CN105721371A - Method for identifying common digital modulation signal based on cyclic spectrum correlation - Google Patents

Abstract

The invention discloses a method for identifying a common digital modulation signal based on cyclic spectrum correlation. The reliability of signal analysis is improved by utilizing the noise-proof feature of a signal cyclic spectrum; the steps of alpha section wavelet de-noising and averaging through superposition are introduced into a calculation process of a signal spectral correlation function, so that the random fluctuation caused by the limited sampling number and the external disturbance in an original spectrum correlation estimation algorithm result is effectively weakened to facilitate identification and extraction of modulation features; and meanwhile, an alpha section and an f section of an obtained spectral correlation diagram are computed by utilizing signal spectral correlation, and appropriate features and parameters (such as a ratio of maximum absolute values of spectral correlation functions, namely the alpha section and the f section, the number of intense lines of the alpha section, a coefficient of fluctuation of the alpha section, the normalized area of the f section, a predominance ratio of spectral lines of the alpha section and the like) are selected to construct a classification method to identify the modulation mode of the communication signal.

Description

Common digital modulation signal identification method based on cyclic spectrum correlation

Technical Field

The invention belongs to the field of communication signal modulation identification, and particularly relates to a common digital modulation signal identification method based on cyclic spectrum correlation.

Technical Field

The identification of the modulation scheme of the communication signal is an important link between signal acquisition and demodulation, and plays an important role in the military and civil communication fields, especially in the management of dynamic radio spectrum and the identification of unknown interference. In addition, the modulation mode identification of the communication signal is also the basis for constructing software radio or cognitive radio application, and technical support is provided for multi-system communication interconnection application.

Fourier transform-based signal power spectrum analysis is a more classical analysis method for stationary signals. Through the power spectrum analysis of the signal, the basic characteristics of the analyzed signal, including signal parameters such as carrier frequency, frequency bandwidth and the like, can be well acquired. However, in the communication process, a large part of signals are converted into non-stationary signals due to manual intervention (such as modulation, coding, scanning, and the like), and the modulation characteristics of the signals cannot be more fully revealed by power spectrum analysis, so that the modulation identification method based on the power spectrum analysis usually needs to use more time domain statistical characteristics, and has larger limitations. The theory related to the cyclic spectrum proposed by Gardner is an important tool for researching non-stationary signals in recent years, fully considers the important property of the signals after periodic artificial interference, namely the cyclic stability, and can more comprehensively disclose the modulation characteristics of communication signals, which is incomparable to classical power spectrum analysis.

Although the theory of the cyclic spectrum correlation is widely applied to the analysis of non-stationary signals in various fields, the theory of the cyclic spectrum correlation is not popularized in the field of modulation and identification of communication signals at present.

Disclosure of Invention

The technical problem to be solved by the invention is as follows: the method overcomes the defects of the prior art, provides a common digital modulation signal identification method based on cyclic spectrum correlation, and improves the reliability of signal analysis by utilizing the anti-noise characteristic of the cyclic spectrum of the signal; the links of alpha section wavelet denoising and superposition averaging are introduced in the calculation process of the signal spectrum correlation function, so that the random fluctuation caused by limited sampling points and external interference in the original spectrum correlation estimation algorithm result is effectively weakened, and the identification and extraction of modulation characteristics are facilitated; meanwhile, an alpha section and an f section of the obtained spectrum correlation diagram are calculated by utilizing signal spectrum correlation, and a classification method is established by selecting proper characteristics and parameters (such as the maximum absolute value ratio of the alpha section and the f section of a spectrum correlation function, the number of strong spectral lines of the alpha section, the fluctuation coefficient of the alpha section, the normalized area of the f section, the significance ratio of the spectral lines of the alpha section and the like) to identify the modulation mode of the communication signal.

The technical scheme adopted by the invention for solving the technical problems is as follows: a common digital modulation signal identification method based on cyclic spectrum correlation comprises the following steps:

(1) dividing the intercepted signal-to-noise ratio modulation signal into n equal parts after sampling, and taking the n equal parts as input signals for performing spectrum correlation operation;

(2) selecting proper Fourier transform point number and smooth window width, and respectively carrying out spectrum correlation operation on the n signals by using a spectrum correlation estimation method based on frequency domain smoothing;

the spectral correlation estimation algorithm based on frequency domain smoothing is as follows:

$S_{x\mathrm{\Δ}t}^{\mathrm{\α}}{(t,f)}_{\mathrm{\Δ}f}=\frac{1}{M}\underset{v=-(M-1)/2}{\overset{(M-1)/2}{\Σ}}\frac{1}{\mathrm{\Δ}t}{X}_{\mathrm{\Δ}t}(t,f+\frac{\mathrm{\α}}{2}+{\mathrm{vF}}_S)X_{\mathrm{\Δ}t}^{*}(t,f-\frac{\mathrm{\α}}{2}+{\mathrm{vF}}_S)$

$X_{\mathrm{\Δ}t}(t,f)=\underset{K=0}{\overset{N-1}{\Σ}}{a}_{\mathrm{\Δ}t}\left({\mathrm{KT}}_S\right)x_k(t-{\mathrm{KT}}_S)\exp (-j2\mathrm{\π}f(t-{\mathrm{KT}}_S\left)\right)$

wherein,is the result after the spectrum correlation operation; x_Δt(t, f) is the signal x_k(t) the result of the short-time fourier transform; signal x_k(t) is one of n equal parts of the intercepted signal; Δ t is x_k(t) a duration of time; a is_ΔtIs a window function; Δ f is the frequency domain smoothing interval; f_SIs the frequency domain minimum increment unit; t is_SIs the time domain sampling interval; n is the signal x_k(t) number of samples, M is the smoothing window width factor, α is the cycle frequency, f is the frequency, t is the time;

(3) the operation results in the step (2) are subjected to alpha section wavelet denoising and then are added to obtain an average value, and an output result of the improved spectral correlation estimation calculation method is obtained;

the improved spectral correlation estimation algorithm is as follows:

$S_{x\mathrm{\Δ}t}^{\mathrm{\α}}\left(f\right)=\frac{1}{N}\underset{k=1}{\overset{N}{\Σ}}{\mathrm{WDN}}_{\mathrm{\α}}\[S_{x\mathrm{\Δ}t}^{\mathrm{\α}}(t_k,f)\]$

wherein WDN_α(. represents a pairThe cyclic frequency α is subjected to wavelet denoising;

(4) extracting 5 modulation identification characteristics according to the α cross section and the f cross section of the spectrum correlation function obtained in the step (3), and identifying several digital modulation signals by combining a signal characteristic construction and classification method based on time domain statistics, wherein the 5 modulation identification characteristics are the maximum absolute value ratio (marked as R) of the α cross section and the f cross section of the spectrum correlation function₁) α number of strong spectral lines in section (denoted as R)₂) α section undulation coefficient (noted as R)₃) F normalized area of cross section (denoted as R)₄) α ratio of the significance of the spectral lines in the cross-section (denoted as R)₅) (ii) a The statistical characteristics based on the signal time domain are as follows: zero center normalizationStandard deviation of absolute value of instantaneous amplitude (denoted as R)₆) (ii) a The digital modulation signals comprise 2ASK, 4ASK, BPSK, QPSK, 8PSK, MSK, 2FSK, 4FSK, 2FSK and 4FSK, wherein FSK represents frequency shift keying with uncorrelated initial phases of code elements;

the identification process for several digital modulation signals is as follows:

if it isJudging the modulation mode of the signal to be MSK;

if it is $\{{\hat{R}}_1>r_{12},{\hat{R}}_2>r_{21}\}$ Or $\{{\hat{R}}_1<r_{12},{\hat{R}}_3<r_{31}\},$ Determining that the modulator of the signal is 4 FSK;

if it isJudging that the modulation mode of the signal is 2 FSK;

if it isJudging the modulation mode of the signal to be BPSK;

if it is $\{{\hat{R}}_1>r_{11},{\hat{R}}_2<r_{22},{\hat{R}}_5<r_{51},{\hat{R}}_6<r_{61}\},$ Judging that the modulation mode of the signal is 2 ASK;

if it is $\{{\hat{R}}_1>r_{11},{\hat{R}}_2<r_{22},{\hat{R}}_5<r_{51},{\hat{R}}_6>r_{61}\},$ Judging that the modulation mode of the signal is 4 ASK;

if it isDetermining the modulation mode 2FSK of the signal;

if it isJudging the modulation mode of the signal to be 4 FSK;

if it is $\{{\hat{R}}_1<r_{12},{\hat{R}}_3>r_{31},{\hat{R}}_4<r_{42},{\hat{R}}_5>r_{52}\},$ Determining the modulation mode QPSK of the signal;

if it is $\{{\hat{R}}_1<r_{12},{\hat{R}}_3>r_{31},{\hat{R}}_4<r_{42},{\hat{R}}_5<r_{52}\},$ Judging the modulation mode 8PSK of the signal;

whereinIs a characteristic R₁Estimated value of r₁₁And r₁₂Is corresponding to the characteristic R₁A threshold value of (d); the rest is the same.

The sampling frequency of the signals in the step (1) meets the requirement

In the step (2), the number of sampling points of each signal is N-1024, and the smoothing window width coefficient is M-63.

The principle of denoising the small waves in the step (3) is as follows: the noisy sequence is decomposed using the sym8 wavelet, and at the fifth layer of the decomposition, the sequence is denoised using a soft sure threshold selection method, and the threshold is adjusted with the noise variance of the first layer wavelet decomposition.

The 5 modulation identification characteristics based on the modulation signal spectrum correlation function in the step (4) are defined as follows:

① spectral correlation function α section and f section maximum absolute value ratio R₁Is defined as:

$R_1=\underset{N\&RightArrow;\mathrm{\&infin;}}{\lim }\frac{1}{N}\underset{n=1}{\overset{N}{\&Sigma;}}\frac{\max \left(\right|S_{xn}^{\mathrm{\&alpha;}}\left(0\right)\left|\right)}{\max \left(\right|S_{xn}^0\left(f\right)\left|\right)}$

whereinAndare respectively a spectral correlation functionα cross section and f cross section functions of;

② α number of strong lines in cross section R₂Is defined as:

$R_2=\underset{N\&RightArrow;\mathrm{\&infin;}}{\lim }\frac{1}{N}\underset{n=1}{\overset{N}{\&Sigma;}}count({\mathrm{\&rho;}}_n\&GreaterEqual;{\mathrm{\&rho;}}_{th}*{\mathrm{\&rho;}}_{max})$

where count (-) represents the number of significant peaks, ρ, for cross-section of the spectral correlation map α_nIs the degree of significance, rho, of the cross-sectional peak_maxIs the maximum significance, rho, of the cross-sectional peak_thTaking a fixed value of 0.27 as a significance threshold value;

wherein the peak saliency ρ is defined as:

$\mathrm{\&rho;}=\frac{h^2}{l*max\left(h\right)}$

wherein h is the difference between the amplitude of the wave peak in the alpha section of the spectrum correlation diagram and the larger of the amplitudes of two adjacent wave troughs, l is the width value of the wave peak, and max (h) is the maximum h value in the alpha section;

③ α coefficient of section fluctuation R₃Is defined as:

$R_3=\underset{N\&RightArrow;\mathrm{\&infin;}}{\lim }\frac{1}{N}\underset{n=1}{\overset{N}{\&Sigma;}}count({\mathrm{\&beta;}}_n\&GreaterEqual;{\mathrm{\&beta;}}_{th}*{\mathrm{\&beta;}}_{max})$

wherein count (·) represents that the section waviness of the statistical spectral correlation function α is more than β_th*β_maxNumber of peaks of β_nUndulation of the crest of the cross-section, β_maxMaximum waviness of the peaks of the cross-section, β_thTaking a fixed value of 0.1 as a fluctuation threshold value;

wherein the peak undulation β is defined as:

$\mathrm{\&beta;}=\frac{h}{l}$

wherein h and l have the same meanings as defined in (ii);

④ f normalized area of cross section R₄Is defined as:

$R_4=\underset{N\&RightArrow;\mathrm{\&infin;}}{\lim }\frac{1}{N}\underset{n=1}{\overset{N}{\&Sigma;}}\left\{\frac{1}{max\left(S_{xn}^0\right(f\left)\right)}{\&Integral;}_{-f_0}^{f_0}\right|S_{xn}^0\left(f\right)\left|df\right\}$

whereinMeans that the maximum of the cross-section function of the spectral correlation function f is found,the area of the f section is obtained;

⑤ α ratio of significance of cross-sectional spectral lines R₅Is defined as:

$R_5=\underset{N\&RightArrow;\mathrm{\&infin;}}{\lim }\frac{1}{N}\underset{n=1}{\overset{N}{\&Sigma;}}\frac{\sec \left({\mathrm{\&rho;}}_n\right)}{max\left({\mathrm{\&rho;}}_n\right)}$

where max (p)_n) The maximum significance value of the spectral line in section of α, sec (ρ)_n) The next largest significance value for the α cross-sectional spectral line;

the standard deviation R of the absolute value of the zero-center normalized instantaneous amplitude of the statistical characteristics based on the signal time domain in the step (4)₆Is defined as:

$\mathrm{\&delta;}=\sqrt{\frac{1}{N}\&lsqb;\underset{i=1}{\overset{N}{\&Sigma;}}{A}_{cn}^2\left(i\right)\&rsqb;-{\&lsqb;\frac{1}{N}\underset{i=1}{\overset{N}{\&Sigma;}}\left|A_{cn}\left(i\right)\right|\&rsqb;}^2}$

wherein N is the number of observation samples, A_cn(i) Normalized instantaneous amplitude for zero center, defined as A_cn(i)＝A_n(i)-1；

WhereinDivisorAnd A (i) is the sample point instantaneous amplitude;

the threshold value of each debugging identification characteristic in the step (4) is as follows:

characteristic R₁The threshold value of (2): r is₁₁＝0.67，r₁₂0.36; characteristic R₂The threshold value of (2): r is₂₁＝4.8，r₂₂3.2; characteristic R₃The threshold value of (2): r is₃₁60.46; (ii) a Characteristic R₄The threshold value of (2): r is₄₁＝903.51，r₄₂498.54; characteristic R₅The threshold value of (2): r is₅₁＝0.073，r₅₂＝7.95×10^-4(ii) a Characteristic R₆The threshold value of (2): r is₆₁＝0.29。

The principle of the invention is as follows: the theory of the cyclic spectrum correlation is a non-stationary signal analysis method and is suitable for analyzing communication modulation signals with cyclostationarity. Aiming at the random fluctuation phenomenon caused by external interference and limited sample points in the output result of the original spectrum correlation estimation algorithm, a link of wavelet de-noising and superposition averaging is introduced to optimize the spectrum correlation operation process, so that the random fluctuation of the output result is effectively weakened, and the spectrum characteristics are obviously enhanced. On the basis, 5 identification characteristics based on the cyclic spectrum of the modulation signal can be obtained, and various digital modulation signals are identified by combining a statistical characteristic based on the signal time domain.

The invention has the advantages over the prior art that: compared with the traditional power spectrum analysis, the method inherits the advantages of the cyclic spectrum analysis, can fully explore the modulation characteristics of the communication signals, ensures that the modulation identification process does not depend on the signal time domain statistical characteristics with stronger noise sensitivity excessively, and improves the identification precision when the signal to noise ratio is low; compared with the original spectrum correlation estimation algorithm, the invention introduces links of wavelet de-noising and superposition averaging to optimize the spectrum correlation operation process, thereby weakening the random fluctuation caused by external interference and limited sample points in the output result, strengthening the spectrum characteristics and being beneficial to the extraction of modulation identification characteristics; compared with the identification characteristic form based on signal time domain statistics, the 5 modulation identification characteristics based on the spectral correlation function provided by the invention are simple in form and weak in noise sensitivity.

Drawings

FIG. 1 is a flow chart of a modulation identification method based on cyclic spectrum correlation according to the present invention;

fig. 2 is a flow chart of modulation identification after spectral correlation operation.

Detailed Description

The invention is described in detail below with reference to the figures and the detailed description.

As shown in fig. 1, the method for identifying a common digital modulation signal based on cyclic spectrum correlation according to the present invention comprises the following steps:

(1) because the invention introduces the link of accumulation and averaging in the original spectrum correlation estimation algorithm, the original signal is required to be sampled and then is equally divided, and the sampling frequency is set asT_SIs a time domain sampling interval, and each equal part of signals generally takes 1024 sample points;

(2) the theory of circular spectrum correlation is a theoretical tool proposed by Gardner for non-stationary signal analysis, and is defined by an estimation algorithm based on frequency domain smoothing as follows:

$S_{x\mathrm{\&Delta;}t}^{\mathrm{\&alpha;}}{(t,f)}_{\mathrm{\&Delta;}f}=\frac{1}{M}\underset{v=-(M-1)/2}{\overset{(M-1)/2}{\&Sigma;}}\frac{1}{\mathrm{\&Delta;}t}{X}_{\mathrm{\&Delta;}t}(t,f+\frac{\mathrm{\&alpha;}}{2}+{\mathrm{vF}}_S)X_{\mathrm{\&Delta;}t}^{*}(t,f-\frac{\mathrm{\&alpha;}}{2}+{\mathrm{vF}}_S)---\left(1\right)$

in the above formula x_k(t) is one of n equal parts of the intercepted signal; Δ t is x_k(t) a duration of time; Δ f is the frequency domain smoothing interval; f_SIs the frequency domain minimum increment unit; t is_SIs the time domain sampling interval, M is the smoothing window width coefficient with a fixed value of 63, α is the cycle frequency, f is the frequency, t is the time, X is the frequency_Δt(t, f) is the aliquot signal X in step (1)_Δt(t) a short-time fourier transform, defined as:

$X_{\mathrm{\&Delta;}t}(t,f)=\underset{K=0}{\overset{N-1}{\&Sigma;}}{a}_{\mathrm{\&Delta;}t}\left({\mathrm{KT}}_S\right)x_k(t-{\mathrm{KT}}_S)\exp (-j2\mathrm{\&pi;}f(t-{\mathrm{KT}}_S\left)\right)---\left(2\right)$

in the above formula a_ΔtIs a window function, here taken as a rectangular window function of equal length to the aliquot signal; taking the number of points of short-time Fourier transform as N as 1024;

(3) due to the influence of external interference and limited sample point number, the spectral correlation calculation result in the step (2) has large fluctuation, which is not beneficial to the observation and extraction of weak features, so that the cyclic frequency alpha of the spectral correlation function of the n equal parts of signals is subjected to wavelet denoising and then added to obtain an average, as shown in the following formula:

$S_{x\mathrm{\&Delta;}t}^{\mathrm{\&alpha;}}\left(f\right)=\frac{1}{N}\underset{k=1}{\overset{N}{\&Sigma;}}{\mathrm{WDN}}_{\mathrm{\&alpha;}}\&lsqb;S_{x\mathrm{\&Delta;}t}^{\mathrm{\&alpha;}}(t_k,f)\&rsqb;$

(3)

WDN in the formula_α(. represents a pairThe cyclic frequency α in (1) is wavelet denoised as follows:

y＝wden(x,'heursure','s','sln',5,'sym8')(4)

wherein x and y are discrete sequences before and after denoising, the whole formula represents that a sym8 wavelet is used for decomposing a noisy signal sequence, on the decomposed fifth layer, a soft sure threshold value selection method is used for denoising the sequence, and the threshold value is adjusted along with the noise variance of the wavelet decomposition of the first layer.

(4) By analyzing the spectrum correlation function of the modulation signal, the spectral characteristics are mainly concentrated on the α section and the f section of the spectrum correlation diagram, so that the modulation signal can be identified by extracting the spectral characteristics of the two sections to combine the spectral characteristics into a debugging identification characteristic, and the proposed characteristic is the maximum absolute value ratio R of the α section and the f section of the spectrum correlation function₁α number of strong spectral lines in section R₂α coefficient of section waviness R₃F area of normalized cross section R₄α ratio of significance of cross-sectional spectral lines R₅(ii) a In addition, the modulation identification process also relates to a signal time domain statistical characteristic-standard deviation R of zero-center normalized instantaneous amplitude absolute value₆In order to accurately describe the modulation identification characteristics, the section peak undulation degree β and the saliency rho are introducedAnd are used to describe α background jitter and line intensity, respectively,

wherein the peak undulation β is defined as:

$\mathrm{\&beta;}=\frac{h}{l}---\left(5\right)$

wherein h is the difference between the amplitude of the wave peak in the alpha section of the spectrum correlation diagram and the larger of the amplitudes of two adjacent wave troughs, and l is the width value of the wave peak;

the peak saliency ρ is defined as:

$\mathrm{\&rho;}=\frac{h^2}{l*max\left(h\right)}---\left(6\right)$

wherein max (h) is the maximum value of h in the alpha section; on the basis, the modulation identification characteristics are described as follows:

① spectral correlation function α section and f section maximum absolute value ratio R₁Is defined as:

$R_1=\underset{N\&RightArrow;\mathrm{\&infin;}}{\lim }\frac{1}{N}\underset{n=1}{\overset{N}{\&Sigma;}}\frac{\max \left(\right|S_{xn}^{\mathrm{\&alpha;}}\left(0\right)\left|\right)}{\max \left(\right|S_{xn}^0\left(f\right)\left|\right)}---\left(7\right)$

whereinAndare respectively a spectral correlation functionα cross section and f cross section function of the feature, the threshold value of the feature being taken as r₁₁＝0.67，r₁₂＝0.36；

② α number of strong lines in cross section R₂Is defined as:

$R_2=\underset{N\&RightArrow;\mathrm{\&infin;}}{\lim }\frac{1}{N}\underset{n=1}{\overset{N}{\&Sigma;}}count({\mathrm{\&rho;}}_n\&GreaterEqual;{\mathrm{\&rho;}}_{th}*{\mathrm{\&rho;}}_{max})$

where count (-) represents the number of significant peaks, ρ, for cross-section of the spectral correlation map α_nIs the degree of significance, rho, of the cross-sectional peak_maxIs the maximum significance, rho, of the cross-sectional peak_thTaking a fixed value of 0.27 as a significance threshold value; the threshold of this feature is taken as r₂₁＝4.8，r₂₂＝3.2；

③ α coefficient of section fluctuation R₃Is defined as:

$R_3=\underset{N\&RightArrow;\mathrm{\&infin;}}{\lim }\frac{1}{N}\underset{n=1}{\overset{N}{\&Sigma;}}count({\mathrm{\&beta;}}_n\&GreaterEqual;{\mathrm{\&beta;}}_{th}*{\mathrm{\&beta;}}_{max})$

wherein count (·) represents that the section waviness of the statistical spectral correlation function α is more than β_th*β_maxNumber of peaks of β_nUndulation of the crest of the cross-section, β_maxMaximum waviness of the peaks of the cross-section, β_thTaking a fixed value of 0.1 as a fluctuation threshold value; the threshold of this feature is taken as r₃₁＝60.46；

④ f normalized area of cross section R₄Is defined as:

$R_4=\underset{N\&RightArrow;\mathrm{\&infin;}}{\lim }\frac{1}{N}\underset{n=1}{\overset{N}{\&Sigma;}}\left\{\frac{1}{max\left(S_{xn}^0\right(f\left)\right)}{\&Integral;}_{-f_0}^{f_0}\right|S_{xn}^0\left(f\right)\left|df\right\}$

whereinMeans that the maximum of the cross-section function of the spectral correlation function f is found,the area of the f section is obtained; the threshold of this feature is taken as r₄₁＝903.51，r₄₂＝498.54；

⑤ α ratio of significance of cross-sectional spectral lines R₅Is defined as:

$R_5=\underset{N\&RightArrow;\mathrm{\&infin;}}{\lim }\frac{1}{N}\underset{n=1}{\overset{N}{\&Sigma;}}\frac{\sec \left({\mathrm{\&rho;}}_n\right)}{max\left({\mathrm{\&rho;}}_n\right)}$

where max (p)_n) The maximum significance value of the spectral line in section of α, sec (ρ)_n) The second largest significance value of α section spectral line, and the threshold value of the characteristic is taken as r₅₁＝0.073，r₅₂＝7.95×10^-4

⑥ standard deviation R of zero-center normalized instantaneous amplitude absolute value₆Is defined as:

$\mathrm{\&delta;}=\sqrt{\frac{1}{N}\&lsqb;\underset{i=1}{\overset{N}{\&Sigma;}}{A}_{cn}^2\left(i\right)\&rsqb;-{\&lsqb;\frac{1}{N}\underset{i=1}{\overset{N}{\&Sigma;}}\left|A_{cn}\left(i\right)\right|\&rsqb;}^2}$

wherein N is the number of observation samples, A_cn(i) Normalized instantaneous amplitude for zero center, defined as A_cn(i)＝A_n(i) -1; whereinDivisorAnd A (i) is the sample point instantaneous amplitude; the threshold of this feature is taken as r₆₁＝0.29。

And the identification is carried out on several digital modulation signals by combining the above 6 modulation identification characteristics, the specific modulation identification flow is shown in figure 2,

if it isJudging the modulation mode of the signal to be MSK;

if it is $\{{\hat{R}}_1>r_{12},{\hat{R}}_2>r_{21}\}$ Or $\{{\hat{R}}_1<r_{12},{\hat{R}}_3<r_{31}\},$ Judging that the modulation mode of the signal is 4 FSK;

if it isJudging that the modulation mode of the signal is 2 FSK;

if it isJudging the modulation mode of the signal to be BPSK;

if it is $\{{\hat{R}}_1>r_{11},{\hat{R}}_2<r_{22},{\hat{R}}_5<r_{51},{\hat{R}}_6<r_{61}\},$ Judging that the modulation mode of the signal is 2 ASK;

if it is $\{{\hat{R}}_1>r_{11},{\hat{R}}_2<r_{22},{\hat{R}}_5<r_{51},{\hat{R}}_6>r_{61}\},$ Judging that the modulation mode of the signal is 4 ASK;

if it isJudging the modulation mode of the signal to be 2 FSK;

if it isJudging the modulation mode of the signal to be 4 FSK;

if it is $\{{\hat{R}}_1<r_{12},{\hat{R}}_3>r_{31},{\hat{R}}_4<r_{42},{\hat{R}}_5>r_{52}\},$ Determining the modulation mode QPSK of the signal;

if it is $\{{\hat{R}}_1<r_{12},{\hat{R}}_3>r_{31},{\hat{R}}_4<r_{42},{\hat{R}}_5<r_{52}\},$ Judging the modulation mode 8PSK of the signal;

whereinAn estimate of the characteristic R during modulation identification.

According to the identification criterion, the identification rate of each digital modulation mode is more than 88.8% when the signal-to-noise ratio is 10dB, the identification effect is ideal, and the identified modulation modes comprise 2ASK, 4ASK, BPSK, QPSK, 8PSK, MSK, 2FSK, 4FSK, 2FSK and 4FSK, wherein FSK represents the frequency shift keying in which the initial phase of the code element is irrelevant.

Those skilled in the art will appreciate that the invention may be practiced without these specific details.

Although the best mode of the invention and the accompanying drawings are disclosed for illustrative purposes, those skilled in the art will appreciate that: various substitutions, changes and modifications are possible without departing from the spirit and scope of the present invention and the appended claims. Therefore, the technical solution protected by the present invention should not be limited to the disclosure of the best example and the accompanying drawings.

Claims (5)

1. A common digital modulation signal identification method based on cyclic spectrum correlation is characterized in that: the method comprises the following steps:

(1) dividing the intercepted signal-to-noise ratio modulation signal into n equal parts after sampling, and taking the n equal parts as input signals for performing spectrum correlation operation;

(2) selecting proper Fourier transform point number and smooth window width, and respectively carrying out spectrum correlation operation on the n signals by using a spectrum correlation estimation method based on frequency domain smoothing;

the spectral correlation estimation algorithm based on frequency domain smoothing is as follows:

$S_{x\mathrm{\&Delta;}t}^{\mathrm{\&alpha;}}{(t,f)}_{\mathrm{\&Delta;}f}=\frac{1}{M}\underset{v=-(M-1)/2}{\overset{(M-1)/2}{\&Sigma;}}\frac{1}{\mathrm{\&Delta;}t}{X}_{\mathrm{\&Delta;}t}(t,f+\frac{\mathrm{\&alpha;}}{2}+{\mathrm{vF}}_S)X_{\mathrm{\&Delta;}t}^{*}(t,f-\frac{\mathrm{\&alpha;}}{2}+{\mathrm{vF}}_S)$

$X_{\mathrm{\&Delta;}t}(t,f)=\underset{K=0}{\overset{N-1}{\&Sigma;}}{a}_{\mathrm{\&Delta;}t}\left({\mathrm{KT}}_S\right)x_k(t-{\mathrm{KT}}_S)\exp (-j2\mathrm{\&pi;}f(t-{\mathrm{KT}}_S\left)\right)$

wherein,is the result after the spectrum correlation operation; x_Δt(t, f) is the signal x_k(t) the result of the short-time fourier transform; signal x_k(t) is one of n equal parts of the intercepted signal; Δ t is x_k(t) a duration of time; a is_ΔtIs a window function; Δ f is the frequency domain smoothing interval; f_SIs the frequency domain minimum increment unit; t is_SIs the time domain sampling interval; n is the signal x_k(t) number of samples, M is the smoothing window width factor, α is the cycle frequency, f is the frequency, t is the time;

(3) the operation results in the step (2) are subjected to alpha section wavelet denoising and then are added to obtain an average value, and an output result of the improved spectral correlation estimation calculation method is obtained;

the improved spectral correlation estimation algorithm is as follows:

$S_{x\mathrm{\&Delta;}t}^{\mathrm{\&alpha;}}\left(f\right)=\frac{1}{N}\underset{k=1}{\overset{N}{\&Sigma;}}{\mathrm{WDN}}_{\mathrm{\&alpha;}}\&lsqb;S_{x\mathrm{\&Delta;}t}^{\mathrm{\&alpha;}}(t_k,f)\&rsqb;$

wherein WDN_α(. represents a pairThe cyclic frequency α is subjected to wavelet denoising;

(4) extracting 5 modulation identification characteristics according to the α cross section and the f cross section of the spectrum correlation function obtained in the step (3), and identifying several digital modulation signals by combining a signal characteristic construction and classification method based on time domain statistics, wherein the 5 modulation identification characteristics are the maximum absolute value ratio (marked as R) of the α cross section and the f cross section of the spectrum correlation function₁) α number of strong spectral lines in section (denoted as R)₂) α section undulation coefficient (noted as R)₃) F normalized area of cross section (denoted as R)₄) α ratio of the significance of the spectral lines in the cross-section (denoted as R)₅) (ii) a The statistic device based on signal time domainCharacterized in that: standard deviation of zero-centered normalized instantaneous amplitude absolute value (denoted as R)₆) (ii) a The digital modulation signals comprise 2ASK, 4ASK, BPSK, QPSK, 8PSK, MSK, 2FSK, 4FSK, 2FSK and 4FSK, wherein FSK represents frequency shift keying with uncorrelated initial phases of code elements;

the identification process for several digital modulation signals is as follows:

if it isJudging the modulation mode of the signal to be MSK;

if it is $\left\{\begin{array}{cc}{\hat{R}}_1>r_{12},& {\hat{R}}_2>r_{21}\end{array}\right\}$ Or $\left\{\begin{array}{cc}{\hat{R}}_1<r_{12},& {\hat{R}}_3<r_{31}\end{array}\right\},$ Determining that the modulator of the signal is 4 FSK;

if it isJudging that the modulation mode of the signal is 2 FSK;

if it isJudging the modulation mode of the signal to be BPSK;

if it is $\left\{\begin{array}{cccc}{\hat{R}}_1>r_{11},& {\hat{R}}_2<r_{22},& {\hat{R}}_5<r_{51},& {\hat{R}}_6<r_{61}\end{array}\right\},$ Judging that the modulation mode of the signal is 2 ASK;

if it is $\left\{\begin{array}{cccc}{\hat{R}}_1>r_{11},& {\hat{R}}_2<r_{22},& {\hat{R}}_5<r_{51},& {\hat{R}}_6>r_{61}\end{array}\right\},$ Judging that the modulation mode of the signal is 4 ASK;

if it isDetermining the modulation mode 2FSK of the signal;

if it isJudging the modulation mode of the signal to be 4 FSK;

if it is $\left\{\begin{array}{cccc}{\hat{R}}_1<r_{12},& {\hat{R}}_3>r_{31},& {\hat{R}}_4<r_{42},& {\hat{R}}_5>r_{52}\end{array}\right\},$ Determining the modulation mode QPSK of the signal;

if it is $\left\{\begin{array}{cccc}{\hat{R}}_1<r_{12},& {\hat{R}}_3>r_{31},& {\hat{R}}_4<r_{42},& {\hat{R}}_5<r_{52}\end{array}\right\},$ The modulation scheme of the signal is determined to be 8 PSK.

2. The method for identifying common digital modulation signals based on cyclic spectrum correlation according to claim 1, characterized in that: in the step (2), the number of sampling points of each signal is N-1024, and the smoothing window width coefficient is M-63.

3. The method for identifying common digital modulation signals based on cyclic spectrum correlation according to claim 1, characterized in that: the principle of denoising the small waves in the step (3) is as follows: the noisy sequence is decomposed using the sym8 wavelet, and at the fifth layer of the decomposition, the sequence is denoised using a soft sure threshold selection method, and the threshold is adjusted with the noise variance of the first layer wavelet decomposition.

4. The method for identifying common digital modulation signals based on cyclic spectrum correlation according to claim 1, characterized in that: the 5 modulation identification characteristics based on the modulation signal spectrum correlation function in the step (4) are defined as follows:

① spectral correlation function α section and f section maximum absolute value ratio R₁Is defined as:

$R_1=\underset{N\&RightArrow;\mathrm{\&infin;}}{\lim }\frac{1}{N}\underset{n=1}{\overset{N}{\&Sigma;}}\frac{\max \left(\right|S_{xn}^{\mathrm{\&alpha;}}\left(0\right)\left|\right)}{\max \left(\right|S_{xn}^0\left(f\right)\left|\right)}$

whereinAndare respectively a spectral correlation functionα cross section and f cross section functions of;

② α number of strong lines in cross section R₂Is defined as:

$R_2=\underset{N\&RightArrow;\mathrm{\&infin;}}{lim}\frac{1}{N}\underset{n=1}{\overset{N}{\&Sigma;}}count({\mathrm{\&rho;}}_n\&GreaterEqual;{\mathrm{\&rho;}}_{th}*{\mathrm{\&rho;}}_{max})$

where count (-) represents the number of significant peaks, ρ, for cross-section of the spectral correlation map α_nIs the degree of significance, rho, of the cross-sectional peak_maxIs the maximum significance, rho, of the cross-sectional peak_thTaking a fixed value of 0.27 as a significance threshold value;

wherein the peak saliency ρ is defined as:

$\mathrm{\&rho;}=\frac{h^2}{l*max\left(h\right)}$

wherein h is the difference between the amplitude of the wave peak in the alpha section of the spectrum correlation diagram and the larger of the amplitudes of two adjacent wave troughs, l is the width value of the wave peak, and max (h) is the maximum h value in the alpha section;

③ α coefficient of section fluctuation R₃Is defined as:

$R_3=\underset{N\&RightArrow;\mathrm{\&infin;}}{lim}\frac{1}{N}\underset{n=1}{\overset{N}{\&Sigma;}}count({\mathrm{\&beta;}}_n\&GreaterEqual;{\mathrm{\&beta;}}_{th}*{\mathrm{\&beta;}}_{max})$

wherein count (·) represents that the section waviness of the statistical spectral correlation function α is more than β_th*β_maxNumber of peaks of β_nUndulation of the crest of the cross-section, β_maxMaximum waviness of the peaks of the cross-section, β_thTaking a fixed value of 0.1 as a fluctuation threshold value;

wherein the peak undulation β is defined as:

$\mathrm{\&beta;}=\frac{h}{l}$

wherein h and l have the same meanings as defined in (ii);

④ f normalized area of cross section R₄Is defined as:

$R_4=\underset{N\&RightArrow;\mathrm{\&infin;}}{lim}\frac{1}{N}\underset{n=1}{\overset{N}{\&Sigma;}}\left\{\frac{1}{max\left(S_{xn}^0\right(f))}{\&Integral;}_{-f_0}^{f_0}\right|S_{xn}^0\left(f\right)\left|df\right\}$

whereinMeans that the maximum of the cross-section function of the spectral correlation function f is found,the area of the f section is obtained;

⑤ α ratio of significance of cross-sectional spectral lines R₅Is defined as:

$R_5=\underset{N\&RightArrow;\mathrm{\&infin;}}{lim}\frac{1}{N}\underset{n=1}{\overset{N}{\&Sigma;}}\frac{\sec \left({\mathrm{\&rho;}}_n\right)}{max\left({\mathrm{\&rho;}}_n\right)}$

where max (p)_n) The maximum significance value of the spectral line in section of α, sec (ρ)_n) The next largest significance value for the α cross-sectional spectral line;

the standard deviation R of the absolute value of the zero-center normalized instantaneous amplitude of the statistical characteristics based on the signal time domain in the step (4)₆Is defined as:

$\mathrm{\&delta;}=\sqrt{\frac{1}{N}\&lsqb;\underset{i=1}{\overset{N}{\&Sigma;}}{A}_{cn}^2\left(i\right)\&rsqb;-{\&lsqb;\frac{1}{N}\underset{i=1}{\overset{N}{\&Sigma;}}\left|A_{cn}\left(i\right)\right|\&rsqb;}^2}$

wherein N is the number of observation samples, A_cn(i) Normalized instantaneous amplitude for zero center, defined as A_cn(i)＝A_n(i) -1; whereinDivisorAnd A (i) is the sample point instantaneous amplitude.

5. The method for identifying common digital modulation signals based on cyclic spectrum correlation according to claim 1, characterized in that: the threshold value of each debugging identification characteristic in the step (4) is as follows:

characteristic R₁The threshold value of (2): r is₁₁＝0.67，r₁₂0.36; characteristic R₂The threshold value of (2): r is₂₁＝4.8，r₂₂3.2; characteristic R₃The threshold value of (2): r is₃₁60.46; (ii) a Characteristic R₄The threshold value of (2): r is₄₁＝903.51，r₄₂498.54; characteristic R₅The threshold value of (2): r is₅₁＝0.073，r₅₂＝7.95×10^-4(ii) a Characteristic R₆The threshold value of (2): r is₆₁＝0.29。