A Study Of Matching Pursuit On The Suppression Of Magnetotelluric Noise

Magnetotelluric, one of the most mature electromagnetic prospecting methods in recent years, is an important means for the study of geological structure, mineral exploration and engineering prospecting. Because of the extended frequency bandwidth, the weak MT signal is vulnerablly interfered by both environmental and artificial noise. It is vital for the MT data acquisition and processing to suppress noise and improve the SNR. However, the nonlinear, non-stationary and non-minimum phase attribute of MT signal does not meet the initial conditions of Fourier Transform.In order to denoise MT sounding signal, processing methods like short-time Fourier Transform, Wavelet Transform, Hilbert-Huang Transform and S Transform were proposed to achieve more accurate time domain and frequency domain localization.This dissertation introduces a novel approach of signal decomposition, Matching Pursuit method, which decomposes the MT signal into linear representations through the establishment of a redundant, overcomplete waveform dictionary and implementation of greedy algorithm. Due to the application of Wigner-Ville spectrum to magnetolluric signal, the information of time-frequency distribution of environmental noise is proposed to hit the target of denoising.According to the shape of MT noise, the noise is classified into impulse noise,harmonic noise, charged and discharged noise, step noise and rectangle waveform noise, as signal model for denoising; introduces the basic theory of matching pursuit,Gauss function with scaling, time shifting, frequency shifting and phase changing constructs Gabor atoms. greedy algorithm is used to decompose and reconstruct the theoretical signals adding with different kinds of noises laying the foundation of separation of signal and noise. the reconstructed signal spectrum analysis and understanding the spectrum feature of different types of noise,providing the critical basis for signal noise separation. A new method of MT noise suppression is provided by the analysis of the spectrum of theoretical signal loaded with different noises.