A fast convergence algorithm of Volterra adaptive filters

Volterra nonlinear adaptive filters are known to be superior, having characteristics of general applicability to nonlinear systems and global convergence which is guaranteed by a gradient adaptive calculating algorithm. How ever, such convergence can become slow and unstable easily when Volterra nuclei are dense. In this paper, the authors have derived theoretically important properties of the error curve shape of Gaussian signal inputs in a second-order Volterra nonlinear adaptive filter. On the basis of these analytical results, a fast convergence algorithm using a corrected direction and a renewed amplitude of an optimal weighting coefficient vector for white signal inputs and colored signal inputs is proposed and its efficacy is verified by computer simulations. (C) 2001 Scripta Technica.