Improving convergence in finite word length nonlinear active noise control systems

Abstract

An attempt has been made in this paper to improve the convergence of functional link artificial neural network (FLANN) based nonlinear active noise control (ANC) systems. This improvement has been achieved by formulating a recursive least square (RLS) training mechanism. However, FLANN-RLS ANC systems are not effective in noise mitigation when implemented in a finite word length scenario. A QR-RLS based training mechanism has been designed to improved convergence even in reduced word length implementations. A simulation study has been carried out to study the effectiveness of the proposed scheme in improving convergence when finite word length implementation is attempted. The proposed FLANN-QRRLS scheme has been shown to improve convergence behaviour in comparison with other schemes compared.