Design of nonlinear filters using affine projection algorithm based exact and approximate adaptive exponential functional link networks

Abstract

Adaptive exponential functional link network (AEFLN) is a recently developed linear-in-the-parameters nonlinear adaptive filter. It has been observed that the convergence performance of the AEFLN filter deteriorates in the presence of colored and/or correlated inputs. To overcome this issue, an affine projection algorithm (APA) based AEFLN (AEFLN-APA) filter is proposed in this brief. To reduce the hardware complexity of the proposed filter, an approximate AEFLN-APA filter is also developed. The proposed APA-based filters are found to provide improved modeling accuracy in nonlinear system identification scenarios. In this brief, the proposed nonlinear filters are also applied to active noise control (ANC) systems. An adaptive exponential filtered-s APA (AEFsAPA) is developed to enhance the noise mitigation capability of an ANC system. In addition, an approximate AEFsAPA is also formulated to achieve reduced-complexity implementation of AEFsAPA. Simulation results demonstrate the improved convergence behavior of the proposed schemes in nonlinear active noise mitigation scenarios.