A new feed forward multilevel and spectrum difference separation BSS denoising method

Blind signal processing in noisy is difficult because there are a great many unknown terms. In this paper, a new BSS denoising method with additive noise is proposed. This method first collected noise sample become noise specimen collections, then circulation subtract noise sample from the observed signal with additive noise in frequency domain, some groups with higher SNR are selected about using a SNR estimation algorithm by the paper given. And the groups are new sources of feed forward multilevel BSS. Feed forward multilevel BSS is that outputs of last separation be regard as the sources of next separation after proper choice and combination. Then using BSS technique repeatedly, residual parts of the noise can be separated. The simulations verified the effectiveness and adaptability of the proposed. What is more, the novel way of using 3D similitude phase graph is also proposed in this paper.