Low-Power Semantic Fault-Detection in Multi-Sensory Mobile Health Monitoring Systems

Multi-Sensory mobile health monitoring systems promise substantial improvements in the quality of healthcare. However, large-scale trials are uncovering key areas that inhibit long-term large-scale deployments, including power consumption and lifetime issues, and high communication overhead. Traditional techniques can efficiently resolve these issues while maintaining semantic fidelity of the sensed medical signal, but also amplify the signal's sensitivity to sensor faults, thereby reducing system safety. We propose a set of statistical techniques to optimize system power and bandwidth consumption, while adhering to signal fidelity and sensor fault diagnosis requirements. By defining signal fidelity in terms of its semantic value, and formulating the problem as a sensor subset selection wherein mutual information rather than aggregate signal quality is maximized, we show that power consumption in a wireless human gait monitoring system can be reduced by up to 78% while accurately estimating many functional gait assessment metrics and precisely diagnosing semantic faults.