Compressed sensing implementation in cardiac signals

Abstract

This work presents the study of a compressed sensing implementation in a system which acquires three cardiac signals, ballistocardiogram, electrocardiogram and photoplethysmogram. In order to accurately estimate heart rate and its variability, cardiac signals must be acquired at frequencies of about 1 kHz, but since these signals have a sparse representation in some transformation basis, namely in the wavelet domain, compressed sensing paradigm states that they can be recovered from a small number of projections in another basis incoherent with the first. The signals' compressibility was assessed, then TwIST algorithm was applied and reconstruction quality was measured for a number of different signal-to-noise ratios, compression rates, and sparsity basis. The analysis was completed by evaluating the algorithm's computation time and heart rate deviation of the reconstructed signals.