CN104485979A - Blind estimation method for underdetermined hybrid frequency hopping parameters based on time frequency diagram correction - Google Patents

Abstract

The invention discloses a blind estimation method for underdetermined hybrid frequency hopping parameters based on time frequency diagram correction in order to mainly solve the problem of large estimation errors caused by a poor signal to noise ratio in the prior art. The method comprises the following implementation steps: (1) performing short-time Fourier transformation on a signal to obtain a time frequency matrix of an observed signal; (2) calculating an energy matrix of the time frequency matrix; (3) correcting the time frequency diagram; (4) estimating frequency hopping signal parameters according to the corrected time frequency diagram. The method disclosed by the invention can be used for obtaining a clear time frequency diagram at a low signal to noise ratio, has the advantage of improving the estimation precision of the frequency hopping signal parameters, and can be applied to various frequency hopping communication scenes.

Description

Based on the under-determined mixture frequency parameter blind estimating method of time-frequency figure correction

Technical field

The invention belongs to signal transacting field, particularly a kind of under-determined mixture frequency parameter blind estimating method, can be used for the field such as frequency hopping communications scouting, non-cooperating frequency hopping communications, military communication.

Background technology

In recent years Frequency-hopping Communication Technology because antijamming capability is strong, good confidentiality and be easy to the advantages such as networking and be widely used in various communication scenes.Research Frequency-hopping Communication Technology, one of its main task estimates the characteristic parameter of Frequency Hopping Signal.

At present, method based on time frequency analysis is the method that Frequency Hopping Signal parameter blind estimation is conventional, first these class methods adopt certain time frequency analyzing tool such as Fourier in short-term to change WVD, Smoothing Pseudo wigner-ville distribution SPWVD, spectrogram etc., obtain the time-frequency distributions of Frequency Hopping Signal, then utilize each moment of time-frequency distributions to estimate hop cycle along the periodicity of the maximum waveform of frequency axis, and then estimate other parameter.But by the impact of the factors such as time-frequency uncertainty, noise and cross-interference terms, the time-frequency figure obtained is often fuzzyyer, directly carries out data analysis from time-frequency figure, and error is large, and thus this class methods signal to noise ratio adaptive capacity is poor.In addition, the method based on atomic parameter can also be adopted to estimate Frequency Hopping Signal parameter, namely construct a kind of time-frequency atom dictionary mated with Frequency Hopping Signal, then Frequency Hopping Signal is decomposed into the combination of time-frequency atom, finally estimate the parameter of Frequency Hopping Signal according to the parameter value of these time-frequency atom.This class methods time-frequency atom dictionary is huge, and algorithm complex is high.Current most of algorithm is only applicable to the parameter Estimation of single Frequency Hopping Signal, multiple Frequency Hopping Signal can be separated by eigenmatrix associating near optimal algorithm, and then adopt Time-Frequency Analysis Method can estimate multiple Frequency Hopping Signal parameter, but the method is only applicable to the situation of overdetermination or positive definite, and when signal to noise ratio is low poor performance.For owing the parameter Estimation of the Frequency Hopping Signal under shape, although can apply based on the openness method revising time-frequency figure of time-frequency, choosing of the method key parameter is not adaptive, and the impact by signal to noise ratio is very large, and thus parameter estimating error is large.

Summary of the invention

The object of the invention is to the deficiency for above-mentioned prior art, propose a kind of under-determined mixture frequency parameter blind estimate algorithm based on time-frequency figure correction, the adaptive ability to different noise when obtaining characteristic parameter to improve, reduces parameter estimating error.

Realizing technical thought of the present invention is: by the segmentation to time-frequency domain signal, calculates each section of time-frequency energy in short-term.Choose an initial threshold by the short-time energy of each section, observation data is divided into signal segment and noise segment according to this thresholding.The short-time energy of basis signal section, chooses new thresholding, and the noise signal being by mistake divided into signal segment is repartitioned noise segment, obtains time-frequency figure clearly, carries out the parameter Estimation of Frequency Hopping Signal.

According to above-mentioned thinking, performing step of the present invention is as follows:

(1) Fourier is in short-term done to time domain observation signal x (t) received and change STFT, obtain time-frequency matrix

(2) by time-frequency matrix each element delivery value and square after, obtain time-frequency energy matrix B;

(3) press leu according to time-frequency energy matrix B and revise time-frequency matrix obtain revised time-frequency matrix

(3a) by the q column data b of time-frequency energy matrix B _qsegmentation also calculates the short-time energy of every section, obtains short-time energy vector Y, calculates short-time energy mean value again the interval division be made up of with maximum the minimum value of short-time energy vector Y is become several minizones, the length of each minizone is the mean value of short-time energy add up data amount check in the short-time energy vector Y that each minizone comprises, find out the minizone i that front and back data amount check rate of change is maximum _γ, and according to this i _γand choose initial threshold: wherein, * represents and is multiplied;

(3b) according to initial threshold γ ₀divide between noise range and signal spacing, by each component of short-time energy vector Y and initial threshold γ ₀compare, if component is less than initial threshold γ ₀the segmentation that then this component is corresponding belongs between noise range, otherwise belongs to signal spacing; The number-of-fragments that recording noise is interval and signal spacing is corresponding respectively, and check in signal spacing whether have continuous print fragment number, if nothing, then obtain frequency index matrix by the number-of-fragments of signal spacing with signal spacing short-time energy vector otherwise fragment number continuous print adjacent segment, obtains frequency index matrix in combined signal interval with signal spacing short-time energy vector $\tilde{Y};$

(3c) secondary location is carried out to signal spacing, calculate signal spacing short-time energy vector mean value and compare, if signal spacing short-time energy is vectorial in component be less than then this component is joined the noise vector y in signal spacing _win, otherwise this component is joined signal vector y _sin; According to signal vector y _sand mean value choose new thresholding: wherein mean (y _s) represent and ask y _sthe mean value of middle data;

Calculate wherein k _mfor the fragment number between noise range, N ₀for the number of noise Concourse Division; Inspection signal spacing noise vector y _win data y _wjif, in signal spacing, then retain this component y _wj, otherwise, fragment number corresponding for this component is rejoined between noise range, and from frequency index matrix by component y _wjcorresponding frequency index row vector is removed;

(3d) time-frequency matrix is revised: take out frequency index matrix every a line, find out often go minimum value V _minwith maximum V _max, be V by the data extending of often going _minto V _maxbetween all continuous integral numbers, be saved in successively in set omega, revise according to Ω in q row column vector obtain revised q column vector

(3e) original matrix is revised successively according to step (3a)-(3d) in all row, obtain revised time-frequency matrix $\hat{X}=\left[\begin{array}{cccc}{\hat{x}}_1& {\hat{x}}_2& \·\·\·& {\hat{x}}_Q\end{array}\right];$

(4) according to revised time-frequency matrix estimate the source signal number of Frequency Hopping Signal, saltus step window, often jump signal carrier frequency vector, jumping moment and hop period these parameters.

Compared with prior art, the invention has the beneficial effects as follows:

1) the present invention revises time-frequency figure by the energy detection method of adaptive threshold, improves noise resisting ability, makes time-frequency figure more clear, and is applicable to owe shape;

2) the present invention estimates frequency parameter according to revised time-frequency matrix, and the estimated value obtained is more accurate.

Accompanying drawing explanation

Fig. 1 of the present inventionly realizes general flow chart;

During Fig. 2, the present invention carries out the sub-process figure of time-frequency figure correction;

Fig. 3 the present invention is based on energy binned to choose the sub-process figure of initial threshold;

Fig. 4 is the sub-process figure that the present invention divides with signal spacing between noise range;

Fig. 5 is the sub-process figure that the present invention locates signal spacing secondary;

Fig. 6 is the sub-process figure that the present invention estimates saltus step parameter;

Fig. 7 is uncorrected time-frequency figure;

Fig. 8 utilizes the inventive method to carry out revised time-frequency figure;

Fig. 9 is the comparison diagram that existing Time-Frequency Analysis Method and the inventive method change with signal to noise ratio signal carrier frequency evaluated error;

Figure 10 is the comparison diagram that existing Time-Frequency Analysis Method and the inventive method change with signal to noise ratio jumping moment evaluated error;

Figure 11 is the comparison diagram that existing Time-Frequency Analysis Method and the inventive method change with signal to noise ratio hop period evaluated error.

Embodiment

Below with reference to accompanying drawing and example in detail embodiments of the present invention, accompanying drawing described herein is used to provide a further understanding of the present invention, form a application's part, schematic description and description of the present invention, for explaining the present invention, does not form inappropriate limitation of the present invention.

With reference to Fig. 1, specific embodiment of the invention step is as follows:

Step 1 obtains its time-frequency matrix to observation signal x (t) received as Short Time Fourier Transform as shown in the formula:

$\hat{X}(p,q)=\underset{t=p\·L_{\mathrm{step}}+1}{\overset{t=p\·L_{\mathrm{step}}+N_{\mathrm{fft}}}{\mathrm{\Σ}}}x\left(t\right)e^{-j\frac{2\mathrm{\πtq}}{N_{\mathrm{fft}}}},p=\mathrm{1,2},\·\·\·,N_{\mathrm{fft}},q=\mathrm{1,2},\·\·\·,Q$

Wherein N _fftfor length of window, L _stepfor sliding window interval, Q is window number.

Step 2. is by time-frequency matrix each element delivery value after square, obtain time-frequency energy matrix B.

Step 3. is pressed leu according to time-frequency energy matrix B and is revised time-frequency matrix

With reference to Fig. 2, being implemented as follows of this step:

(3a) initial threshold is chosen:

With reference to Fig. 3, choose being implemented as follows of initial threshold:

(3a1) for the q row b of time-frequency energy matrix B _q, by b _qinterior data are divided into C section, and every segment length is L; To the data summation in segmentation, obtain the short-time energy of this section; Short-time energy is asked to every segment data, obtain short-time energy vector Y=[Y (1) ... Y (k) ..., Y (C)]; Wherein, to be integer 8, C be L y (k) is the short-time energy of a kth segmentation, k=1,2 ..., C;

(3a2) short-time energy average is calculated

$\stackrel{\‾}{Y}=\frac{1}{C}\underset{c=1}{\overset{C}{\mathrm{\Σ}}}Y\left(c\right),$

(3a3) interval division be made up of with maximum the minimum value of short-time energy vector Y is become several minizones, the length of each minizone is short-time energy average

(3a4) add up the number that each minizone comprises data in short-time energy vector Y, find out front and back data amount check and change maximum minizone i _γ;

(3a5) according to i _γand calculate initial threshold: wherein, * represents and is multiplied;

(3b) between noise range and signal spacing is divided

With reference to Fig. 4, divide between noise range as follows with the concrete steps of signal spacing:

(3b1) by initial threshold γ ₀compare with the data in short-time energy vector Y, if the data in short-time energy vector Y are less than γ ₀, then the number-of-fragments of its correspondence is joined between noise range, otherwise joins signal spacing;

(3b2) check in signal spacing whether have continuous print number-of-fragments; If nothing, then the frequency index of number-of-fragments corresponding segments in signal spacing is joined frequency index matrix by row in, the short-time energy of segmentation is joined signal spacing short-time energy vector in; Otherwise first merging numbering continuous print is multiple is segmented into a new segmentation, then obtains frequency index matrix as stated above with the short-time energy vector of signal spacing concrete merging method is as follows:

First from column vector b _qin find and will merge data corresponding to segmentation, then by these data by sorting from big to small, choose the data of a front L data as new segmentation, using the short-time energy of this L data sum as new segmentation, record frequency index corresponding to this L data as frequency index corresponding to new segmentation simultaneously; Wherein L is the length of in (3a1) every section;

(3c) secondary location is carried out to signal spacing

With reference to Fig. 5, being implemented as follows of this step:

(3c1) to signal spacing short-time energy vector in data average, obtain the short-time energy mean value in signal spacing

(3c2) by short-time energy vector middle data with compare, if in data be less than then joined the noise vector y in signal spacing _win, otherwise, join signal vector y _sin; Calculate new thresholding γ ₁:

${\mathrm{\γ}}_1=\frac{1}{2}(\mathrm{man}\left(y_s\right)+\stackrel{\stackrel{\‾}{~}}{Y}),$

Wherein mean (y _s) represent and ask y _sin data average;

(3c3) the interval short-time energy average of calculating noise:

${\stackrel{\‾}{Y}}_{\mathrm{noise}}=\frac{1}{N_0}\underset{m=1}{\overset{N_0}{\mathrm{\Σ}}}Y\left(k_m\right)$

Wherein, k _mfor the number-of-fragments between noise range, N ₀for the number of number-of-fragments middle between noise range;

(3c4) according to new thresholding γ ₁and short-time energy average between noise range inspection noise vector y _win data y _wjif, in signal spacing, then retain this component y _wjcorresponding number-of-fragments, otherwise, this number-of-fragments is rejoined between noise range, and from frequency index matrix middle by component y _wjcorresponding frequency index row vector is removed;

(3d) time-frequency matrix is revised: take out subscript matrix successively every a line, find out often go minimum value V _minwith maximum V _max, be V by the data extending of often going _minwith V _maxall integers all are also saved in set omega, finally obtain revised column vector for:

${\hat{x}}_q\left(i\right)=\left\{\begin{array}{cc}0& i\∉\mathrm{\Ω}\\ {\tilde{x}}_q\left(i\right)& i\∈\mathrm{\Ω}\end{array}\right.,$

Wherein, for time-frequency matrix q column vector;

(3e) time-frequency matrix is revised successively according to step (3a)-(3d) in all row, obtain revised time-frequency matrix $\hat{X}=\left[\begin{array}{cccc}{\hat{x}}_1& {\hat{x}}_2& \·\·\·& {\hat{x}}_Q\end{array}\right];$

Step 4, according to revised time-frequency figure, estimates each parameter of Frequency Hopping Signal.

With reference to Fig. 6, being implemented as follows of this step:

(4a) source signal number is estimated: to revised time-frequency matrix the column vector that every window is corresponding carries out frequency cluster, the number of the vector sum cluster frequency that the cluster frequency obtaining this window is formed, note be the cluster frequency vector of q window, num _qit is the cluster frequency number of q window;

The number of times that Statistical Clustering Analysis frequency number occurs, using the estimated value of cluster frequency numbers maximum for occurrence number as source signal number

(4b) saltus step window is estimated:

(4b1) find out all cluster frequency numbers and be greater than source signal number estimated value window;

(4b2) cluster frequency number is greater than to the window q of source signal number estimated value, judges the cluster frequency change of window before and after it, search positive integer n, n is met if the cluster frequency vector of q-n window with the cluster frequency vector of q+n window all satisfied then be judged as it not being saltus step window, otherwise, be saltus step window; Obtain all saltus step windows as stated above;

(4b3) to the multiple saltus step window of window number continuous print, retain first saltus step window, abandon remaining saltus step window; To the discontinuous saltus step window of window number, be then directly left saltus step window;

(4c) estimated signal carrier frequency:

(4c1) according to saltus step window, source signal is divided by jumping figure, to the window between kth-1 saltus step window and a kth saltus step window, check the cluster frequency number of its correspondence, if the cluster frequency number of window equals source signal number estimated value signal then in this window belongs to kth and jumps signal, otherwise does not belong to kth jumping signal;

(4c2) to often jump signal as Fourier transform obtain this jumping signal carrier frequency vector;

(4d) a kth jumping moment is estimated according to carrier frequency vector by the method for circulation adjustment detection window

(4d1) using a kth saltus step window as the first detection window T ₁home window, if detection window T ₁starting point be s, terminal is d, to the first detection window T ₁the first spectrogram F is obtained as Fourier transform ₁;

(4d2) kth is jumped the carrier frequency of signal at the first spectrogram F ₁the maximum ε of middle amplitude _kwith the first spectrogram F ₁in to jump except kth and except kth+1 jumping signal carrier frequency, the maximum λ of amplitude compares; If ε _k> λ, then by the first detection window T ₁starting point s with step-length step ₁move backward, and upgrade the first spectrogram F ₁and described two maximum ε _kand λ, circulation performs this step until ε _kstop during < λ or s>d, record move frequency n ₁with the first detection window T ₁starting point s; Otherwise, directly perform (4d3), wherein step-length step ₁during for carrying out Short Time Fourier Transform window size 1/20th;

(4d3) with the first detection window T ₁starting point s centered by, with the step-length step of 2 times ₁for length configuration second detection window T ₂, to the second detection window T ₂the second spectrogram F is obtained as Fourier transform ₂;

(4d4) by the second spectrogram F ₂middle kth jumps the average λ of the corresponding amplitude of signal carrier frequency ₁and kth+1 jumps the average λ of the corresponding amplitude of signal carrier frequency ₂compare, if λ ₁> λ ₂, then by the second detection window T ₂center with step-length step ₂move backward, and upgrade the second spectrogram F ₂and described two average λ ₁and λ ₂, circulation performs this step until λ ₁< λ ₂stop, record move frequency n ₂and perform step (4d6); Otherwise, perform (4d5), wherein step-length step ₂for step-length step ₁1/10th;

(4d5) by the second detection window T ₂center with step-length step ₂move forward, and upgrade the second spectrogram F ₂and described two average λ ₁and λ ₂, circulation performs this step until λ ₁> λ ₂stop, record move frequency n ₂and perform (4d6);

(4d6) jumping moment is calculated

${\hat{t}}_h^{\left(k\right)}=(h_k-1)\&times;L_{\mathrm{step}}+{\mathrm{step}}_1\&times;n_1\&PlusMinus;{\mathrm{step}}_2\&times;(n_2-1)\stackrel{\&OverBar;}{+}{\mathrm{step}}_2,$

Wherein, symbol " ± " and symbol express possibility and get "+" and also may get "-": when adopting the circulation time in (4d4), symbol " ± " is got "+", symbol get "-"; When adopting the circulation time in (4d5), symbol " ± " is got "-", symbol get "+"; h _kfor the window number of a kth saltus step window, L _stepfor carrying out sliding window interval during Short Time Fourier Transform;

(4e) according to jumping moment, hop period is calculated

${\hat{T}}_h=\frac{1}{\hat{K}-1}\underset{k=1}{\overset{\hat{K}-1}{\mathrm{\&Sigma;}}}({\hat{t}}_h^{(k+1)}-{\hat{t}}_h^{\left(k\right)}),$

Wherein, for the number of jumping moment.

The estimated value of source signal number is obtained by above-mentioned steps (4a) estimate saltus step window by above-mentioned steps (4b), obtained the estimated value of signal carrier frequency by above-mentioned steps (4c), obtained the estimated value of jumping moment by above-mentioned steps (4d) the estimated value of hop period is obtained by above-mentioned steps (4e) so far the estimated value of all frequency parameters is obtained.

Advantage of the present invention can be further illustrated by following emulation experiment:

(1) simulated conditions

Initialize signal is the Frequency Hopping Signal under the synchronized orthogonal networking mode produced by matlab, and source signal number N is 4, hop period T _hfor 0.001s, transition times K is 2, first jumping moment t ⁽¹⁾for 0.001s, second jumping moment t ⁽²⁾for 0.002s, source signal real Hopping frequencies collection in observation time is $f_c=\left[\begin{array}{ccc}5& 1& 2\\ 6.5& 2.5& 4\\ 7.5& 5.5& 4.5\\ 8& 6& 7\end{array}\right]\mathrm{MHz},$ The experiment number U of emulation experiment 3 to emulation experiment 5 is 5000 times.

Frequency Hopping Signal parameter Estimation performance measure index:

Carrier frequency evaluated error e _f:

$e_f=\frac{1}{\mathrm{NKU}}\underset{n=1}{\overset{N}{\mathrm{\&Sigma;}}}\underset{k=1}{\overset{K}{\mathrm{\&Sigma;}}}\underset{u=1}{\overset{U}{\mathrm{\&Sigma;}}}\frac{|f_{\mathrm{nk}}^{\left(u\right)}-{\hat{f}}_{\mathrm{nk}}^{\left(u\right)}|}{f_{\mathrm{nk}}^{\left(u\right)}}$

Wherein, be the true carrier frequency of the u time experiment n-th source signal kth jumping, the carrier frequency estimated value of the u time experiment n-th source signal kth jumping;

Jumping moment evaluated error e _t:

$e_t=\frac{1}{\mathrm{KU}}\underset{k=1}{\overset{K}{\mathrm{\&Sigma;}}}\underset{u=1}{\overset{U}{\mathrm{\&Sigma;}}}\frac{|t^{\left(k\right)}-{\hat{t}}_{\mathrm{hu}}^{\left(k\right)}|}{{t_h}^{\left(k\right)}}$

Wherein, the estimated value of the u time experiment kth jumping moment;

Hop cycle evaluated error

$e_{T_h}=\frac{1}{U}\underset{u=1}{\overset{U}{\mathrm{\&Sigma;}}}\frac{|T_h-{{\hat{T}}_h}^{\left(u\right)}|}{T_h}$

Wherein, it is the estimated value of the u time experiment hop period;

(2) content is emulated

Emulation experiment 1: as Short Time Fourier Transform, time-frequency figure is obtained to initialize signal, result is as shown in Figure 7;

Emulation experiment 2: utilize the time-frequency figure of method of the present invention to signal to revise, result as shown in Figure 8;

Emulation experiment 3: under different signal to noise ratio condition, estimate signal carrier frequency respectively by existing Time-Frequency Analysis Method and the inventive method and calculate carrier frequency evaluated error, the carrier frequency evaluated error of two kinds of methods contrasted, result is as shown in Figure 9;

Emulation experiment 4: under different signal to noise ratio condition, estimate jumping moment respectively by existing Time-Frequency Analysis Method and the inventive method and calculate jumping moment evaluated error, the jumping moment evaluated error of two kinds of methods contrasted, result as shown in Figure 10

Emulation experiment 5: under different signal to noise ratio condition, estimate hop period respectively by existing Time-Frequency Analysis Method and the inventive method and calculate hop period evaluated error, the hop period evaluated error of two kinds of methods contrasted, result is as shown in figure 11;

(3) analysis of simulation result

In as can be seen from Fig. 7 to Figure 11, method noise resisting ability of the present invention is strong, and the time-frequency figure obtained is more clear; Obtain signal carrier frequency, jumping moment according to revised time-frequency matrix, the estimated value of hop period is more accurate;

The above is only the preferred embodiment of the present invention; it should be pointed out that for those skilled in the art, under the premise without departing from the principles of the invention; can also make some improvements and modifications, these improvements and modifications also should be considered as protection scope of the present invention.

Claims (6)

1., based on a under-determined mixture frequency parameter blind estimating method for time-frequency figure correction, comprise the steps:

(1) Fourier is in short-term done to time domain observation signal x (t) received and change STFT, obtain time-frequency matrix

(2) by time-frequency matrix each element delivery value and square after, obtain time-frequency energy matrix B;

(3) press leu according to time-frequency energy matrix B and revise time-frequency matrix obtain revised time-frequency matrix

(3a) by the q column data b of time-frequency energy matrix B _qsegmentation also calculates the short-time energy of every section, obtains short-time energy vector Y, calculates short-time energy mean value again the interval division be made up of with maximum the minimum value of short-time energy vector Y is become several minizones, the length of each minizone is the mean value of short-time energy add up data amount check in the short-time energy vector Y that each minizone comprises, find out the minizone i that front and back data amount check rate of change is maximum _γ, and according to this i _γand choose initial threshold: wherein, * represents and is multiplied;

(3b) according to initial threshold γ ₀divide between noise range and signal spacing, by each component of short-time energy vector Y and initial threshold γ ₀compare, if component is less than initial threshold γ ₀the segmentation that then this component is corresponding belongs between noise range, otherwise belongs to signal spacing; The number-of-fragments that recording noise is interval and signal spacing is corresponding respectively, and check in signal spacing whether have continuous print fragment number, if nothing, then obtain frequency index matrix by the number-of-fragments of signal spacing with signal spacing short-time energy vector otherwise fragment number continuous print adjacent segment, obtains frequency index matrix in combined signal interval with signal spacing short-time energy vector

(3c) secondary location is carried out to signal spacing, calculate signal spacing short-time energy vector mean value and compare, if signal spacing short-time energy is vectorial in component be less than then this component is joined the noise vector y in signal spacing _win, otherwise this component is joined signal vector y _sin; According to signal vector y _sand mean value choose new thresholding: wherein mean (y _s) represent and ask y _sthe mean value of middle data;

Calculate wherein k _mfor the fragment number between noise range, N ₀for the number of noise Concourse Division; Inspection signal spacing noise vector y _win data y _wjif, in signal spacing, then retain this component y _wj, otherwise, fragment number corresponding for this component is rejoined between noise range, and from frequency index matrix by component y _wjcorresponding frequency index row vector is removed;

(3d) time-frequency matrix is revised: take out frequency index matrix every a line, find out often go minimum value V _minwith maximum V _max, be V by the data extending of often going _minto V _maxbetween all continuous integral numbers, be saved in successively in set omega, revise according to Ω in q row column vector obtain revised q column vector

(3e) original matrix is revised successively according to step (3a)-(3d) in all row, obtain revised time-frequency matrix $\hat{X}=\left[\begin{array}{cccc}{\hat{x}}_1& {\hat{x}}_2& \&CenterDot;\&CenterDot;\&CenterDot;& {\hat{x}}_Q\end{array}\right];$

(4) according to revised time-frequency matrix estimate the source signal number of Frequency Hopping Signal, saltus step window, often jump signal carrier frequency vector, jumping moment and hop period these parameters.

2. method according to claim 1, estimating the source signal number of Frequency Hopping Signal in wherein said step (4), is for revised time-frequency matrix the advanced line frequency cluster of every window, the number of the vector sum cluster frequency that the cluster frequency obtaining this window is formed, note be the cluster frequency vector of q window, num _qit is the number of q window cluster frequency; The number of times that Statistical Clustering Analysis frequency number occurs, using the estimated value of values maximum for cluster frequency number occurrence number as source signal number

3. method according to claim 1, saltus step window is estimated in wherein said step (4), estimate according to the cluster frequency vector sum cluster frequency number of window, namely cluster frequency number is greater than to the window q of source signal number estimated value, judge the cluster frequency change of window before and after it, search positive integer n, n is met if the cluster frequency vector of q-n window with the cluster frequency vector of q+n window all satisfied then be judged as it not being saltus step window, otherwise, be saltus step window.

4. method according to claim 1, in wherein said step (4), the carrier frequency vector of signal is often jumped in estimation, be first according to saltus step window, source signal is divided by jumping figure, then make Fourier transform to often jumping signal, often jumped the carrier frequency vector of signal.

5. method according to claim 1, estimates jumping moment in wherein said step (4) carry out as follows:

(5a) using a kth saltus step window as the first detection window T ₁home window, if detection window T ₁starting point be s, terminal is d, to the first detection window T ₁the first spectrogram F is obtained as Fourier transform ₁;

(5b) kth is jumped the carrier frequency of signal at the first spectrogram F ₁the maximum ε of middle amplitude _kwith the first spectrogram F ₁in to jump except kth and except kth+1 jumping signal carrier frequency, the maximum λ of amplitude compares, if ε _k> λ, then by the first detection window T ₁starting point s with step-length step ₁move backward, and upgrade the first spectrogram F ₁and described two maximum ε _kand λ, circulation performs this step until ε _kstop during < λ or s>d, record move frequency n ₁with the first detection window T ₁starting point s; Otherwise, directly perform (5c); Wherein step-length step ₁during for carrying out Short Time Fourier Transform window size 1/20th;

(5c) with the first detection window T ₁starting point s centered by, with the step-length step of 2 times ₁for length configuration second detection window T ₂, to the second detection window T ₂the second spectrogram F is obtained as Fourier transform ₂;

(5d) by the second spectrogram F ₂middle kth jumps the average λ of the corresponding amplitude of signal carrier frequency ₁and kth+1 jumps the average λ of the corresponding amplitude of signal carrier frequency ₂compare, if λ ₁> λ ₂, then by the second detection window T ₂center with step-length step ₂move backward, and upgrade the second spectrogram F ₂and described two average λ ₁and λ ₂, circulation performs this step until λ ₁< λ ₂stop, record move frequency n ₂and perform step (5f); Otherwise, perform (5e); Wherein step-length step ₂for step-length step ₁1/10th;

(5e) by the second detection window T ₂center with step-length step ₂move forward, and upgrade the second spectrogram F ₂and described two average λ ₁and λ ₂, circulation performs this step until λ ₁> λ ₂stop, record move frequency n ₂and perform (5f);

(5f) jumping moment is calculated

Wherein, symbol " ± " and symbol represent and get "+" or get "-", when adopting the circulation time in step (5d), symbol " ± " is got "+", symbol get "-"; When adopting the circulation time in step (5e), symbol " ± " is got "-", symbol get "+"; h _kfor the window number of a kth saltus step window, L _stepfor carrying out sliding window interval during Short Time Fourier Transform.

6. method according to claim 1, is calculated hop period in wherein said step (4), is calculated by following formula:

${\hat{T}}_h=\frac{1}{\hat{K}-1}\underset{k=1}{\overset{\hat{K}-1}{\mathrm{\&Sigma;}}}({\hat{t}}_h^{(k+1)}-{\hat{t}}_h^{\left(k\right)}),$

Wherein, for the number of jumping moment.