Reference weighted filtered-x LMS algorithm for active control of impulsive noise

This paper proposes a novel reference-weighted filtered-x least mean squares (RWFxLMS) algorithm for active control of impulsive noise. The paper briefly discusses the existing generalized step-size normalized filtered-x least mean-square (FxGSNLMS) algorithm that inspires the development of the proposed algorithm, which simplifies the computations without apparently sacrificing the control performance. The RWFxLMS algorithm uses a quadratic function of the reference signal obtained by curve-fitting the probability distribution of an impulsive signal to weight the step-size at that instant. Numerical simulations are conducted to validate the effectiveness of the proposed algorithm for the active noise control under non-Gaussian noise environment including impulsive noise with symmetric a-stable (s-a-s) distribution. Further, the convergence property and the computational complexity of the proposed algorithm are briefly discussed and compared with three existing effective impulsive noise control algorithms. The simulation results demonstrate that the proposed RWFxLMS algorithm is able to achieve comparable reduction of impulsive noise for a wide range of impulsiveness with a considerably reduced computational cost as compared to the existing adaptive algorithms for impulsive noise control.