A Low-Cost Discrete Wavelet Transform Framework for ECG Signal Denoising on Resource-Constrained Embedded Platforms

Electrocardiogram (ECG) signals acquired in semi-rural primary health centers using low-cost AD8232 front-ends are routinely corrupted by power-line interference (50 Hz), baseline wander caused by electrode motion, and broadband electromyographic noise. This paper presents a lightweight Discrete Wavelet Transform (DWT) denoising pipeline benchmarked on 30 records of the MIT-BIH Arrhythmia Database (fs = 360 Hz). Seven wavelet families are compared at decomposition levels L = 2–7 using Donoho's universal soft-threshold rule with a robust median absolute deviation estimator. The proposed db6 wavelet at L = 5 yields an output SNR of 7.46 dB from a 1.17 dB input — an improvement of 6.29 dB — while preserving the QRS morphology with only 0.0019 mean-square error. The algorithm has a complexity of O(N) and executes in 11.3 ms on an ESP32 (240 MHz), making it deployable on rural tele-cardiology kiosks costing under ₹2,500.