Research on magnetotelluric signal noise separation based on improved SVD

As the impact of cultural noise on magnetotelluric impedance estimation becomes increasingly severe, we propose a magnetotelluric signal-to-noise separation method based on improved Singular Value Decomposition (SVD). Compared to cultural noise, the amplitude of the useful electromagnetic signal is weaker. Based on the difference in amplitude between the noise and the useful signal, the SVD method can separate the noisy components from the original signal. However, traditional SVD struggles to completely extract and remove noise when separating different components, and residual noise often exists at the boundary of noise waveforms. Therefore, by combining the Complementary Ensemble Empirical Mode Decomposition (CEEMD) to improve the SVD method, the goal of effectively separating useful magnetotelluric signals and cultural noise is achieved. The proposed method is applied to synthetic cases and actual data processing. Compared with traditional SVD and CEEMD, the results showed that the proposed method improved the problem of incomplete noise suppression in the SVD method. It achieved good processing results for large-scale noise and impulse noise in low signal-to-noise ratio data, and can reduce the distortion of the apparent resistivity phase curve. Furthermore, compared with the processing results of traditional robust methods, the proposed method exhibits superior noise resistance and restores smoother response curves.