Development of SA-LightCS for Lightweight ECG Signal Reconstruction Using Compressive Sensing

ECG monitoring systems generate extensive data, challenging storage and transmission in resource-constrained environments. While compressed sensing offers data reduction potential, existing approaches struggle to balance computational efficiency with signal fidelity. Our SA-LightCS framework addresses these limitations through three innovations: depthwise separable convolutions reducing parameters by 63.5%, parallel residual structures preserving multi-scale features, and self-attention mechanisms capturing global signal dependencies. Tested on MIT-BIH datasets, our model outperforms conventional methods (CoSaMP, SOMP) and recent deep learning approaches (CAE, CSNet), achieving 12.9% lower PRD and 1.99 dB higher SNR at 5% compression while requiring significantly fewer computational resources. SA-LightCS provides an effective solution for real-time ECG monitoring in resource-limited healthcare applications.