Adaptive discrete ECG representation - comparing variable depth decimation and continuous non-uniform sampling

This paper compares two methods of non-uniform ECG sampling: the variable depth decimation (VDD) and the continuous non-uniform sampling (CNU). The VDD algorithm uses the wavelet-based time-scale decomposition of the segmented ECG in which the high frequency scales representation is eliminated for the signal sections of narrower bandwidth (e.g. T-P segment). In result, the signal is locally decimated down to the level depending on the expected bandwidth. The CNU algorithm uses a soft estimation of the length for each subsequent sampling interval on a basis of expected local bandwidth of the signal. For ECG records from the CSE Multilead Database the average efficiency of the VDD algorithm is significantly higher (4.26) than the efficiency computed for the CNU method (3.01). Unfortunately, the global reconstruction error (PRD) is also higher for the VDD (0.40%) than for the CNU algorithm (0.22%).