SIGNAL PROCESSING AND ACOUSTIC EVENT DETECTION FOR WIRELESS SMART SENSORS

A wide range of Wireless Sensor Network (WSN) applications are event-driven, therefore event detectors are a crucial component that ought to furnish opportune and reliable event estimations. Particularly, event detection computing demands from sensor nodes intensive signal processing tasks. Rather, the wireless sensor platform resources are limited and can not execute efficiently complex signal processing algorithms. Moreover, available radio chips can not transmit into the communication channel large amounts of raw sensor data. In this paper the development of a lightweight event detector for acoustic events is presented. Threshold-based event detection methods require, on one hand off-line time consuming parameter tuning processes, and on the other hand the detector execution should perform continuously energy expensive point-to-point comparisons. As an alternative, we compute events occurrences, on the change of rate of the sum of the signal energy average plus the energy standard deviation at two separated time instants. Our approach works, tinder the assumption that for long time periods of unknown signal energy noise background levels, the rate slope will present flat or stable behavior, and it will experiment noticeable changes under sudden signal energy level changes for brief time instants. The detector reactivity is presented as function of two parameters: a) the change of rate slope tolerance and b) the time distance length among consecutive slope rate computations.