Edge AI-Based Real-Time Physiological Signal Processing for Wearable Health Monitoring

Abstract

Wearable health monitoring has moved from episodic measurement toward continuous, longitudinal observation of physiological state, generating new requirements for real-time signal processing, privacy preservation, and system-level governance. Edge artificial intelligence has emerged as a promising architectural strategy for processing electrocardiographic, photoplethysmographic, electrodermal, and inertial signals near the body rather than relying exclusively on centralized cloud infrastructure. This paper presents a system-oriented analysis of edge AI for wearable physiological monitoring, emphasizing architectural trade-offs, resource governance, robustness, fairness, deployment sustainability, and policy implications. Instead of treating edge AI as a narrow algorithmic advance, the paper examines it as a distributed sociotechnical infrastructure in which sensing, inference, communication, and clinical decision support must be jointly governed. The discussion addresses the layered structure of wearable edge systems, the challenge of motion artifact resilience in ambulatory settings, and the tension between local model autonomy and centralized learning. It further considers federated learning as a governance mechanism, the regulatory complexity of automated health inference, and the clinical integration difficulties that arise when models trained in controlled settings are exposed to heterogeneous real-world conditions. The paper argues that sustainable and equitable wearable edge-AI systems require adaptive model management, transparent subgroup evaluation, staged clinical validation, and clear accountability across device manufacturers, software providers, health systems, and regulators. Future research should focus on energy-aware inference, model drift monitoring, cross-device interoperability, and governance frameworks that account for the continuous and intimate nature of wearable physiological data.