Parameter Estimation Of Lfm Signal In Alpha Stable Noise

Linear frequency modulation signals are an important class of low probability signals,and they have been widely used in sonar,radar,communications and seismic surveys.In addition,as a kind of typical non-stationary signal,the frequency modulation of an linear frequency modulation signal is first-order time-varying,and can be used as the basis for processing high-order polynomial phase signals.Therefore,it is of great significance to study the parameter estimation method of linear frequency modulation signals.Traditional analysis methods for linear frequency modulation signals are based on Gaussian noise.In recent years,studies have shown that alpha stable distribution is more suitable for some real noise or man-made noise.As the noise model changes,conventional algorithms based on second-order statistics are no longer valid.Therefore,it is of great theoretical significance and engineering value to carry out research on detection and parameter estimation of linear frequency modulation signals,especially in alpha stable noise environment.In the environment of alpha stable noise,traditional time-frequency analysis methods may degrade or even fail in the analysis of linear frequency modulation signals,which is not conducive to extract signal parameter.In this thesis,according to the characteristics of alpha stable distribution,a Gaussian-type function which can suppress large pulses is constructed.Based on this,a new method for parameter estimation of linear frequency modulation signals based on Short time Gaussian-Fourier transformation time-frequency distribution is proposed.This method adds a sliding window to the observed signal first,and then uses the constructed Gaussian-type function to process the signal in the short-time window to obtain the Short time Gaussian-Fourier time-frequency distribution.The method proposed in this thesis can effectively suppress the alpha stable noise and obtains the time-frequency distribution of the signal.It also has good robustness to impulse noise with different intensities.Using the feature that the linear frequency modulation signal is a straight line in the time-frequency domain,Hough transform can be further applied to the Gaussian-Fourier transform to obtain the parameter estimation information of the linear frequency modulation signal.Simulation experiments show that the proposed method can effectively estimate the signal parameters in alpha stable noise environment and has good robustness.The fractional order correlation method can estimate the chirp rate of the linear frequency modulation signal,but its estimation performance degrades in alpha stable noise.In this regard,the thesis proposes an improved fractional correlation method based on the correlation entropy.This method utilizes the local correlation properties of the correlation entropy and combines it with the fractional order correlation function to suppress the effect of noise on the signal in the fractional correlation domain.In alpha stable noise,the proposed method firstly used to estimate the chirp rate of the linear frequency modulation signal,then demodulates the observed signals,uses the correntropy and the correntropy spectrum to process the short-time stationary signal after demodulating,thereby the initial frequency can be obtained.Simulation experiments demonstrate the effectiveness of the proposed method.LVD has become a powerful tool for analyzing linear frequency modulation signals in recent years.It can directly transform the signal to the centroid frequency-chirp rate domain.After the LVD transform,the linear frequency modulation signal has a distinct peak in the centroid frequency-chirp rate domain,and the signal parameters can be estimated from the peak coordinates.However,in alpha stable noise,the peak of the linear frequency modulation signal in the centroid frequency-chirp rate domain is submerged by noise.Based on this,the thesis proposes a method,Generalized Cauchy Radial Basis Function-LVD,to estimate the parameters of linear frequency modulation signals.Utilizes the nonlinear properties of radial basis function network and employs generalized Cauchy distribution as the radial basis function,the method can suppress the alpha stable noise through continuous training for network.Then the LVD is used to transform the signal with the impulses being suppressed to the centroid frequency-chirp rate domain and directly estimate the signal parameters.