Efficient Implementation of GMDA-based DOA Technique Using Pre-training Phase Unwrapping for Source Localization

Abstract

In this paper, a novel technique that improves the performance of generalized mixture decomposition algorithm (GMDA) based on pre-training phase unwrapping. From the investigation of the GMDA scheme, it was discovered that the conventional GMDA technique cannot fully consider phase unwrapping, because the estimated inter-channel phase difference (IPD) slope is initialized randomly. To avoid this phenomenon, the proposed GMDA approach initialized the IPD slope from the data of low-frequency bins. Experimental results show that comparing to the conventional GMDA technique, the proposed GMDA technique based on pre-training phase unwrapping obtains a lower estimation error. When integrated into a source localization system, the result of source localization is improved.