Active noise control of impulsive noise with selective outlier elimination

Traditional active noise control (ANC) methods are based on adaptive filtering algorithms designed to minimize the noise variance. The convergence of such algorithms may be jeopardized in the presence of non-Gaussian noise signals, characterized by a marked impulsiveness (and infinite second-order moments), such as are frequently encountered in real-world acoustic settings. ANC methods have been recently extended to deal with such signals, modifying the weight update of the adaptive filter so that out-of-range samples are discarded or discounted. These methods require precise a priori knowledge of the impulsive characteristics of the noise and are generally not suitable for signals where such characteristics are time-varying. This work introduces an algorithm, based on an adaptive box-plot approach for outlier detection, which does not rely on any a priori information and yields uniformly high attenuation performance in all conditions tested in simulation.