On the wavelet-based compressibility of continuous-time sampled ECG signal for e-health applications.

Abstract

This paper presents a compression study of electrocardiogram (ECG) signals for e-Health cardiac online diagnostic systems. The study uses 75 real electrocardiogram records sampled with continuous-time level-crossing (LC) analog-to-digital converter (ADC). This signal-dependent LC-ADC compresses signals compared to conventional ADC but further compression is needed especially for long-time monitoring applications. The orthogonal matching pursuit algorithm is simulated to evaluate ECG compression with 54 orthogonal and biorthogonal wavelets. For LC-ADC amplitude output compression, Biorthogonal3.1 (bior3.1) wavelet achieves optimal performances in terms of compression ratio (CR) while ensuring 2-% percentage root-mean-square difference (PRD). The PRD must be limited to this value to ensure a very good quality signals after decompression. For circuit implementation purposes, bior3.1 wavelet is proposed as a multiplier-free decomposition step and a noncomplex global and hard thresholding process is achieved. The average CR is 63% and PRD varies between 0.1 and 2.1% leading to a very good diagnostic quality.