Online Data and Execution Profiling for Dynamic Energy-Fidelity Optimization in Body Sensor Networks

Power consumption in many BSN devices is dominated by the wireless transmission of raw sensed data. On-node data reduction techniques can be employed to enhance energy efficiency but often come at the expense of application fidelity. This energy-fidelity relationship, however, is subject to both inter- and intra-individual variations, calling for dynamically adaptable on-node signal processing techniques that adjust key parameters based on real-time phenomena. This work presents the tools, methods, and framework for dynamic energy-fidelity optimization, including online data profiling for information content, on-node data rate reduction techniques, energy and resource profiling for executing those techniques, and application fidelity assessment. This approach is demonstrated on an inertial BSN platform using data collected in a clinical study of tremor. Compared to static data reduction settings that are based on patient-specific data profiling, dynamic energy-fidelity optimization through online profiling is shown to reduce energy by 76% for a given data distortion or reduce distortion by 90% for a given energy.