DOA detection from HOS by FOD beamforming and joint-process estimation

Abstract

Array Beamforming is a powerful technique in Speech Enhancement, Noise Reduction, Source Separation, etc. for which powerful techniques have been developed [8]. Nevertheless, large Arrays present several inconveniences, as are sensor equalization, complex DOA algorithms, high costs, large computational requirements, etc. This lead to exploring other possible structures based on paired sensors, as First-Order Differential Beamformers (FODB) [2]. These structures may be steered to aim their sharp notch to the desired source, which may be removed from the output, and complementarily reconstructed using several methods, as direct or spectral subtraction [1], or joint-process estimation [5]. The main problem that these systems present is DOA estimation in the presence of reverberation. Through this paper it is shown that the use of Higher Order Statistics may help in detecting DOA's. Results for simulated source separation, and DOA detection in a real room are given and discussed. © Springer-Verlag 2004.