Detecting simultaneous directions of arrival in an Ambisonic signal with REVEB-ESPRIT

Paper i proceeding, 2023

We present a purely real-valued variant of the extended vector-based EB-ESPRIT (REVEB-ESPRIT), an algorithm that estimates multiple simultaneous directions of arrival (DOAs) from Ambisonic signals, which are either encoded mono sounds or captured via a spherical microphone array. Our proposal uses fully real-valued spherical harmonics and DOA vectors and presents the required extended set of recurrence relations. Moreover, we propose a real-valued joint Schur decomposition using inverse iterations to efficiently solve the simultaneous diagonalization problem that is inherent in EB-ESPRIT algorithms. We evaluate the proposed algorithm in free-field conditions with a varying number of simultaneously estimated DOAs and varying signal-to-noise ratios. Our analysis shows a slight increase in speed and accuracy due to the proposed real-valued formalism, and in particular a noticeable increase in speed and accuracy when detecting many simultaneous DOAs. A reference implementation of the proposed algorithm is provided online.