Robust Expectation–Maximization Direction-of-Arrival Estimation Algorithm for Wideband Source Signals

Direction-of-arrival (DOA) estimation for wideband source signals using far-field acoustic sensors has recently drawn much research interest. A wide variety of DOA estimation approaches are based on the maximum-likelihood objective. In this paper, we tackle the DOA estimation problem based on the realistic assumption that the sources are corrupted by spatially nonwhite noise. We explore the respective limitations of two popular DOA methods to solve this problem-the stepwise-concentrated maximum-likelihood (SC-ML) and approximately concentrated maximum-likelihood (AC-ML) algorithms-and design a novel expectation-maximization (EM) algorithm. In addition, we provide the Cramer-Rao lower bound (CRLB) and the computational-complexity analyses for the aforementioned DOA estimation schemes. Through Monte Carlo simulations and our derived CRLB and computational-complexity analyses, it is demonstrated that our proposed EM algorithm outperforms the SC-ML and AC-ML methods in terms of the DOA estimation accuracy and computational complexity.