Hardware-Software Inexactness in Noise-aware Design of Low-Power Body Sensor Nodes

Wireless Body Sensor Nodes (WBSNs) are miniaturized and ultra-low-power devices, able to acquire and wirelessly transmit biosignals such as electrocardiograms (ECG) for extended periods of times and with little discomfort for subjects [1]. Energy efficiency is of paramount importance for WBSNs, because it allows a higher wearability (by requiring a smaller battery) and/or an increased mean time between charges. In this paper, we investigate how noise-aware design choices can be made to minimize energy consumption in WBSNs. Noise is unavoidable in biosignals acquisitions, either due to external factors (in case of ECGs, muscle contractions and respiration of subjects [2]) or to the design of the frontend analog acquisition block. From this observation stems the opportunity to apply inexact strategies such as on-node lossy compression to minimize the bandwidth over the energyhungry wireless link [3], as long as the output quality of the signal, when reconstructed on the receiver side, is not constrained by the performed compression. To maximise gains, ultra-low-power platforms must be employed to perform the above-mentioned Digital Signal Processing (DSP) techniques. To this end, we propose an under-designed (but extremely efficient) architecture that only guarantees the correctness of operations performed on the most significant data (i.e., data most affecting the final results), while allowing sporadic errors for the less significant data [4]. In particular, the paper describes how the knowledge of the noise corrupting biosignals can be leveraged to minimize power consumption on WBSNs. Section II puts our work into context by acknowledging related efforts in the field, Section III describes an application-level solution based on lossy compression and low-power sensing, while Section IV proposes significance-based DSP as an architectural technique for inexact and ultra-low-power biosignals processing. Judicious implementation of these approaches results in substantial increases in WBSN energy efficiency, with minimal degradation of the quality of biosignal acquisitions.