Reduction of correlation computation in the permutation of the frequency domain ICA by selecting DOAs estimated in subarrays

Abstract

This paper addresses the permutation problem in blind source separation by frequency domain independent component analysis. We propose a method to reduce the correlation calculation in aligning permutation by selecting the DOAs (direction of arrival) estimated in several subarrays. When several subarrays are available, we have more choices to select a reliable DOA to determine the permutation. More reliable DOAs, more permutations can be determined by DOAs instead of the correlation of the separated signals between neighbouring frequency bins, which is much more complex than DOA estimation. To obtain accurate DOA in each subarray, we also propose an improved DOA estimation method by averaging the phase differences in each subarray. Experimental results show that about 58% of the correlation computation was reduced when compared with the case of a single array, and a more accurate DOA was estimated by the proposed method.