Energy-Accuracy Tradeoff for Efficient Noise Monitoring and Prediction in Working Environments

Abstract

We explore the tradeoff between energy consumption and measurement accuracy for noise monitoring and prediction based on continuously collected data by wireless, energy-constrained IoT nodes. This tradeoff can be controlled by the sampling interval between measurements and is of interest for energy-efficient operation, but most of ten ignored in the literature. We study the influence of the sampling intervals on the accuracy of various noise indicators and metrics. To provide a context for the tradeoff, we consider the use case of noise monitoring in working environments and present a learning algorithm to also predict sound indicators. The results indicate that a proper tradeoff between energy consumption and accuracy can save considerable energy, while only leading to acceptable or insignificant reductions in accuracy, depending on the specific use case. For instance, we show that a system for monitoring and prediction can perform well for users and only uses around 7% of the energy compared to full sampling.