Adaptive recurrent fuzzy neural networks for active noise control

Abstract

This paper discussed nonlinear active noise control (ANC). Some adaptive nonlinear noise control approaches using recurrent fuzzy neural networks (RFNNs) were derived. The proposed RFNNs were feed-forward fuzzy neural networks (NNs) with different local feedback connections that are used to construct dynamic fuzzy rules. Different recurrent connection strategies, diagonal recurrent and full connected recurrent ones, were considered. In addition, different fuzzy operation strategies, product (multiply) inference and “summation” (addition) inference, were proposed. Because RFNN-based ANC systems can capture the dynamic behavior of a system through the feedback links, the exact lag of the input variables need not be known in advance. Online dynamic back-propagation learning algorithms based on the error gradient descent method were proposed, and the local convergence of a closed-loop system was proven using the discrete Lyapunov function. A nonlinear simulation example showed that an adaptive ANC system based on an RFNN with summation inference is superior to a system based on other fuzzy NNs.