Magnetotelluric Sounding Signal Higher Order Spectral Estimation And Applied Research

In the magnetotelluric data processing, it is often difficult because theresponse functions (such as apparent resistivity and phase) appear dispersionat individual frequencies, rather big error bar and shape abnormity etc.Although we can use much more sampling and stacking average means infield to improve data quality, but cannot radically solve the problems.Additionally many hypothesis and limits was introduced on signal andgeology model in previous kinds of data processing methods, based onpower spectrum, such as signal is Gauss signal, noise is Gauss white noiseand tetluric system is linear non-minimum phase system. But, actualsituation does not like this. Moreover, power spectrum does not suppressnoise very well and loses phase information of signal when processing databecause of the inherent computational mode of power spectrum.Study showed that higher-order statistics can extract many informationfrom signal that can not by two-order statistic(correlation). It can not onlyautomatically suppress the influence of Gauss color noise, but also suppressthe influence of non-Gauss color noise. It has more excellence than classicalpower spectrum estimation and correction function. It is said that it isworth resorting to higher-order statistics method to analyze and process anyproblems that cannot be solved satisfactorily by using power spectrum orcorrelation function.In this paper, through the use of MATLAB as the numerical platform,the study of power spectrum estimation for magnetotelluric by using higherorder statistic analysis is performed. From reading the time serieas files ofEH-4 high frequentcy magnetotelluric image system, the power spectum andcross power spectrum were reconstructed using higher order statistics, thepower spectrum file is formed, and the result generated by using classicalpower spectrum estimation method such as periodogram is compared. Forthe special property that the higher order statistics (higher order spectrumand higher order cumulants) of Gaussian process is zero, so power spectrumestimation based on higher order cumulants can suppress color gauss noisenaturely. So this method can be used to improve the quality of highfrequentcy MT data processing.