Signal-Aware Diagnostic Intelligence: Edge-Centric Signal Processing for Home-Based IoT Health Monitoring

Abstract

With the growing demand for decentralized and patient-centric healthcare, Diagnostic Internet of Things (D-IoT) systems have emerged as a promising solution for continuous health monitoring in home environments. However, existing architectures often rely on cloud-based processing, which introduces latency, power inefficiencies, and privacy concerns, particularly in real-time diagnostic scenarios. This paper proposes a novel edge-centric signal processing framework designed for ultra-low-power, on-device physiological monitoring using wearable IoT devices. The framework integrates real-time denoising using wavelet transforms and adaptive filtering, hybrid feature extraction across time, frequency, and nonlinear domains, and dimensionality reduction via Principal Component Analysis (PCA). Lightweight AI models, including 1D-CNNs and TinyLSTM, are deployed directly on microcontroller-class hardware, enabling accurate anomaly detection with <50 ms latency and <20 mW power usage. The system was evaluated across four benchmark datasets—MIT-BIH Arrhythmia, Sleep-EDF, PPG-DaLiA, and AudioSet—demonstrating high diagnostic accuracy, robustness under noisy conditions, and operational feasibility on embedded platforms. Compared to cloud-dependent solutions, this edge-centric approach ensures real-time responsiveness, data privacy, and long-term battery efficiency. The results validate the viability of performing clinically meaningful diagnostics directly on wearable devices, marking a critical advancement toward intelligent, autonomous, and accessible home-based healthcare.