An ML algorithm for outliers detection and source localization

The problem of simultaneous detection of outliers and localization of multiple sources is addressed. This is motivated by the performance degradation observed when quadratic beamformers operate under those conditions. The approach relies on maximum likelihood (ML) methods where outliers are modeled as a space/time impulsive noise process with unknown statistics. The maximization algorithm follows a strategy based on sequential estimation and detection schemes, and it is initialized by an I₁ beamformer, yielding efficient detection of spikes and accurate estimates of their statistics. This makes it possible to design a model-based beamformer for bearing estimation. The derivation of the algorithm is presented, and its efficiency is discussed using the results obtained from computer simulations