Detection and classification of spectrally equivalent processes: a parametric approach

Abstract

The detection of two spectrally equivalent (SE) processes is addressed. The two SE processes are modeled using two SE parametric models: the noisy AR model and the ARMA model. Higher-order statistics are shown to be an efficient tool for the SE process detection problem. A new detector based on the higher-order Yule-Walker matrix singularity is studied. The detector performance is compared in supervised and unsupervised learning. The model order mismatch is then studied.