ApproxBioWear: Approximating Additions for Efficient Biomedical Wearable Computing at the Edge

Wearables in the biomedical domain have been of extensive use in the current era. Given the importance of wearable computing, it has become necessary to innovate on enhancing hardware efficiency. The domain of approximate computing offers a conclusive method to lower area, power and delay in hardware in addition to a marginal loss in accuracy. In this paper, we investigate ApproxBioWear, a technique which enables the use of approximate computing for efficient biomedical wearable computing at the edge. The methodology involves approximating additions during the functional stages of an error-resilient biomedical signal processing algorithm and determining the application accuracy. Upon evaluating the Pan-Tompkins algorithm, which is used to detect QRS peaks in ECG signals, it is observed that the ApproxBioWear approach reduces the power consumption and chip area by 19.27% and 19.71% respectively on an average with a marginal loss in accuracy.