Single channel fetal ECG recovery using sparse redundant representations

A novel approach to fetal ECG signal recovery using a single abdominal record is proposed. Following transforming the original ECG waveform into a 2D array, we apply an image denoising procedure based on sparse representations over overcomplete dictionaries obtained by the K-SVD algorithm. A modified version of this algorithm using an L₁-norm fidelity term is shown to improve the performances over the classical L₂-norm approach. Experimental results indicate that the quality of the recovered fetal ECG signals is comparable to those employing multi-channel recordings and Independent Components Analysis or adaptive filtering algorithms.