Abstract
Two sensor broadband direction of arrival (DOA) estimation suffers from an inherent lack of dimensionality due to having just two sensors, yet humans and other animals are able to overcome this limitation using subtle variations introduced by the ears. Application of existing DOA estimation techniques to such systems becomes complicated due to the ill-behaved nature of the Head Related Transfer Function (HRTF). In this paper we present a subband signal extraction and focussing technique which retains the diversity information of the HRTF. We then develop a framework for combining these signals for subspace DOA estimation and investigate the constraints imposed on the single and multi-source DOA estimation problems. Finally, estimation performance is compared with existing techniques and we find performance has improved to be comparable to human localisation abilities. © 2012 IEEE.