A new eigenvector-based 3D wideband acoustic DOA estimator

Abstract

Many multiple narrow-band source detection and DOA methods have been proposed. In the past, we have used an Approximate Maximum-Likelihood (AML) method needing considerable computational complexity for the detection and 3D DOA estimation of multiple broad-band sources. Now, we propose a novel eigen system-based array detection and DOA estimation of multiple broad-band sources with significantly reduced computational complexity. After appropriate STFT, all N sensor array data for any subband index k can be used to form a cross-covariance matrix Ck. An eigen-decomposition yields [Vk, Dk] — eig(C k), where the ratios of the larger eigen values in Dk relative to the smallest eigenvalue provide a possible detection criterion for various sources and the corresponding eigenvectors can be used to estimate the DOA's of multiple sources. An N = 4 element array was used to collect a two-bird wide-band acoustic source to illustrate the proposed method. An intuitive clustering of different A-indexed subband was also used. The proposed method offers many advantages compared to AML for estimating the azimuth and elevation angles for multiple sources. The primary advantage is that source direction can be estimated based on a single subband frequency for up to N-1 sources. The eigen-analysis also avoids a processing complexity that is proportional to the number of snapshots. The impact of signal to interference ratio greatly reduced the ill-effect of the interference since different eigenvectors are essentially independent of each other. The integration over frequency to improve SNR can be put off until the sources have been isolated.