Improved functional link artificial neural network filters for nonlinear active noise control

Abstract

The functional link artificial neural network (FLANN) structure using trigonometric function expansion is used successfully in nonlinear active noise control (NANC) system. However, there are still two potential shortcomings to deteriorate its performance. One is that the FLANN filter may perform poorly when the cross-terms exist in NANC system, and the other is that the FLANN filter cannot obtain its optimal solution in some complex nonlinear cases because its nonlinear parts have intrinsic coefficient dependences between each other. To surmount these shortcomings, improved FLANN (IFLANN) filter and simplified IFLANN (SIFLANN) filter are proposed in this paper. To reduce the computational complexity further, the filtered-error least mean square (FELMS) algorithm is considered in the NANC system using these two proposed filters. IFLANN and SIFLANN filters solve these problems by inserting a corrective filter before trigonometric function expansion to offer suitable cross-term delay samples and counterbalance the nonlinear coefficients calculation. Moreover, detailed computational complexity analysis and extensive simulations with different reference signal and acoustic path are provided to demonstrate the effectiveness of the proposed IFLANN and SIFLANN filters. Compared to FLANN filter and its other improved versions, the proposed IFLANN and SIFLANN filters have better performance on computational burdens and noise attenuation.