Towards At-the-Edge ECG Signal Processing with Accuracy-tunable Approximate Adders

Edge devices used in healthcare require high energy efficiency due to battery operation and replacement challenges. Approximate computing is effective in reducing energy consumption for applications tolerant to errors. Specifically, we investigate the application of approximate addition in the Pan-Tompkins algorithm for ECG processing. Our initial findings suggest that certain processes within the algorithm, such as high-pass filtering and derivative, are well-suited for approximation. We also observed that low-pass filtering is not as amenable to approximation. One noteworthy observation was that the handling of negative numbers through approximate addition did not introduce any adverse effects, reinforcing the viability of using this approach in the context of ECG processing for edge devices in healthcare applications. Our preliminary experiments demonstrate that a 16.5% reduction in power during additions is achieved.