Improving Energy Efficiency in Medical Edge Devices for ECG Feature Detection via Approximate Computing

Improving the energy efficiency of battery-powered medical edge devices is a challenge. In this study, we explored electrocardiogram (ECG) feature detection using the Pan-Tompkins algorithm and integrated approximate computing techniques involving addition and multiplication into the process to enhance power efficiency. We adopted eight approximate multipliers from EvoApproxLib along with a Carry-Maskable Adder (CMA). We evaluated the impact of applying approximate computing on feature detection accuracy, as well as its influence on enhancing the power efficiency of the arithmetic circuits. When replacing just one of the arithmetic circuits with its approximate counterpart, the most power-efficient multiplier had dynamic power reduced by 48.3% in multiplications while maintaining a detection accuracy of 99.96%, while the CMA reduced dynamic power reduction by 95.2% while maintaining a detection accuracy 99.4%. When the CMA and approximate multiplication were applied, detection accuracy was maintained at 99.4%. In terms of power efficiency, the multiplier played a predominant role and the overall dynamic power consumption was reduced by 48.7%.