Low-Complexity Compressed Analysis in Eigenspace with Limited Labeled Data for Real-Time Electrocardiography Telemonitoring

Abstract

To achieve real-time electrocardiography (ECG) telemonitoring, one of the major obstacles to overcome is the scarce bandwidth. Compressed sensing (CS) has emerged as a promising technique to greatly compress the ECG signal with little computation. Furthermore, with edge-classification, the data rate can be reduced by transmitting abnormal ECG signals only. However, there are three main limitations: limited amount of labeled ECG data, tight battery constraint of edge devices and low response time requirement. Task-driven dictionary learning (TDDL) appears as an appropriate classifier to render low complexity and high generalization. Combining CS with TDDL directly (CA-N) will degrade classification and require higher complexity model. In this paper, we propose an eigenspace-aided compressed analysis (CA-E) integrating principal component analysis (PCA), CS and TDDL, sustaining not only light complexity but high performance under exiguous labeled ECG dataset. Simulation results show that CA-E reduces about 67% parameters, 76% training time, 87% inference time and has a smaller accuracy variance to the CA-N counterpart.