Abstract
To reduce the interference of impulsive noise when the spline adaptive filter (SAF) algorithm is used to identify nonlinear systems, this paper proposes a family of SAF algorithms using the Heaviside step function (HSF). The suitability of those cost functions proposed are investigated; those cost functions are design based on some HSF’s approximate functions. Then based on that, four SAF algorithms have been developed: SAF-HSF-sigmoid, SAF-HSF-erfc, SAF-HSF-atan, and SAF-HSF-tanh. Also, the bound of the learning rate has been derived for those proposed algorithms. The proposed SAF-HSF algorithms have been evaluated for nonlinear system identification and simulation studies to demonstrate their robustness.