A design methodology for the optimal estimation of the spectral characteristics of non-stationary random signals

The authors outline a methodology for the design of estimators for the evolutive (or Wigner-Ville) spectrum of a harmonizable Gaussian random process, which are optimized to a priori specified properties. Such a property might be the minimization of bias in a specified region of the time-frequency plane consistent with a minimum desired resolution in frequency. The design utilizes Karhunen-Loeve type expansions for the Wigner-Ville distribution of the process to specify the one-dimensional probability distributions for the proposed estimates. An iterative scheme based on tailoring the properties of the estimators to specified requirements is suggested. Various types of update strategies have been mentioned based upon evidence obtained by simple simulations.<
>