Robust Direction Estimation in the Presence of Spatially Correlated Noise

Abstract

In most algorithms for direction estimation of signal wavefronts, the additive noise term is assumed to be spatially white or known to within a multiplicative scalar. Since the surrounding environment and orientation of the array may be time varying, the requirement of known noise statistics is seldom satisfied in practice. At high signal-to-noise ratio (SNR) the deviation from these assumption are not critical. However, at low SNR, the degradation may be severe. By introducing a banded structure noise model, it is possible to estimate the noise covariance simultaneously as the direction parameters are estimated. This technique considerably reduces the bias on the direction estimates, that are induced by the colored noise.In this paper such a parameterization is proposed, and the asymptotic bias is investigated with respect to small perturbations in the noise model.