Date of Award
Master of Science (MS)
Chester E. Grosch
Predicting sonar performance, critical to using any sonar to its maximum effectiveness, is computationally intensive and typically the results are based on data from the past and may not be applicable to the current water conditions. This paper discusses how Beowulf clustering techniques were investigated and applied to achieve real-time sonar performance prediction capabilities based on commercially off the shelf (COTS) hardware and software. A sonar system measures ambient noise in real-time. Based on the active sonar range scale, new ambient measurements can be available every 1 to 24 seconds. Traditional sonar performance prediction techniques operated serially and often took approximately 120 seconds of computing time per prediction. These predictions were outdated by potentially several sonar measurements. Using Beowulf clustering techniques, the same prediction now takes approximately 2 seconds. Analysis of measured data using a sonar hardware suite reveals that there is a set of sonar system parameters where a serial approach to sonar performance prediction is more efficient than Beowulf clustering. Using these parameters, a sonar engineer can make the best decision for system prediction capability based on the number of sonar beams and the expected operational range. The paper includes a discussion on the taxonomies of parallel computing, the historical developments leading to measuring the speed of light, and how those measurements enable acoustic paths to be computed in ocean environments.
Cartledge, Charles L..
"Investigating Real-Time Sonar Performance Predictions Using Beowulf Clustering"
(2007). Master of Science (MS), thesis, Computer Science, Old Dominion University, DOI: 10.25777/w2ry-5163