Wavelet Data-Driven Extreme Learning Machine Auto-Encoder based Compressive Sensing for Physiological Signal Reconstruction over Wireless Body Area Network
The exponentially rise in the demands of Internet-of-Things (IoTs) enabled Wireless Body Area Network (WBAN) and Personalized e-Health systems for which compressive sensing (CS) technique has played decisive role. CS being potential towards low redundant data communication and resource efficient transmission is of great significance; however, majority of classical CS methods don’t address non-linear sparsity problems in physiological signals that confines it to exhibit low reconstruction quality and high compression error. In this paper a novel wavelets data-driven Extreme Learning Machine Auto-Encoder (ELM-AE) based CS model is developed for multiple physiological signal reconstruction. The proposed CS model at first estimates different wavelets containing approximated coefficient and detailed coefficient values, where the first is learnt over the modified ELM-AE to obtain sparse representation of the input physiological-signal. Executing ELM-AE learning over the approximated coefficients, we performed threshold-adaptive optimal sparse feature generation, which was subsequently processed for Inverse-SWT in conjunction with the SWT-detailed coefficient to perform signal-reconstruction. Simulation over ECG and PPG signals revealed that the proposed CS model achieves better performance in terms of Percent Root Mean Square Difference (PRD), Signal to Noise Ratio (SNR), Compression Ratio and compression quality score (QS) for the different physiological signals.