Low power programmable architecture for periodic activity monitoring

Body sensor networks (BSNs) are considered a great example for cyber-physical systems due to their close coupling with human body. Activity monitoring is one of the numerous applications of BSNs. Continuous and real-time monitoring of human activities has many applications in health-care and wellness domains. BSNs utilizing light-weight wearable computers and equipped with inertial sensors are highly suitable for real-time activity monitoring. However, power requirement is a major obstacle for miniaturization of these wearable systems, due to the need for sizable batteries, and also limits the life time of the system. In this paper, we propose a low-power programmable signal processing architecture for dynamic and periodic activity monitoring applications which utilizes the properties of the physical world (i. e., human body movements) to reduce the power consumption of the system. The significant power reduction is achieved by performing signal processing in a tiered-fashion and removing the signals that are not of interest as early as possible. Our proposed architecture uses wavelet decomposition and is favorable for the discrimination of periodic activities. The experimental results show that our architecture achieves 75.7% power saving while maintaining 96.9% sensitivity in the detection of target actions, compared with the scenario where the signal processing is not performed in tiered-fashion. This creates opportunities to enable the next generation of self-powered wearable computers.