Adaptive Optimal Dictionary Construction Scheme for Multi-Scale Joint Compression and Recovery of MECG Signals in WBAN

Abstract

In tele-cardiac monitoring systems deploying Compressive Sensing (CS) in Wireless Body Area Network (WBAN) needs to ensure true representation of ECG data for accurate prognosis and diagnosis. In this context data preparation strategy plays a critical role prior to sub-sampling phase in CS algorithms at codecs for energy efficient, least distortion based signal compression and recovery. Most of the previous designed Adaptive CS algorithms used single dictionary only for the full length segment of Multi-channel ECG (MECG) signals. Objective of the present work is to perform simultaneous compression of  MECG signals and recovery with least  distortion .This MMV (Multiple Measurement Vector) problem was solved by applying the proposed low distortion based Adaptive Dictionary (AD) construction technique to obtain clinically acceptable high fidelity data within the CS frame work. An optimal dictionary is generated for each block of ECG data of each channel concurrently. The results illustrate that the proposed adaptive best mother wavelet-based optimal dictionary (ABMOD) construction algorithm achieved least average PRD (Percentage root mean square difference) and WEDD (Wavelet energy based diagnostic distortion) values of 4.03 and 3.59 compared to existing 6.60 and 6.07 for PTB database with 70 compressed measurements and PRD1 of 0.475 against 1.13 for MIT-BIH ECG database with 192 measurements respectively relative to that of landmark Sub-band weighted mixed norm minimization (SWMNM) algorithm. The CR was 90.06 % or 7.14 against 6.34 of SWMNM algorithm which clearly indicates the relatively higher compression ability of proposed ABMOD algorithm and subsequent energy preservation This demonstrates the superior performance of the proposed ABMOD algorithm in terms of Joint compression and reconstruction for MECG signals. Further same studies were performed  with using  pan-tompkin’s algorithm for filtering purposes and results reveal that SNR was enhanced for the proposed algorithm for both MECG data bases.