A Method for Robustifying Classical Nonparametric Spectral Estimation Techniques

Abstract

In this study, robust nonparametric spectral estimation methods for non-Gaussian environments are proposed. For this aim, the autocorrelation function estimator obtained from sample spatial sign covariance matrix is used together with classical nonparametric spectral estimation methods such as periodogram and Blackman-Tukey. Performances of classical spectral estimation methods and robust methods suggested in this study are compared by applying them to one Gaussian process and one non-Gaussian heavy-tailed stochastic process. The results obtained show that, for non-Gaussian environments, the proposed robust nonparametric spectral estimation methods could perform better compared to the classical methods.