Denoising Method of Magnetotelluric Signals based on Multiple-threshold Sparse Decomposition

Abstract

The magnetotelluric (MT) sounding method is a passive-source electromagnetic detection technology, which features the wide frequency band and weak observation signals, more susceptible to noise and interference. To solve the problem of weak broadband electromagnetic exploration data processing, and extract effectively information with a strong interference, MT observational data preprocessing is made with sparse signal decomposition. The key problem is to construct a reasonable criterion that can effectively describe the noise signal. However, the fixed iteration number is taken as a criterion in the traditional single-threshold criterion, causing poor execution efficiency, and the obtained matching atoms are not the best in many cases. Therefore, we propose a multiple-threshold criterion based on the iteration number, residual projection and the residual projection gradient. Experimental results show that the multiple-threshold criterion can separate the MT signal from noise quickly and effectively. In addition, the reconstructed signal has significant improvements in terms of the normalized cross correlation (NCC), signal-to-noise ratio (SNR), error (E) and the running time of the algorithm (T).