Universal nonuniform sampling of ECG signals: Opportunities and obstacles

Abstract

Sampling of ECG signals has been studied for several decades. However, nonuniform sampling at low average rate has received much less attention. While uniform sampling rate needs to be at least twice the bandwidth of a signal (Nyquist rate), in theory, nonuniform sampling at sub-Nyquist rate should be adequate for faithful reconstruction if the signal is sufficiently sparse. Noting the known sparsity ECG signals, we attempt to demonstrate sub-Nyquist sampling of such signals. However, alongside this opportunity, reconstruction from nonuniform samples also poses a particular challenge. While uniform sampling admits universal (signal-independent) reconstruction (via sinc interpolation), reconstruction from nonuniform samples is generally data-dependent. In this paper, we make use of compressive sampling theory to demonstrate that nonuniform sampling does provide universal performance guarantee. In our experiments, we use ANSI/AAMI standard test ECG signals, and tune the orthogonal mathing pursuit algorithm for those to obtain superior performance. Our results potentially facilitates development of low-power ECG recording devices with potential application in remote locations without grid power.