Algorithm Development for a Real-Time Military Noise Monitor

Abstract : The long-range goal of this 1-year SERDP Exploratory Development (SEED) project was to create an improved real-time, high-energy military impulse noise monitoring system that can detect events with peak levels (Lpk) as low as 100 dB with a high degree of accuracy and post the results in readily usable format. Toward this goal, this phase of the project was concerned with field data collection of noise measurements to support algorithm development, processing of the signals with software to extract standard signal metrics, development of new metrics that would improve classification accuracy, and the development, training and evaluation of an artificial neural network (ANN) that used the signal metrics to determine whether a particular noise source was military impulse noise or not. Current noise monitors suffer from inaccuracies since they look for a particular shape in the noise waveform, which can be highly variable. It is reported that detection of signals with Lpk below 115 dB is difficult or impossible and that false positives (as high as 10%) can occur (primarily from wind triggers) [SERDP, 2003]. Data collection trips were conducted to the US Marine Corps Base Camp Lejeune (MCBCL), NC (military and wind noise sources), Fort Indiantown Gap (FTIG), PA (military and wind noise sources), central Ohio (wind noise source) and suburban Pittsburgh (wind noise source). These data collection trips yielded approximately 1,000 usable waveforms (330 military impulse and 670 non-impulse events). A custom data collection system was assembled from various pieces of hardware and software to permit the collection of intermittent (impulse) or continuous (wind) sampling of signals.