Abstract
Chi (see Sixth European Signal Processing Conference, EUSIPCO-92, Belgium, p.755-8, vol.2, Aug. 1992) proposed a higher order statistics (HOS) based linear prediction error (LPE) filter, a moving average (MA) filter, for nonGaussian stationary processes. The authors present a generalization of this filter to an autoregressive moving average (ARMA) filter and an algorithm for practically solving the ARMA parameters. Then, by simulation, they show that the smaller signal-to-noise ratio (SNR), the more Chi's HOS based LPE filter outperforms the conventional correlation based LPE filter for the case of finite nonGaussian measurements in the presence of additive Gaussian noise