Decision- and classification-directed methods in nonstationary signal analysis

Abstract

An examination is made of alternatives to traditional frequency analysis of nonstationary signals using a decision-directed estimation methodology. This methodology is used to estimate the probability structure of signal energy at discrete frequencies. The method presented utilizes possible harmonic structure in signals of interest by using adaptive coupling of detectors at harmonically related frequencies. This coupling is directed according to signal classifications made on the marginal detector decision outputs. The method given is less computationally intensive than estimation of the full joint probability distribution. Results show improvement over marginal detection alone given true Bayesian statistics.<>