Robust Expectation-Maximization Algorithm for DOA Estimation of Acoustic Sources in the Spherical Harmonic Domain

TL;DR

This paper introduces a robust EM algorithm for accurate DOA estimation of acoustic sources using spherical microphone arrays, effectively handling noise and reverberation in the spherical harmonic domain.

Contribution

It proposes a new EM-based method that reduces computational complexity and improves robustness in DOA estimation within noisy, reverberant environments, along with a derived CRB for performance benchmarking.

Findings

At least 6dB improvement in robustness over existing methods.

RMSE close to the derived CRB in noisy, reverberant conditions.

Significant reduction in computational complexity.

Abstract

The direction of arrival (DOA) estimation of sound sources has been a popular signal processing research topic due to its widespread applications. Using spherical microphone arrays (SMA), DOA estimation can be applied in the spherical harmonic (SH) domain without any spatial ambiguity. However, the environment reverberation and noise can degrade the estimation performance. In this paper, we propose a new expectation maximization (EM) algorithm for deterministic maximum likelihood (ML) DOA estimation of L sound sources in the presence of spatially nonuniform noise in the SH domain. Furthermore a new closed-form Cramer-Rao bound (CRB) for the deterministic ML DOA estimation is derived for the signal model in the SH domain. The main idea of the proposed algorithm is considering the general model of the received signal in the SH domain, we reduce the complexity of the ML estimation by breaking it down into two steps: expectation and maximization steps. The proposed algorithm reduces the complexity from 2L-dimensional space to L 2-dimensional space. Simulation results indicate that the proposed algorithm shows at least an improvement of 6dB in robustness in terms of root mean square error (RMSE). Moreover, the RMSE of the proposed algorithm is very close to the CRB compared to the recent methods in reverberant and noisy environments in the large range of signal to noise ratio.