Semantic ECG Communication for Personalized Arrhythmia Monitoring Over Human Body Communication Links

Abstract

This paper presents a semantic ECG transmission framework for personalized arrhythmia monitoring over human body communication (HBC) links. ECG beats are compressed into low-dimensional latent features using a deep autoencoder and transmitted through an HBC-based wireless link. At the receiver, arrhythmia detection is performed using either ECG reconstruction or direct latent feature classification using a support vector machine (SVM). Simulation results based on the MIT-BIH Arrhythmia Database show that latent feature classification is substantially more robust against communication errors than reconstruction-based detection. A latent dimension of 32 achieves a favorable trade-off between communication efficiency and diagnostic performance, while comparison with a conventional 2.4-GHz GFSK link demonstrates the advantage of HBC under identical transmit power conditions. The proposed framework provides an effective semantic communication solution for low-power wearable arrhythmia monitoring.