On Estimating the Spectral Density Function of a Stochastic Process

Abstract
Summary: On the basis of an expected-mean-square-error criterion, the authors discuss the various hitherto proposed estimators of the spectral density function of a discrete-parameter process; they also introduce a new estimator. Any of these estimators depends on some parameter, and indications are given as to how this parameter should match those of the process in order to optimize the estimator. The alleged conflict between resolvability and statistical reliability is discussed. By way of illustration, the case of a first-order Markov process is treated in detail. Numerical applications are also given.