperative communication system, poor robustness under Low SNR, unstable properties, discrimination are low.

Summary of the invention

The invention provides a kind of modulation identification method and system to solve existing signal modulate technology in non-cooperative communication system, poor robustness under Low SNR, unstable properties, the problem that discrimination is low.

In order to achieve the above object, technical scheme of the present invention is achieved in that

According to an aspect of the present invention, provide a kind of identification of signal modulation method, method comprises:

Treat identification signal and carry out preliminary treatment, extract the characteristic parameter of predetermined quantity, and the characteristic vector using the characteristic parameter of described predetermined quantity to form characterizes described signal to be identified;

Utilize optimum projection matrix to carry out feature extraction to described signal to be identified, described signal to be identified is projected in low-dimensional proper subspace; Wherein, described optimum projection matrix is obtained by locality preserving projections algorithm;

Calculate signal to be identified in described low-dimensional proper subspace and the Euclidean distance between the known training signal of signal modulation style, the nearest neighbor algorithm based on Euclidean distance determines the signal modulation style of signal to be identified.

Alternatively, treating before identification signal carries out preliminary treatment, the method also comprises: training obtains optimum projection matrix; Described training obtains optimum projection matrix and comprises:

The pre-treatment step same with described signal to be identified is carried out to the training signal that signal modulation style is known, extract the characteristic parameter of described predetermined quantity, and the characteristic vector using described characteristic parameter to form characterizes the known training signal of described signal modulation style, obtain the sample of signal matrix comprising each modulation type training signal;

Utilize locality preserving projections algorithm, described sample of signal matrix is processed, calculates optimum projection matrix.

Alternatively, the described training signal known to signal modulation style carries out the pre-treatment step same with described signal to be identified, and the characteristic parameter extracting described predetermined quantity comprises:

Step 31, training signal s (t) received is sampled, obtains instantaneous amplitude a (i) of described training signal s (t), instantaneous phase φ (i), instantaneous frequency f (i) and power spectrum S (i); Wherein, sample frequency is f _s, sampling number is N _s;

Step 32, calculates the standard deviation sigma of normalization instantaneous amplitude _a, the standard deviation of normalization instantaneous amplitude and the ratio R of average _aand the divergence μ of normalization instantaneous amplitude ^a;

Step 33, calculates the standard deviation sigma of normalization instantaneous frequency _f, normalization instantaneous frequency divergence μ ^fand the standard deviation sigma of normalized nonlinear instantaneous phase _φ;

Step 34, the symmetry P of rated output spectrum;

The characteristic parameter of described predetermined quantity comprises totally 7 characteristic parameters that step 32, step 33 and step 34 calculate.

Alternatively, described step 32, calculates the standard deviation sigma of normalization instantaneous amplitude _a, the standard deviation of normalization instantaneous amplitude and the ratio R of average _aand the divergence μ of normalization instantaneous amplitude ^acomprise:

Utilize the standard deviation sigma of following formulae discovery normalization instantaneous amplitude _a:

wherein,

Utilize the standard deviation of following formulae discovery normalization instantaneous amplitude and the ratio R of average _a:

wherein,

Utilize the divergence μ of following formulae discovery normalization instantaneous amplitude ^a:

Alternatively, described step 33, calculates the standard deviation sigma of normalization instantaneous frequency _f, normalization instantaneous frequency divergence μ ^fand the standard deviation sigma of normalized nonlinear instantaneous phase _φcomprise:

Utilize the standard deviation sigma of following formulae discovery normalization instantaneous frequency _f:

wherein, f _n=f (i)/f _max, f _max=max{f (i) };

Utilize the divergence μ of following formulae discovery normalization instantaneous frequency ^f:

Utilize the standard deviation sigma of following formulae discovery normalized nonlinear instantaneous phase _φ:

wherein, the non-linear partial that φ (i) is instantaneous phase.

Alternatively, described step 34, the symmetry P of rated output spectrum comprises:

Utilize the symmetry P of following formulae discovery power spectrum:

Wherein, s (i)=DFT [s (i)], N _fc=f _cn _s/ (f _s-1).

Alternatively, describedly utilize locality preserving projections algorithm, described sample of signal matrix processed, calculates optimum projection matrix and comprise:

If sample of signal matrix X=is [x ₁, x ₂..., x _n], N represents the number of training signal sample, x _i∈ R ^{7 × 1}(i=1,2 ..., N) and represent i-th training signal sample;

Build similar adjacent map G={Z, B}, wherein, vertex set Z={x ₁, x ₂x _n, weight matrix B ∈ R ^{n × N};

Structure target function is as follows:

Wherein, weight matrix B is a symmetrical matrix, and its i-th row jth column element is defined as follows:

Y _itraining signal sample x _ilow dimension formulation, α represents projecting direction, makes y=α ^tx brings target function into, obtains:

Wherein, L=D-B is Laplacian Matrix;

D is a diagonal matrix, or

XLX ^trepresent the local dispersion matrix of training signal sample.

Increase a constraints to described target function and make α ^txDX ^tα=1, in conjunction with described constraints, described target function can be write as:

Solve following characteristics equation:

αXLX ^Tα＝λXDX ^Tα

If A=XLX ^t, B=XDX ^t, then λ is B ^-1a corresponds to the characteristic value of characteristic vector α;

Optimum projection matrix α=[α ₁α ₂α _l], the column vector of α is by B ^-1front l the eigenvalue of maximum characteristic of correspondence vector composition of A, wherein, l represents the quantity of projecting direction.

Alternatively, the signal to be identified in described calculating described low-dimensional proper subspace and the Euclidean distance between the known training signal of signal modulation style, the nearest neighbor algorithm based on Euclidean distance determines that the signal modulation style of signal to be identified comprises:

Utilize optimum projection matrix α by signal x to be identified ^*project in described low-dimensional proper subspace, obtain described signal x to be identified ^*low dimension formulation y in low-dimensional proper subspace ^*;

Utilize described optimum projection matrix to carry out feature extraction to each training signal in described sample of signal matrix, described training signal is projected in low-dimensional proper subspace;

Calculate signal x to be identified ^*with the training signal x that signal modulation style is known _kbetween Euclidean distance;

If described Euclidean distance meets following formula, then determine signal x to be identified ^*with the training signal x that modulation type is known _kbelong to same class;

Wherein, d (y ^*, y _i)=|| y ^*-y _i|| ², (i=1,2 ..., k).

Corresponding with above-mentioned identification of signal modulation method, present invention also offers a kind of identification of signal modulation system, described system comprises:

Pretreatment unit, carrying out preliminary treatment for treating identification signal, extracting the characteristic parameter of predetermined quantity, and the characteristic vector using the characteristic parameter of described predetermined quantity to form characterizes described signal to be identified;

Feature extraction unit, for utilizing optimum projection matrix to carry out feature extraction to described signal to be identified, projects in low-dimensional proper subspace by described signal to be identified; Wherein, described optimum projection matrix is obtained by locality preserving projections algorithm;

Judgement output unit, for calculating signal to be identified in described low-dimensional proper subspace and the Euclidean distance between the known training signal of signal modulation style, the nearest neighbor algorithm based on Euclidean distance determines the signal modulation style of signal to be identified.

Alternatively, described system also comprises:

Optimum projection matrix training unit, preliminary treatment is carried out for the training signal known to signal modulation style, extract the characteristic parameter of described predetermined quantity, and the characteristic vector using described characteristic parameter to form characterizes described training signal, obtain the sample of signal matrix comprising each modulation type training signal; Utilize locality preserving projections algorithm, described sample of signal matrix is processed, calculates optimum projection matrix.

The invention has the beneficial effects as follows: this identification of signal modulation method and system of the embodiment of the present invention utilizes the optimum projection matrix obtained in advance, signal to be identified is projected in low-dimensional proper subspace, and in this proper subspace, calculates the modulation type determining signal to be identified.This compared with directly calculating the scheme of the modulation type judging signal to be identified in higher dimensional space, can greatly reduce amount of calculation, the efficiency improving identification of signal modulation and stability.In addition, the present invention obtains optimum projection matrix by locality preserving projections algorithm, obtains optimum projection matrix after namely utilizing the sample of signal matrix of locality preserving projections algorithm to the training signal place comprising each modulation type to carry out locality preserving projections calculating.So in locality preserving projections process, not only maintain the otherness feature of sample of signal, also maintain the local similarity feature of sample, identification of signal modulation method of the present invention there is robustness preferably, even if also can obtain higher discrimination when making, signal to noise ratio more at signal modulation style to be identified lower.

Accompanying drawing explanation

Fig. 1 is the schematic flow sheet of a kind of identification of signal modulation method of one embodiment of the invention;

Fig. 2 is the principle schematic of a kind of identification of signal modulation method of the present invention;

Fig. 3 is the employing characteristic vector number of one embodiment of the invention when being 2, and training signal sample is schemed in the drop shadow effect of low-dimensional proper subspace;

Fig. 4 is that the discrimination of identification of signal modulation method of the present invention and prior art PCA method contrasts schematic diagram;

Fig. 5 be prior art PCA method under signal to noise ratio is the condition of 10db, for the discrimination result schematic diagram acquired by unlike signal modulation type;

Fig. 6 is identification of signal modulation method of the present invention is under the condition of 10db in signal to noise ratio, for the discrimination result schematic diagram acquired by unlike signal modulation type;

Fig. 7 is the block diagram of a kind of identification of signal modulation system that one embodiment of the invention provides.

Embodiment

Core concept of the present invention is: propose a kind of Corpus--based Method Pattern recognition principle identification of signal modulation scheme, solve identification of signal modulation poor robustness in non-cooperative communication system, and the problem that under low signal-to-noise ratio, discrimination is low.Technical scheme of the present invention carries out preliminary treatment by treating identification signal, extracts 7 kinds of characteristic parameters to characterize signal to be identified, and each training signal sample adopting locality preserving projections algorithm process signal modulation style known, obtain optimum projection matrix, this optimum projection matrix is utilized to carry out further feature extraction to pretreated signal to be identified and training signal, signal to be identified and training signal (signal to be identified and training signal can regard the point in higher dimensional space as) are projected in the proper subspace of low-dimensional, after locality preserving projections, training signal has similar compactness preferably in low-dimensional proper subspace, inhomogeneity discreteness, and then utilize the arest neighbors method of discrimination based on Euclidean distance calculate the distance between the signal to be identified training signal known with each modulation type after projection and compare, if the Euclidean distance of signal to be identified and wherein the training sample point that a certain class signal modulation style is known is minimum, then the signal modulation style of signal to be identified is defined as identical with this training sample signal modulation style, then the signal modulation style treating identification signal carries out judgement and exports.Thus, when calculating optimum projection matrix by locality preserving projections algorithm, both the otherness of training signal sample had been considered, take into account again the local similarity of training signal sample, the robustness of Signal analysis is better, thus makes radar recognition also can obtain higher discrimination under Low SNR.

Fig. 1 is the schematic flow sheet of a kind of identification of signal modulation method of one embodiment of the invention, and see Fig. 1, this identification of signal modulation method of the embodiment of the present invention comprises:

Step S110, treats identification signal and carries out preliminary treatment, extract the characteristic parameter of predetermined quantity, and the characteristic vector using the characteristic parameter of described predetermined quantity to form characterizes described signal to be identified;

Step S120, utilizes optimum projection matrix to carry out feature extraction to described signal to be identified, projects in low-dimensional proper subspace by described signal to be identified; Wherein, described optimum projection matrix is obtained by locality preserving projections algorithm;

Step S130, calculate signal to be identified in described low-dimensional proper subspace and the Euclidean distance between the known training signal of signal modulation style, the nearest neighbor algorithm based on Euclidean distance determines the signal modulation style of signal to be identified.

Through the step shown in Fig. 1, the signal type modulator approach of one embodiment of the invention, in locality preserving projections process, not only maintain the otherness feature of training signal sample, also maintain the local similarity feature of training signal sample, there is robustness preferably, even if so more at signal modulation style to be identified, also higher discrimination can be obtained when signal to noise ratio is lower.

In one embodiment of the invention, treating before identification signal carries out preliminary treatment, the method also comprises: training obtains optimum projection matrix; Training obtains optimum projection matrix and comprises: the training signal known to signal modulation style carries out the pre-treatment step same with signal to be identified, extract the characteristic parameter of predetermined quantity, and the training signal that the characteristic vector characterization signal modulation type of use characteristic parameter composition is known, obtain the sample of signal matrix comprising each modulation type training signal; Utilize locality preserving projections algorithm, sample of signal matrix is processed, calculate optimum projection matrix.

Fig. 2 is the principle schematic of a kind of identification of signal modulation method of the present invention, see Fig. 2, the square ratio juris of identification of signal modulation of the present invention is: sample to training signal sample, preliminary treatment → extraction 7 kinds of signal characteristic parameter representative training signal samples → utilize locality preserving projections, training produces optimum projection matrix, and optimum projection matrix is exported to nearest neighbor classifier; The quantity of training signal sample is here multiple, and multiple training signal sample constitutes sample of signal matrix.Next can treat identification signal to carry out same preprocessing process and namely treat that identification signal carries out signal sampling, the signal to be identified of preliminary treatment → extract 7 kinds of signal characteristic parameter representative signals to be identified → characterized by 7 kinds of characteristic parameters sends into nearest neighbor classifier; Then treat identification signal according to the optimum projection matrix obtained after training signal sample training to project, all project in same proper subspace by signal to be identified and training signal sample, in this proper subspace, nearest neighbor classifier is utilized to calculate the Euclidean distance of signal to be identified and which training signal sample minimum, so just be defined as by the modulation type of signal to be identified identical with the signal modulation style of this training signal sample, the signal modulation style then treating identification signal carries out judgement and exports.In the present embodiment, be the Euclidean distance calculating signal to be identified and the known multiple training signals of signal modulation style, and signal to be identified be defined as that training signal minimum with Euclidean distance and belong to same signal modulation style.Euclidean distance also claims Euclidean distance, and it is a distance definition usually adopted, and it represents the actual distance in m-dimensional space between two points.Euclidean distance can regard the similarity degree of signal as, and distance is nearer more similar.In one embodiment of the invention, the training signal known to signal modulation style carries out the pre-treatment step same with signal to be identified, and the characteristic parameter extracting predetermined quantity comprises:

Step 31, training signal s (t) received is sampled, obtains instantaneous amplitude a (i) of training signal s (t), instantaneous phase φ (i), instantaneous frequency f (i) and power spectrum S (i); Wherein, sample frequency is f _s, sampling number is N _s;

Step 32, calculates the standard deviation sigma of normalization instantaneous amplitude _a, the standard deviation of normalization instantaneous amplitude and the ratio R of average _aand the divergence μ of normalization instantaneous amplitude ^a;

Step 33, calculates the standard deviation sigma of normalization instantaneous frequency _f, normalization instantaneous frequency divergence μ ^fand the standard deviation sigma of normalized nonlinear instantaneous phase _φ;

Step 34, the symmetry P of rated output spectrum;

The characteristic parameter of predetermined quantity comprises totally 7 characteristic parameters that step 32, step 33 and step 34 calculate, and namely 7 characteristic parameters are respectively: the standard deviation sigma of normalization instantaneous amplitude _a, the standard deviation of normalization instantaneous amplitude and the ratio R of average _a, the divergence μ of normalization instantaneous amplitude ^a, the standard deviation sigma of normalization instantaneous frequency _f, the divergence μ of normalization instantaneous frequency ^f, the standard deviation sigma of normalized nonlinear instantaneous phase _φ, the symmetry P of power spectrum.

In one embodiment of the invention, step 32, calculates the standard deviation sigma of normalization instantaneous amplitude _a, the standard deviation of normalization instantaneous amplitude and the ratio R of average _aand the divergence μ of normalization instantaneous amplitude ^acomprise:

Utilize the standard deviation sigma of following formulae discovery normalization instantaneous amplitude _a:

wherein,

Utilize the standard deviation of following formulae discovery normalization instantaneous amplitude and the ratio R of average _a:

wherein,

Utilize the divergence μ of following formulae discovery normalization instantaneous amplitude ^a:

In one embodiment of the invention, step 33, calculates the standard deviation sigma of normalization instantaneous frequency _f, normalization instantaneous frequency divergence μ ^fand the standard deviation sigma of normalized nonlinear instantaneous phase _φcomprise:

Utilize the standard deviation sigma of following formulae discovery normalization instantaneous frequency _f:

wherein, f _n=f (i)/f _max, f _max=max{f (i) };

Utilize the divergence μ of following formulae discovery normalization instantaneous frequency ^f:

Utilize the standard deviation sigma of following formulae discovery normalized nonlinear instantaneous phase _φ:

wherein, the non-linear partial that φ (i) is instantaneous phase.

In one embodiment of the invention, step 34, the symmetry P of rated output spectrum comprises:

Utilize the symmetry P of following formulae discovery power spectrum:

Wherein, s (i)=DFT [s (i)], N _fc=f _cn _s/ (f _s-1).

Thus, 7 kinds of characteristic parameters of each training signal sample can be obtained by above-mentioned steps, utilize these 7 kinds of characteristic parameters can characterize a training signal sample, the dimension of training signal sample is regarded as by 7 kinds of characteristic parameters, each training signal sample can be regarded as a point in 7 dimension spaces, and 7 kinds of characteristic parameters are coordinate figures of this training signal sample point.

It should be noted that, in the embodiment of the present invention, same preprocessing process is carried out to training signal sample and signal to be identified, namely treat identification signal and also will carry out signal sampling, preliminary treatment, extract ratio, the divergence of normalization instantaneous amplitude, the standard deviation of normalization instantaneous frequency, the divergence of normalization instantaneous frequency, the standard deviation of normalized nonlinear instantaneous phase, symmetry totally 7 kinds of characteristic parameters of power spectrum of the standard deviation of the normalization instantaneous amplitude of signal to be identified, the standard deviation of normalization instantaneous amplitude and average.Preprocessing process due to signal to be identified is identical with the pre-treatment step of training signal sample, and therefore, the pre-treatment step treating identification signal see the pre-treatment step to training signal sample, can not repeat them here.

As from the foregoing, preliminary treatment is carried out to each training signal sample, extract 7 kinds of characteristic parameters of this training signal, represent the training signal sample of correspondence with 7 kinds of characteristic parameters after, just can obtain the sample of signal matrix be made up of these training signal samples.

In one embodiment of the invention, utilize locality preserving projections algorithm, sample of signal matrix processed, calculate optimum projection matrix and comprise:

If sample of signal matrix X=is [x ₁, x ₂..., x _n], N represents the number of training signal sample, x _i∈ R ^{7 × 1}(i=1,2 ..., N) and represent i-th training signal sample;

In the present embodiment, get N number of training signal as sample, namely N number of training signal sample is obtained, this N number of training signal constitutes a sample of signal matrix X, wherein, each the training signal sample matrix in X belongs to (7 kinds of primitive character parameters that corresponding previous calculations goes out) of 7 dimensions;

Build similar adjacent map G={Z, B}, wherein, vertex set Z={x ₁, x ₂x _n, weight matrix B ∈ R ^{n × N};

Similar adjacent map is for representing the space structure relation between the training signal sample in sample of signal matrix.Vertex set Z in similar adjacent map is the training signal sample matrix composition in sample of signal matrix X, and namely each training signal sample can be regarded as a summit.Weight matrix B has measured the similitude (i.e. coefficient) in similar adjacent map between summit, the weight coefficient referred in each summit and this set between other all summits of the similitude between summit.

Then, the target function constructing the embodiment of the present invention is as follows:

Wherein, weight matrix B is a symmetrical matrix, and its i-th row jth column element is defined as follows:

The feature of symmetrical matrix is: element take leading diagonal as symmetry axis correspondent equal.In symmetrical matrix B, the value of its i-th row jth column element has two kinds of situations: a kind of situation is: namely work as x _iat x _jk neighborhood in, or x _jat x _ik neighborhood in time, B _ijvalue for calculating 2*exp (-d ²(x _i, x _j)/t)-1 value, result of calculation is defined as B _ijvalue.In addition, 2*exp (-d ²(x _i, x _j)/t) d in-1 represents distance, namely calculates x _iand x _jbetween distance, t is a suitable constant, when practical application, can calculate B according to the value (such as, 1,10,100) by choosing different t _ijdiscrimination, and get and make B _ijthe t value corresponding when reaching maximum of discrimination.Here k neighborhood can be understood as k nearest sample point.

The second situation is: work as x _inot at x _jk neighborhood in, or x _jnot at x _ik neighborhood in time, B _ijvalue be 0.

Y in target function _itraining signal sample x _ilow dimension formulation, α represents projecting direction, makes y=α ^tx brings target function into, obtains:

Wherein, L=D-B is Laplacian Matrix;

D is a diagonal matrix, the element on its diagonal correspond to each row of weight matrix B or every a line and, namely or

XLX ^trepresent the local dispersion matrix of training signal sample.

In order to avoid the impact of yardstick, increase a constraints to target function and make α ^txDX ^tα=1, in conjunction with described constraints, target function can be write as:

Solve following characteristics equation:

αXLX ^Tα＝λXDX ^Tα

If A=XLX ^t, B=XDX ^t, then λ is B ^-1a corresponds to the characteristic value of characteristic vector α;

Optimum projection matrix α=[α ₁α ₂α _l], the column vector of α is by B ^-1front l the eigenvalue of maximum characteristic of correspondence vector composition of A, wherein, l represents the quantity of projecting direction.In practical application, usually need l (l >=2) individual projecting direction.

In one embodiment of the invention, the quantity of projecting direction is 2, namely sets 2 projecting directions, gets B ^-1the characteristic vector of front 2 eigenvalue of maximum composition of A as the column vector of the optimum projection matrix α in one embodiment of the invention, thus obtains the optimum projection matrix of the present embodiment.After obtaining optimum projection matrix, this optimum projection matrix is utilized to project in low-dimensional proper subspace by training signal sample, owing to utilizing the optimum projection matrix calculated through locality preserving projections to carry out feature extraction and dimensionality reduction, so training signal sample to have in better class discreteness between compactness and class.Namely the training signal sample that signal modulation style is identical substantially all can project in the region of a close together, and distant between the different training signal sample of signal modulation style.

In one embodiment of the invention, the signal to be identified in calculating low-dimensional proper subspace and the Euclidean distance between the known training signal of signal modulation style, the nearest neighbor algorithm based on Euclidean distance determines that the signal modulation style of signal to be identified comprises:

Utilize optimum projection matrix α by signal x to be identified ^*project in low-dimensional proper subspace, obtain signal x to be identified ^*low dimension formulation y in low-dimensional proper subspace ^*;

Utilize optimum projection matrix to carry out feature extraction to each training signal in sample of signal matrix, training signal is projected in low-dimensional proper subspace; Here, training signal sample x ^kall project in a low-dimensional proper subspace with signal to be identified, so can reduce signal dimension, be equivalent to the point in higher dimensional space, project in a lower dimensional space, amount of calculation can be greatly reduced like this, simplify computational process.

Then, signal x to be identified is calculated ^*with the training signal x that signal modulation style is known ^kbetween Euclidean distance;

If Euclidean distance meets following formula, then determine signal x to be identified ^*with the training signal x that modulation type is known _kbelong to same class;

Wherein, d (y ^*, y _i)=|| y ^*-y _i|| ², (i=1,2 ..., k).

Through said process, this signal modulate method of the present invention can determine the signal modulation style of a signal to be identified fast, easily.It should be noted that, when calculating the Euclidean distance between signal to be identified and training signal sample, be by signal x to be identified ^*and training signal sample x ^kall project in low-dimensional proper subspace, in this proper subspace, then calculate the some y that signal to be identified is corresponding ^*the point y corresponding with training signal sample _ieuclidean distance, thus obtain the signal x to be identified in higher dimensional space ^*signal modulation style.

In order to verify and further illustrate the beneficial effect of this identification of signal modulation method of the embodiment of the present invention, below describing utilizes software Matlab to carry out process and the result of emulation experiment to this identification of signal modulation method of the present invention, and the recognition result of the recognition result of identification of signal modulation method of the present invention and PCA method of the prior art is carried out contrast illustrates.

The emulation experiment process that the embodiment of the present invention provides is: produce binary system amplitude-shift keying 2ASK, quaternary amplitude-shift keying 4ASK, binary phase shift keying 2PSK, quaternary phase shift keying 4PSK, Binary Frequency Shift Keying 2FSK, quaternary frequency shift keying 4FSK, octal system quadrature amplitude modulation 8QAM, the hexadecimal quadrature Modulation and Amplitude Modulation 16QAM digital modulation signals that 8 classes are conventional altogether by Matlab software.Wherein, signal(-) carrier frequency f _c=100KHz, sample frequency f _s=3000KHz, chip rate r _b=10240bit/s.Random generation 200 symbols are as baseband signal, and the frequency deviation of 2FSK and 4FSK is respectively 50KHz and 25KHz.After modulation signal produces, add Gaussian random white noise.In order to meet requirement of engineering, raised cosine roll-off function is adopted to form process to baseband signal.The each random generation of 8 class modulation signal 500 samples, 4000 modulation signal samples are used for training and testing altogether, extract above-mentioned 7 characteristic parameters to characterize each sample of signal.In every class, random selecting 30 is as training sample, and remaining is as test sample book.

This emulation experiment is under different signal to noise ratio condition, and such as [-30-20-1001020] db tests the performance of the signal modulate method of the embodiment of the present invention, and contrasts with the recognition methods based on PCA PCA.Fig. 3 is the employing characteristic vector number of one embodiment of the invention when being 2, and training signal sample is schemed in the drop shadow effect of low-dimensional proper subspace; See Fig. 3, under the condition of signal to noise ratio snr=10db, wherein neighborhood k=1, when front 2 characteristic vectors of getting eigenmatrix are as optimum projection matrix, training sample is in the drop shadow effect of proper subspace.In Fig. 3, 31 represent that signal modulation style is drop shadow effect's signal of the training signal sample point of 2ASK, 32 represent that signal modulation style is drop shadow effect's signal of the training signal sample point of 4ASK, 33 represent that signal modulation style is drop shadow effect's signal of the training signal sample point of 16QAM, 34 represent that signal modulation style is drop shadow effect's signal of the training signal sample point of 4PSK, 35 represent that signal modulation style is drop shadow effect's signal of the training signal sample point of 2FSK, 36 represent that signal modulation style is drop shadow effect's signal of the training signal sample point of 4FSK, 37 represent that signal modulation style is drop shadow effect's signal of the training signal sample point of 2PSK, 38 represent that signal modulation style is drop shadow effect's signal of the training signal sample point of 8QAM.As can be seen from Figure 3, except indivedual 2ASK and 4ASK two class training signal sample, in the class of all the other signal modulation styles, compactness is all fine, can significantly be distinguished.By locality preserving projections, in luv space (i.e. 7 dimension spaces), after the projection of nearer sample point (training signal), still keep close, thus make all kind of modulations signal distribution within class compacter.

Emulation experiment tests the discrimination of identification of signal modulation method of the present invention under different signal to noise ratio, and contrast with the Modulation Identification method of Based PC A, Fig. 4 is that the discrimination of identification of signal modulation method of the present invention and prior art PCA method contrasts schematic diagram; See Fig. 4, identification of signal modulation method of the present invention and the contrast of prior art PCA method on overall discrimination.The discrimination under different signal to noise ratio of 41 expression identification of signal modulation method of the present invention; The discrimination of PCA method under different signal to noise ratio of 42 expression prior aries; As can be seen from Figure 4, along with the improvement of signal to noise ratio, the discrimination of two kinds of methods all has greatly improved, but signal modulate method of the present invention obtains better recognition effect when low signal-to-noise ratio, overall performance is better, has better robustness.

Fig. 5 be prior art PCA method under signal to noise ratio is the condition of 10db, for the discrimination result schematic diagram acquired by unlike signal modulation type; Fig. 6 is a kind of identification of signal modulation method that one embodiment of the invention provides is under the condition of 10db in signal to noise ratio, for the discrimination result schematic diagram acquired by unlike signal modulation type.As can be known from Fig. 5 and Fig. 6, the stability of this signal modulate method of the embodiment of the present invention is better, and the signal being very particularly 2ASK, 4ASK to signal modulation style can obtain higher discrimination (be respectively: 96.0 and 98.3 and discrimination that in prior art, PCA method is corresponding only has 82.1 and 91.3).

To sum up, through emulation experiment contrast find, in non-cooperative communication system and Low SNR under method of the present invention higher compared to existing Modulation Identification method discrimination, stability and robustness better.

Corresponding with above-mentioned identification of signal modulation method, present invention also offers a kind of identification of signal modulation system.Fig. 7 is the block diagram of a kind of identification of signal modulation system that one embodiment of the invention provides, and see Fig. 7, this identification of signal modulation system 700 of one embodiment of the invention comprises:

Pretreatment unit 710, carrying out preliminary treatment for treating identification signal, extracting the characteristic parameter of predetermined quantity, and the characteristic vector using the characteristic parameter of described predetermined quantity to form characterizes described signal to be identified;

Feature extraction unit 720, for utilizing optimum projection matrix to carry out feature extraction to described signal to be identified, projects in low-dimensional proper subspace by described signal to be identified; Wherein, described optimum projection matrix is obtained by locality preserving projections algorithm;

Judgement output unit 730, for calculating signal to be identified in described low-dimensional proper subspace and the Euclidean distance between the known training signal of signal modulation style, the nearest neighbor algorithm based on Euclidean distance determines the signal modulation style of signal to be identified.

In one embodiment of the invention, system 700 also comprises:

Optimum projection matrix training unit, preliminary treatment is carried out for the training signal known to signal modulation style, extract the characteristic parameter of predetermined quantity, and the characteristic vector of use characteristic parameter composition characterizes training signal, obtains the sample of signal matrix comprising each modulation type training signal; Utilize locality preserving projections algorithm, sample of signal matrix is processed, calculate optimum projection matrix.

It should be noted that, this identification of signal modulation system of the present invention is corresponding with aforesaid identification of signal modulation method, thus in the present embodiment, the course of work of identification of signal modulation system can illustrating see aforementioned signal modulation identification method part, do not repeat them here.

In sum, this identification of signal modulation method of the present invention utilizes the optimum projection matrix obtained in advance, is projected in low-dimensional proper subspace by signal to be identified, and in this proper subspace, calculate the modulation type determining signal to be identified.This with directly calculate in higher dimensional space, judge signal to be identified modulation type scheme compared with, amount of calculation can be greatly reduced, improve the efficiency of identification of signal modulation and stability.In addition, of the present inventionly obtain optimum projection matrix by locality preserving projections algorithm, namely the sample of signal matrix of locality preserving projections algorithm to the training signal place comprising each modulation type is utilized to carry out locality preserving projections calculating, in projection process, not only maintain the otherness feature of training signal sample, also maintain the local similarity feature of training signal sample, thus there is robustness preferably, even if make, signal to noise ratio more at signal modulation style to be identified also can obtain higher discrimination lower.And, present invention also offers a kind of identification of signal modulation system corresponding with this identification of signal modulation method, this identification of signal modulation system also has the advantage that discrimination is high, robustness better, performance is more stable of above-mentioned identification of signal modulation method.

The foregoing is only preferred embodiment of the present invention, be not intended to limit protection scope of the present invention.All any amendments done within the spirit and principles in the present invention, equivalent replacement, improvement etc., be all included in protection scope of the present invention.

Claims (10)

1. an identification of signal modulation method, is characterized in that, described method comprises:

Treat identification signal and carry out preliminary treatment, extract the characteristic parameter of predetermined quantity, and the characteristic vector using the characteristic parameter of described predetermined quantity to form characterizes described signal to be identified;

Utilize optimum projection matrix to carry out feature extraction to described signal to be identified, described signal to be identified is projected in low-dimensional proper subspace; Wherein, described optimum projection matrix is obtained by locality preserving projections algorithm;

Calculate signal to be identified in described low-dimensional proper subspace and the Euclidean distance between the known training signal of signal modulation style, the nearest neighbor algorithm based on Euclidean distance determines the signal modulation style of signal to be identified.

2. identification of signal modulation method as claimed in claim 1, is characterized in that, treating before identification signal carries out preliminary treatment, the method also comprises: training obtains optimum projection matrix; Described training obtains optimum projection matrix and comprises:

The pre-treatment step same with described signal to be identified is carried out to the training signal that signal modulation style is known, extract the characteristic parameter of described predetermined quantity, and the characteristic vector using described characteristic parameter to form characterizes the known training signal of described signal modulation style, obtain the sample of signal matrix comprising each modulation type training signal;

Utilize locality preserving projections algorithm, described sample of signal matrix is processed, calculates optimum projection matrix.

3. identification of signal modulation method as claimed in claim 2, it is characterized in that, the described training signal known to signal modulation style carries out the pre-treatment step same with described signal to be identified, and the characteristic parameter extracting described predetermined quantity comprises:

Step 31, training signal s (t) received is sampled, obtains instantaneous amplitude a (i) of described training signal s (t), instantaneous phase φ (i), instantaneous frequency f (i) and power spectrum S (i); Wherein, sample frequency is f _s, sampling number is N _s;

Step 32, calculates the standard deviation sigma of normalization instantaneous amplitude _a, the standard deviation of normalization instantaneous amplitude and the ratio R of average _aand the divergence μ of normalization instantaneous amplitude ^a;

Step 33, calculates the standard deviation sigma of normalization instantaneous frequency _f, normalization instantaneous frequency divergence μ ^fand the standard deviation sigma of normalized nonlinear instantaneous phase _φ;

Step 34, the symmetry P of rated output spectrum;

The characteristic parameter of described predetermined quantity comprises totally 7 characteristic parameters that step 32, step 33 and step 34 calculate.

4. identification of signal modulation method as claimed in claim 3, is characterized in that, described step 32, calculates the standard deviation sigma of normalization instantaneous amplitude _a, the standard deviation of normalization instantaneous amplitude and the ratio R of average _aand the divergence μ of normalization instantaneous amplitude ^acomprise:

Utilize the standard deviation sigma of following formulae discovery normalization instantaneous amplitude _a:

${\mathrm{\σ}}_a=\sqrt{\frac{1}{N_s}\left|\underset{i=1}{\overset{N_s}{\Σ}}{a}_n^2\left(i\right)\right|-{\left|\frac{1}{N_s}\underset{i=1}{\overset{N_s}{\Σ}}{a}_n\left(i\right)\right|}^2}$ Wherein, $a_n\left(i\right)=a\left(i\right)/\stackrel{\‾}{a};$

Utilize the standard deviation of following formulae discovery normalization instantaneous amplitude and the ratio R of average _a:

$R_a=\frac{{\mathrm{\σ}}_a}{\stackrel{\‾}{a}}$ Wherein, $\stackrel{\‾}{a}=E\[a\left(i\right)\]=N_s^{-1}\underset{i=1}{\overset{N_s}{\Σ}}a\left(i\right);$

Utilize the divergence μ of following formulae discovery normalization instantaneous amplitude ^a:

${\mathrm{\μ}}^a=E\[a_n^4\left(i\right)\]/{\{E\[a_n^2\left(i\right)\]\}}^2.$

5. identification of signal modulation method as claimed in claim 3, is characterized in that, described step 33, calculates the standard deviation sigma of normalization instantaneous frequency _f, normalization instantaneous frequency divergence μ ^fand the standard deviation sigma of normalized nonlinear instantaneous phase _φcomprise:

Utilize the standard deviation sigma of following formulae discovery normalization instantaneous frequency _f:

wherein, f _n=f (i)/f _max, f _max=max{f (i) };

Utilize the divergence μ of following formulae discovery normalization instantaneous frequency ^f:

${\mathrm{\μ}}^f=E\[f_n^4\left(i\right)\]/{\{E\[f_n^2\left(i\right)\]\}}^2$

Utilize the standard deviation sigma of following formulae discovery normalized nonlinear instantaneous phase _φ:

${\mathrm{\σ}}_{\mathrm{\φ}}=\sqrt{\frac{1}{N_s}\left|\underset{i=1}{\overset{N_s}{\Σ}}{\mathrm{\φ}}_c^2\left(i\right)\right|-{\left|\frac{1}{N_s}\underset{i=1}{\overset{N_s}{\Σ}}{\mathrm{\φ}}_c\left(i\right)\right|}^2}$ Wherein, ${\mathrm{\φ}}_c\left(i\right)=\mathrm{\φ}\left(i\right)-N_s^{-1}\underset{i=1}{\overset{N_s}{\Σ}}\mathrm{\φ}\left(i\right),$ The non-linear partial that φ (i) is instantaneous phase.

6. identification of signal modulation method as claimed in claim 3, is characterized in that, described step 34, and the symmetry P of rated output spectrum comprises:

Utilize the symmetry P of following formulae discovery power spectrum:

$P=\frac{P_L-P_U}{P_L+P_U}$

Wherein, $P_L=\underset{i=1}{\overset{N_{fc}}{\Σ}}{\left|S\left(i\right)\right|}^2,$ $P_U={\underset{i=1}{\overset{N_{fc}}{\Σ}}\left|S\left(i+N_{fc}+1\right)\right|}^2,$ S(i)＝DFT[s(i)]，N _fc＝f _cN _s/(f _s-1)。

7. identification of signal modulation method as claimed in claim 6, is characterized in that, describedly utilizes locality preserving projections algorithm, processes, calculate optimum projection matrix and comprise described sample of signal matrix:

If sample of signal matrix X=is [x ₁, x ₂..., x _n], N represents the number of training signal sample, x _i∈ R ^{7 × 1}(i=1,2 ..., N) and represent i-th training signal sample;

Build similar adjacent map G={Z, B}, wherein, vertex set Z={x ₁, x ₂x _n, weight matrix B ∈ R ^{n × N};

Structure target function is as follows:

$min\underset{ij}{\Σ}{\left(y_i-y_j\right)}^2{B}_{ij}$

Wherein, weight matrix B is a symmetrical matrix, and its i-th row jth column element is defined as follows:

x _iat x _jk neighborhood in, or x _jat x _ik neighborhood in

Y _itraining signal sample x _ilow dimension formulation, α represents projecting direction, makes y=α ^tx brings target function into, obtains:

$\begin{array}{c}\frac{1}{2}\underset{ij}{\mathrm{\Σ}}{\left(y_i-y_j\right)}^2{B}_{ij}=\frac{1}{2}{\mathrm{\Σ}}_{ij}{\left({\mathrm{\α}}^T{x}_i-{\mathrm{\α}}^T{x}_j\right)}^2{B}_{ij}\\ =\[{\mathrm{\Σ}}_i{\mathrm{\α}}^T{x}_i{D}_{ij}{x_i}^T\mathrm{\α}-{\mathrm{\α}}^T{\mathrm{XBX}}^T\mathrm{\α}\]\\ =\[{\mathrm{\α}}^T{\mathrm{XDX}}^T\mathrm{\α}-{\mathrm{\α}}^T{\mathrm{XBX}}^T\mathrm{\α}\]\\ ={\mathrm{\α}}^T{\mathrm{XLX}}^T\mathrm{\α}\end{array}$

Wherein, L=D-B is Laplacian Matrix;

D is a diagonal matrix, $D_{ii}=\underset{j}{\Σ}{B}_{ij}$ Or $D_{ii}=\underset{j}{\Σ}{B}_{ji};$

XLX ^trepresent the local dispersion matrix of training signal sample.

Increase a constraints to described target function and make α ^txDX ^tα=1, in conjunction with described constraints, described target function can be write as:

$f\left(\mathrm{\α}\right)=\arg \underset{{\mathrm{\α}}^T\mathrm{\α}=1}{min}\left(\frac{{\mathrm{\α}}^T{\mathrm{XLX}}^T\mathrm{\α}}{{\mathrm{\α}}^T{\mathrm{XDX}}^T\mathrm{\α}}\right)$

Solve following characteristics equation:

αXLX ^Tα＝λXDX ^Tα

If A=XLX ^t, B=XDX ^t, then λ is B ^-1a corresponds to the characteristic value of characteristic vector α;

Optimum projection matrix α=[α ₁α ₂α _l], the column vector of α is by B ^-1front l the eigenvalue of maximum characteristic of correspondence vector composition of A, wherein, l represents the quantity of projecting direction.

8. identification of signal modulation method as claimed in claim 7, it is characterized in that, signal to be identified in described calculating described low-dimensional proper subspace and the Euclidean distance between the known training signal of signal modulation style, the nearest neighbor algorithm based on Euclidean distance determines that the signal modulation style of signal to be identified comprises:

Utilize optimum projection matrix α by signal x to be identified ^*project in described low-dimensional proper subspace, obtain described signal x to be identified ^*low dimension formulation y in low-dimensional proper subspace ^*;

Utilize described optimum projection matrix to carry out feature extraction to each training signal in described sample of signal matrix, described training signal is projected in low-dimensional proper subspace;

Calculate signal x to be identified ^*with the training signal x that signal modulation style is known _kbetween Euclidean distance;

If described Euclidean distance meets following formula, then determine signal x to be identified ^*with the training signal x that modulation type is known _kbelong to same class;

$d(y_k,y^{*})=\underset{i}{\min }\left\{d\left(y_i,y^{*}\right)\right\}$

Wherein, d (y ^*, y _i)=|| y ^*-y _i|| ², (i=1,2 ..., k).

9. an identification of signal modulation system, is characterized in that, described system comprises:

Pretreatment unit, carrying out preliminary treatment for treating identification signal, extracting the characteristic parameter of predetermined quantity, and the characteristic vector using the characteristic parameter of described predetermined quantity to form characterizes described signal to be identified;

Feature extraction unit, for utilizing optimum projection matrix to carry out feature extraction to described signal to be identified, projects in low-dimensional proper subspace by described signal to be identified; Wherein, described optimum projection matrix is obtained by locality preserving projections algorithm;

Judgement output unit, for calculating signal to be identified in described low-dimensional proper subspace and the Euclidean distance between the known training signal of signal modulation style, the nearest neighbor algorithm based on Euclidean distance determines the signal modulation style of signal to be identified.

10. identification of signal modulation system as claimed in claim 9, it is characterized in that, described system also comprises:

Optimum projection matrix training unit, preliminary treatment is carried out for the training signal known to signal modulation style, extract the characteristic parameter of described predetermined quantity, and the characteristic vector using described characteristic parameter to form characterizes described training signal, obtain the sample of signal matrix comprising each modulation type training signal; Utilize locality preserving projections algorithm, described sample of signal matrix is processed, calculates optimum projection matrix.