Power Spectrum Estimation Using Maximum Entropy Method

Abstract

The present paper is a review and an appraisal of the Maximum Entropy Method (MEM) of power spectrum estimation. Originally suggested by Burg, this method eliminates the necessity of some of the artificial assumptions associated with conventional methods of power spectrum estimation and provides a better resolution performance. After a brief discussion of these methods, the principle of MEM is discussed. It is shown that the antoregressive (AR) spectral estimator of Parzen or Wiener filter approach is similar to MEM, though the two approaches make different assumptions on the time series. This leads to a recursive procedure for implementing the MEM method. Criteria for establishing the order of the AR model are also discussed. The study includes quite a few numerical results illustrating the behaviour of MEM spectral estimator, a brief review of the applications of MEM to other signal processing situations and an extensive bibliography.