According to non-stationary characteristic of Electrocardiogram(ECG) signal, a new method combined non-stationary signal processing and nonlinear estimation was proposed to extract the Fetal ECG(FECG) signal from the abdominal composite signal of the pregnant woman. The non-stationary Maternal ECG(MECG) signal was firstly decomposed into finite Intrinsic Mode Functions(IMFs) and one residue utilizing Empirical Mode Decomposition(EMD). The MECG component in the abdominal composite signal was a nonlinear transformation of the MECG signal and Least Squares Support Vector Machine(LSSVM) was used to fitting the nonlinear transformation. An optimal estimation of the MECG component in the abdominal composite signal was obtained by the finite IMFs and one residue undergoing the nonlinear transformation. The FECG was extracted by removing the optimal estimation of the MECG component from the abdominal composite signal. To evaluate the performance of FECG extraction method, eigenvalue analysis and cross correlation was used to estimate the Signal-to-Noise(SNR) ratio for an ensemble of FECG waveforms. The experimental results show that the clear FECG can be extracted even under the condition of the fetal QRS wave being entirely overlapped with the maternal QRS wave in the abdominal composite signal and its SNR ratio was significantly increased compared with traditional methods. The experimental results verify the proposed method.
