Groggy Wakeup - Automated Generation of Power-Efficient Detection Hierarchies for Wearable Sensors

Abstract

We present a framework for the automated generation of power-efficient state detection in wearable sensor nodes. The core of the framework is a decision tree classifier, which dynamically adjusts the activation and sampling rate of the sensors (termed groggy wakeup), such that only the data necessary to determine the system state is collected at any given time. This classifier can be tuned to trade-off accuracy and power in a structured fashion. Use of a sensor set which measures the phenomena of interest in multiple fashions and with various accuracies further improves the savings by increasing the possible choices for the above decision process.

An application based on a wearable gait monitor provides quantitative results. Comparing the decision tree classifier to a Support Vector Machine, it is shown that groggy wakeup allows the system to achieve the same detection accuracy for less average power. A simulation of real-time operation demonstrates that our multi-tiered system detects states as accurately as a single-trigger (binary) wakeup system, drawing substantially less power with only a negligible increase in latency.