Energy-efficient collection of wearable sensor data through predictive sampling

Abstract

Wearable sensors have emerged as viable and attractive solutions for monitoring the health of people under risk of major problems such as hypertension, heart attacks, and athletes overstressing their bodies. These devices would report on the status of certain body organs to a gateway node over wireless links. A major challenge for effective use of these miniaturized devices is sustaining their operation using a limited energy supply. Therefore, minimizing energy consumption is a crucial design goal. In this paper we propose a novel approach for reducing the volume and frequency of data transmission through data sample prediction. Our methodology is based on applying advanced machine learning techniques to determine when data transmissions are skipped, and by implicitly making the gateway aware of the omitted samples in order to achieve accurate signal reconstruction. The paper also presents a data quantization technique for increased throughput and reduced energy overhead while sustaining desired medical assessment accuracy. Furthermore, a packet formation algorithm is proposed to leverage the available buffering space to improve bandwidth and energy utilization subject to latency constraints. The effectiveness of our approach is validated using publicly available Electrocardiography and Electromyography datasets and is shown not only to outperform conventional data compression methods but can also be applied in conjunction with them.