Theory and applications of parametric estimation methods for sensor array signal processing

Sensor array signal processing has been an area of great research interest over the last thirty years. Techniques developed in tins area have played important roles for a wide variety of applications such as radar, sonar, navigation, and geophysics. Recent, developments in the integrated circuit technologies have further allowed the construction of low-cost, high performance sensor nodes with communication and signal processing capabilities. These technologies not only open up many possibilities but also bring in many challenging issues. The goal of the research performed in this dissertation is to address some of these issues and demonstrate our efforts and accomplishments over the past few years.;The focus of this dissertation is mainly on the parametric methods for array signal processing. We have proposed a new model order section rule under the Bayesian's framework which estimates the number of wideband sources impinging the array within a single observation frame. We have also applied the Maximum Likelihood (ML) algorithm to nonuniform noise scenarios and derived a new deterministic wideband ML estimator for the Direction of Arrival (DOA) estimation. The asymptotic performance of the resulting estimator is then studied by comparing its Mean Square Error (MSE) to the derived theoretical Cramer-Rao Bound (CRB). In addition to the deterministic ML estimator, we have proposed a stochastic nonuniform ML estimator for narrowband DOA estimation. Combining the stepwise concentration technique and a novel modified inverse iteration procedure, we are able to obtain a solution that attains the CRB at a moderate computational complexity.;In the latter part of the dissertation, we apply several stochastic techniques to the wideband source localization/tracking problem. We derive a stochastic optimization procedure for the ML estimator using the Cross-Entropy method. Fast convergence is observed through computer simulation, and the convergence rate appears to be insensitive to the coherence of signals. We have also developed an acoustic source tracking algorithm that combines the ML estimator with the particle filtering technique. By exploiting the temporal consistency of target location estimates, the spurious location estimates caused by the room reverberation can be effectively mitigated.;The developed techniques and algorithms have been validated using data collected by the designed microphone array. The success in the localization and tracking experiments suggests both the practicality and feasibility of this dissertation.