Fetal ECG (FECG) extraction utilizing non-invasive, real-time maternal abdomen recordings is optimized in this work. Early pregnancy fetal signals are low-amplitude and typically obscured by the dominant maternal ECG (MECG), making extraction challenging. Traditional high-order statistics approaches like ICA and PCA are useful but require several abdominal channels. Clinically, few recordings are available, and the direct fetal signal is rare. Increased electrodes on the mother’s belly were formerly considered, but they might harm both mother and fetus. This study proposes a novel method that creates realistic synthetic signals simulating fetal activity without electrodes to circumvent these limits. Using deep learning, specifically Autoencoders, the proposed technique reconstructs fetal ECG signals with a 0.96 correlation coefficient to the observed signal. This breakthrough allows safe and effective FECG extraction utilizing a single belly signal, advancing biomedical signal processing and non-invasive prenatal surveillance.