Compressive Sensing for Secure and Energy Efficient ECG Signals in Healthcare Applications

Abstract

In recent years, there has been a significant focus on advancing continuous health monitoring solutions through wireless technologies, particularly wearable devices, which facilitate the remote analysis of physiological data. While these innovations have revolutionized healthcare, they also present substantial challenges, including the transmission of large data volumes within the constraints of devices with limited battery life. Addressing these challenges requires the development of energy-efficient, secure, and diagnostically accurate telemonitoring systems, incorporating strategies to reduce data transmission while supporting real-time analysis. This thesis explores the integration of compressive sensing and compressed learning frameworks to address these challenges, with a specific focus on electrocardiogram (ECG) signal processing. By leveraging the inherent sparsity of biomedical signals, this research develops innovative methodologies to achieve secure data transmission, efficient signal reconstruction, and accurate diagnosis. The proposed frameworks balance theoretical advancements with practical implementations, bridging gaps in wearable healthcare technology and remote monitoring systems. Key contributions include the design of a dual-layer cryptographic framework for secure and energy-efficient ECG data transmission, validated on low-power hardware platforms. The thesis also introduces a data-driven sensing matrix and overcomplete dictionaries tailored to ECG signals, enabling high compression ratios without compromising reconstruction quality. Further, it integrates deep learning into the CS framework, proposing autoencoder-based models for efficient sensing and reconstruction. These models dynamically adapt to data, achieving superior performance in accuracy and computational efficiency with respect to the state-of-the-art methods. Extending beyond traditional reconstruction, the research develops a compressed learning framework that directly classifies ECG signals in the compressed domain, bypassing the reconstruction step. This approach significantly reduces computational overhead while maintaining diagnostic accuracy, with hardware validation demonstrating its feasibility for real-world applications. The findings highlight the transformative potential of compressive sensing and compressed learning in enabling scalable, secure, and energy-efficient healthcare solutions. This thesis sets a robust foundation for next-generation wearable technologies and remote monitoring systems, contributing to improved healthcare accessibility and diagnostic reliability. Future research directions include extending these methodologies to other physiological signals, optimizing hardware implementations, and exploring multimodal health monitoring systems.