Signal detection and estimation using classification-directed adaptive modeling

The topic of simultaneous signal detection and classification according to time-varying frequency content is examined. Decision-directed empirical Bayes procedures are used with scalar and vector Markov chain models for modeling the time-varying a priori probability structure. Discrete-time point processes are invoked as a modeling tool for signal presence and detection processes. To take advantage of harmonic signal structure in the nonstationary probability estimates, a new method using classification-directed adaptive modeling for the scalar Markov chain model is developed. Estimation methods employing independent estimation of the marginal probabilities and a distributed network realization of a coupled vector estimation procedure are also developed for comparative purposes. Monte Carlo analysis methods and actual sonar data are used in the evaluation of the various detection/estimation procedures.