Active noise control using adaptive POLYnominal Gaussian WinOwed wavelet networks

The capabilities of wavelet networks in function approximation make them appealing for black box system identification. In this paper, a new active noise control (ANC) algorithm is developed based on adaptive wavelet networks. The proposed adaptive nonlinear noise control approach employs frames from POLYnominal WinOwed with Gaussian wavelets. Also, a novel network structure for active noise control is derived incorporating a nonlinear static mapping cascaded with an infinite impulse response filter to model the dynamic part of the network. Online dynamic backpropagation learning algorithms based on gradient descent method are applied to adjust the network parameters. Local convergence of the closed-loop system is proved using discrete Lyapunov function.The performance of the proposed ANC system is examined for typical linear/nonlinear cases. The simulation results demonstrate superior performance of this method in terms of stability, fast convergence rate and noise attenuation while avoiding curse of dimensionality.