A Novel Convolutive ICA for Seismic Data Denoising

A main task of geophysical exploration is to remove random noises in seismic data processing to improve the SNR. Recently blind source separation (BSS) theory is applied to remove seismic random noises. But most are based on the instantaneous mixture model and limited to the synthetic seismic records. A novel denoising method called fast fixed-point convolutive ICA was presented in this paper for enhancing noisy seismic records, which was based on the FastICA method developed by Hyvarinen and Oja for instantaneous mixtures. The novel method aims at filling the gap existing in the other methods and applying a fast-converging kurtotic method for convolutive mixtures to seismic data denoising. The validity and feasibility of the proposed method was verified by both the synthetic and real seismic records. Results show that the average SIR of the seismic signals was improved about 7 dB after processed by the novel method.