Robust BSBL recovery method of physiological signals with application to fetal ECG

Abstract

Compressive Sensing (CS) techniques have emerged with the increasing demand of high data rate transmissions. Recently, block sparse Bayesian learning (BSBL) framework was introduced which has a superior performance over conventional CS methods. In this paper, the BSBL-Expectation Maximization (BSBL-EM) and BSBL-Bound Optimization (BSBL-BO) methods were deployed. The performance, mainly quality and speed, of recovering a block sparse signal was analyzed. Results showed that the two algorithms performance is almost the same in terms of NMSE. However, BSBL-BO achieved better efficiency since the required recovery time was less than BSBL-EM. To further investigate the algorithms performance, they were deployed to recover a real world FECG segment. They achieved a satisfactory quality where the distortion is negligible and does not affect the clinical diagnosis. Nevertheless, using BSBL-BO is more suitable for wireless tele-monitoring based systems since it is more efficient.