Impact of Approximate Computing Adder on Arrhythmia Identification Accuracy

The electrocardiogram (ECG), which is useful in identifying arrhythmias, should be measured and diagnosed continuously due to the infrequent occurrence of these conditions. Therefore, ECG devices should be wearable, facing the challenges of battery life and its replacement risk, and their improved power efficiency is needed. Approximate computing techniques have emerged as a promising solution to achieve power savings, with some compromises in computing accuracy, and they are applicable in biomedical signal processing. In this study, we applied an approximate addition to an algorithm for detecting and classifying arrhythmias as a power-saving solution for wearable ECG devices. We investigated approximate adders from EvoApproxLib to assess their impact on the accuracy of arrhythmia detection and classification, and to verify whether the approximation is acceptable. Our experimental evaluation demonstrated that approximate adders can be effectively applied to arrhythmia detection and classification, achieving high accuracy when the appropriate portions of the algorithm are selected for approximation. This approach offers the potential for power-efficient ECG devices by reducing the power consumption associated with additions.