Abstract
The standard maximum entropy method developed by J.P. Burg (1967) and the resulting autoregressive model have been widely applied to spectrum estimation and prediction. A generalization of the maximum entropy formalism in a nonparametric setting is presented, and the class of the resulting solutions is identified to be a class of Markov processes. The proof is based on a string of information theoretic arguments developed in a derivation of Burg's maximum entropy spectrum by B.S. Choi and T.M. Cover (1984). A framework for the practical implementation of the proposed method is presented in the context of both continuous and discrete data