A novel grading noise-pretreatment algorithm based on time-frequency blind source separation

The blind source separation (BSS) problem under noise is known as a hard problem. The performance of separation algorithm degrades with the decrease of SNR significantly. The key solution is the noise pretreatment. Wavelet transform (WT) and empirical mode decomposition (EMD), two typical analysis methods especially for the processing practical nonstationarity signals in time frequency domain, are chosen as the pretreatment methods in this paper. Based on the analysis of the denoising performances by the two methods, a grading noise pretreatment project is proposed which automatically selects a method according to different SNR. Simulation results shows that such flexible scheme could enhance the BSS performance by effectively denoising, and also makes the existing blind source separation apply to larger range of SNR and enhances the robustness of algorithm.