ABSTRACT
Modern spectral analysis dates back at least to Sir Isaac Newton, whose prism experiments with sunlight led him to discover that each color represented a particular wavelength of light and that the sunlight contained all wavelengths. Newton used the word spectrum, a variant of the Latin word specter, to describe the band of visible light colors. The Bayesian, minimax, minimum variance unbiased, and maximum likelihood estimation schemes are members of the class of parametric parameter estimation procedures. Nonparametric estimation schemes may result as solutions of certain saddle-point games, whose payoff function originates from parametric maximum likelihood formalizations. An estimation algorithm uses the observations and the model to improve its assessment of the base state, to compute uncertainty regions for the state, and to predict the future evolution of the system. Pseudo-noise desensitizes the estimates to the modeling errors but sacrifices the specificity that careful modeling could provide. Additionally, the computed error covariances are broader than they should be.
