An experimental comparison between maximum entropy and minimum relative-entropy spectral analysis

摘要

It is known that the minimum relative-entropy spectral analysis (MRESA) of J.E. Shore and R.W. Johnson (1980) and J.E. Shore (1981) provides an even better power spectrum estimation than the maximum entropy spectral analysis (MESA) by its explicit inclusion of a prior estimate that catches certain important shape features of the unknown true power spectrum. Since the kinds of prior estimates that catch certain important shape features of the unknown true power spectrum are generally hard to determine, a comparison between MESA and MRESA is undertaken in order to justify MRESA. The results indicate that unless the shape information of the true spectrum is correctly caught by the prior spectrum, it is safer to use MESA.