Application of the Maximum Entropy method to estimating parameter distributions for sonar signal processing

Abstract

The Maximum Entropy (MaxEnt) method and Bayesian inference have been employed to incorporate environmental knowledge into the signal processor for a sonar detection application. The sonar receiver is a new Estimator‐Correlator structure that requires only that the probability density function (pdf) of the observation conditioned on the signal belongs to the exponential class, a requirement met by application of the MaxEnt method. Random but statistically correct realizations of the environment are constructed from the pdfs, and an acoustic propagation code is used to propagate acoustic energy through each realization of the environment in a Monte Carlo simulation. From the ensemble of received signals, statistical moments of the signal parameters are estimated and the MaxEnt method is again used to construct signal parameter pdfs. Using Bayesian inference, the predicted parameter pdfs are incorporated into the detection algorithm as a priori information. To evaluate the fidelity of the approach, the statistics of acoustic measurements made during a 1996 experiment in the Strait of Gibraltar are compared with MaxEnt pdfs and simulation predictions.