Adaptive active noise control with online secondary path modelling and variable step-size learning

Abstract

The performance of conventional filtered-X least mean squares (FXLMS)-based active noise control (ANC) systems may degrade owing to uncertainties in the secondary path model and constant learning gains (step-size parameters) used in the FXLMS algorithms. In this paper, a new ANC feedforward system design with online secondary path modelling and variable step-size parameters (VSSPs) is proposed. To improve the convergence performance of FXLMS algorithms, an online tuning scheme of step-size parameters (or learning gains) for the adaptive learning is developed based on the residual errors. In particular, this paper shows how to analyse the stability of the proposed closed-loop ANC systems and to study the convergence of the presented adaptations. Moreover, a modified online modelling strategy is introduced to address the modelling of secondary path dynamics. The modelling and adaptive control are all online implemented without any offline learning phase; faster convergence and better noise elimination can be achieved. Appropriate comparisons to other methods and their computational complexities are also investigated. Comparative simulation results illustrate the improved performance of the proposed methods.