Doctoral Thesis: Localization and Structure Learning in Reverberant Environments
Passive localization and tracking of a mobile emitter, and the joint learning of its reverberant 3D environment, is an important yet challenging task in a wide variety of application domains. In underwater acoustic monitoring with a receiver array, for example, a submarine may need to be tracked in a difficult setting with natural and man-made obstacles, such as seamounts or piers. If such obstacles occlude the line of sight from this vessel to the receivers, then the non-line of sight reflected arrivals of the reverberant environment must be leveraged for localization. Hence, we need to also precisely map these reflective boundaries in order to deliver robust localization performance. We propose a multi-stage global optimization and tracking architecture to approach this problem. Each stage of this architecture seeks to establish domain knowledge such as synchronization and initial environment estimation, which are inputs for the following stages of more refined algorithms. We also introduce a robust 2D neural network-based boundary estimation method, with the results indicating that it outperforms its alternatives in the literature. We seek to analyze the performance and reliability of this holistic approach, both in simulation and in real-life reverberant watertank testbeds that model the shallow-water underwater acoustic setting.
Thesis Supervisor(s): Professor Gregory W. Wornell
- Date: Wednesday, April 26
- Time: 10:00 am - 12:00 pm
- Category: Thesis Defense
- Location: 36-144