Efficient electrocardiogram signal coding for medical devices using linear prediction

Abstract

The growing adoption of wearable health monitoring and internet of things (IoT)-based medical devices has increased the demand for efficient electrocardiogram (ECG) compression techniques that reduce data size while preserving diagnostic fidelity. This paper proposes a compression method that combines linear predictive coding (LPC) with a residual thresholding mechanism. The ECG signal is modeled using a tenth-order LPC framework, and the residual is sparsified by retaining only the most significant components. The novelty of this work lies in applying threshold-based residual sparsification tailored to ECG signals, enabling high compression ratios (CR) while preserving clinically relevant waveform features. Experimental results using PhysioNet ECG data show that the proposed method achieves CRs of up to 38:1, while a moderate threshold (alpha is 0.005) provides a practical trade-off, achieving around 20:1 compression with preserved P wave, QRS complex, and T wave. The method offers a low-complexity solution suitable for wearable and IoT-based healthcare applications.