Experimental comparison of two information-theoretic spectral estimators

Abstract

The use of the minimum cross-entropy (MCE) principle in spectrum estimation, resulting in an information-theoretic method that explicitly includes prior spectral information, has followed two philosophically different pathways leading to two distinct estimators, cross-entropy (CE) and spectral cross-entropy (SCE) respectively. These estimators are compared experimentally, regarding resolvability and fidelity. The data model assumed consists of two equal-amplitude sinusoidal signals, immersed in 1/f noise. The CE estimator appears to have evident superiority in this study.