Parametric modeling and estimation of time-varying spectra

In recent years research in spectrum estimation has increasingly shifted toward studies of nonstationary random signals or noisy signals with time-varying parameters. One Bayesian approach to this problem corresponds to finding the minimum mean square error estimate of the theoretical spectrum, which is the expected value of the theoretical spectrum over the joint posterior density function of the unknown parameters. In this paper we present an approach of estimation of time-varying spectra, where the signal is modeled as a superposition of chirps with Gaussian envelopes and the applied methodology is Markov chain Monte Carlo sampling.