Second-order Volterra Filter based on DFP Technique

Abstract

There have some problems caused by selection of improper parameters when LMS, NLMS or RLS algorithms are used to evaluate coefficients of a second-order Volterra filter. We propose a Davidon-Fletcher-Powell-based second-order Volterra filter (DFP-SOVF). The proposed filter is based on a posteriori error assumption and has a variable convergence factor. We give the recursive inverse auto-correlation matrix formulation and present computational complexity analysis. Then applying the proposed DFP-SOVF filter to single step predictions for pure Lorenz chaotic series, prediction results show that the DFP-SOVF filter can guarantee its convergence and stability and there have not divergence problems when using LMS and NLMS algorithms.