Date of Award
Thesis - Open Access
Master of Software Engineering
Electrical, Computer, Software, and Systems Engineering
Dr. Brian Butka
First Committee Member
Dr. Eric Coyle
Second Committee Member
Dr. Jianhua Liu
The intent of the work conducted was to build a neural network for the purposes of acoustic localization. The target of this localization is a sound source underwater. For our purposes, it is an acoustic pinger, as it produces consistent sound at a fixed rate making it ideal for testing. The network was intended to ingest raw data streams and output location information based on the arrangement of sensors employed. To achieve an accurate network, a simulation factoring in the environment was to be created to produce a data set large and diverse enough to describe the unique parameters of the signals, including: frequency, environmental reflections, and range.
This problem will be approached in multiple steps. Initial models will consider simplified problem spaces, such as individual frequencies and less descriptive training sets. Through development, this will be refined and extended. Where required, simplifications will be kept managing the scope of the problem to allow for a demonstration of the technology to be made at all. Discussion of what is the root cause of the issue navigated will be presented when this occurs. Results will then be shown to demonstrate the performance of the network created as compared to the classical approach to this problem, time difference of arrival.
This paper will demonstrate the performance of a neural network as applied to the problem of acoustic localization. The network developed can accurately localize an acoustic sound source to the same order of magnitude of accuracy and execution time as the current approaches to the problem. However, the network also showed a lacking in some areas of robustness due to training factors not considered, hampering the full potential.
Scholarly Commons Citation
Cronin, Stephen, "Application of Neural Networks to Acoustic Localization" (2017). Doctoral Dissertations and Master's Theses. 324.