An improved pre-processing approach for convex-geometry based blind source separation

Existing studies have shown that the problem of blind source separation of quasi-stationary sources (BSSQSS) with local dominance property can be solved with the convex geometry method. However, it is also illustrated that its performance can seriously degrade if sources are correlated. To address the source cross-correlation problem, in this paper, an effective pre-processing approach is proposed to suppress the cross-correlation component. In contrast to the previous research, we make a tradeoff between the degree of source cross-correlation suppression and the violation of convex geometry, by introducing a novel control parameter. Experiments in white noise and interference environments illustrate that the proposed method offers not only competitive efficiency and robustness in comparison to state-of-the-art methods, but also a better accuracy for high signal-to-noise ratio regime. (C) 2021 Elsevier Inc. All rights reserved.

Automated summary

This paper proposes an effective preprocessing approach to address the source cross-correlation problem in blind source separation, achieving competitive efficiency, robustness, and accuracy in comparison to existing methods.