Abstract
The idea of employing harvested energy from human motion to run electronic devices such as self-powered sensors in fitness gadgets is attracting increased attention of many researchers. However, there is still limited knowledge of energy characteristics generated by human motions. Moreover, the level of human activities varies during a day from sitting for several hours to running on a treadmill. This highlights a vital need for energy conservation strategies. This thesis aims to first obtain insight into the characteristics and availability of harvested energy from human activities, and second provide energy conservation techniques to guarantee self-powered nodes' stability and real-time data transmission.
To gain an understanding of energy harvested from human motion, we analyse two datasets including daily human routines and largest inertial gait dataset. Experimental results show that performing both data collection and real-time transmission using energy generated from human motion is difficult even in intense activities. However, real-time data transmission of high-volume sensory data employing the harvested energy is feasible if data compression is used. To guarantee self-powered node stability, we develop an energy harvesting model to estimate the required time to store sufficient energy to operate the gadget with a certain level of confidence.
Considering the dense coupling between energy availability and data transmission in self-powered sensors, we propose a buffer management model. The buffer management model benefits from a closed-form solution to obtain a reliable battery size for mobile self-powered sensors with an intermittent connection. We investigate the impact of different acceptable energy depletion and overflow rates on the battery size. We also use simulation to examine the applicability of this model in a real-world scenario and obtain adequate data buffer size.
Data reduction scheme is employed as another approach toward the efficient utilisation of both available energy and connection self-powered sensors. We investigate the impact of various compression ratios on both energy efficiency and precision of extracted information by Support Vector Machine (SVM) classifier. A statistical test is employed to compare the effectiveness of reconstruction algorithms on SVM classifier performance. Finally, we quantify the power saving associated with various compression ratios.