Abstract
In this paper we present a neural-based algorithm to cancel the nonlinear narrowband and broadband noise in an active noise control (ANC) system. The improved method, including the ways to decide the learning rate and optimal initial weighting values of the neural network (NN), are presented to enhance the noise reduction performance. The proposed approach does not need mathematical transfer functions of the duct plant and a method of avoiding the premature saturation problem of NNs is also provided. A comparison with conventional neural methods by simulation shows that the proposed method can effectively cancel the undesired noise very well. The proposed improved method is also versatile to the other applications in NN filter design.
Get full access to this article
View all access options for this article.
