Deflationary blind source extraction using an exact solution subspace searching scheme

Abstract

In this paper, we develop a new deflation-based blind source extraction (BSE) algorithm to estimate and extract source signals in a sequential fashion from the mixtures. At the beginning of this algorithm, the first source signal is estimated by a constrained optimization and its efficient updating rule. Then, the other source signals are estimated and extracted by an exact solution subspace searching-based deflation technique. The key merit is that this new technique can greatly alleviate the error accumulation problem suffered by traditional deflation methods. Using speech source signal separation as example, simulations are conducted to demonstrate the superior performance of the proposed algorithm in both source extraction and error-accumulation mitigation over typical simultaneous blind source separation and deflation-based BSE algorithms.