Analysis of explicit regularization in affine projection algorithm: robustness and optimal choice

HERNÁN REY; LEONARDO REY VEGA; SARA TRESSENS; BRUNO CERNUSCHI-FRÍAS

Viena, Austria

Conferencia; European Signal Processing Conference, EUSIPCO; 2004

European Association for Signal Processing, EURASIP

The affine projection algorithms (APA) are a family of adaptive filters with decorrelating properties. In this paper we propose the use of a time-varying explicit regularization factor instead of the classic  step size control. We show that APA results stable and has robust performance against disturbances and model uncertainties. This fact is shown by the H$infty$ optimality and error energy  bounds we provide. Finally, we find the optimal regularization choice for maximum speed of convergence. We consider first the case under independence assumptions and then with gaussian hypothesis.