A Coordinated Ambient/Dedicated Radio Frequency Energy Harvesting Scheme Using Machine Learning

This paper proposes a coordinated ambient/ dedicated (CA/D) protocol where backscatter-enabled combination sensors are optimized to harvest energy from intended radio frequency (RF) sources when available and fall back to harvesting energy from unintended sources when only unintended sources are available. The CA/D protocol can use either of the two new machine learning techniques proposed, the linear forecaster with near-time linear regression-based enhancer (LFNTLRE) algorithm and an artificial neural network (ANN), to determine the optimum energy harvesting (EH) schedule. These machine learning algorithms can reliably operate in environments where there is unpredictable availability of unintended sources and ongoing changes in channel conditions between the sensors and the unintended sources. Numerical results show that sensors using the ANN and LFNTLRE algorithm can achieve up to 99.5% and 99.6% hit count relative to that of ideal sensors, respectively, compared to 100% by an ideal sensor. Sensors using the ANN and LFNTLRE algorithm have an accuracy percentage of up to 99% and 100%, respectively, as well. Ensuring reliable operation of RF EH sensors in all environments will enable its future widespread adoption in the real world.