A Class of Affine Projection Filters that Exploit Sparseness under Symmetric alpha-Stable noise

A new algorithmic framework for sparse underwater acoustic channel estimation under the presence of Symmetric alpha-Stable (S alpha S) noise is proposed. The framework is inspired by the low-complexity affine projection algorithm (APA) and employs a novel cost function that includes robust methods for impulsive noise suppression and a sparseness constraint. The constraint exploits the non-isotropic (Riemannian) structure of the channel, which leads to natural gradient adaptation. Based on this framework, new algorithms are derived with linear algorithmic complexity. The effectiveness of the new algorithms is confirmed by computer simulations.