Extracting foetal electrocardiogram (fECG) plays an important role in diagnosing foetus’s health. However, in real clinical tests, a clean extraction of fECG is difficult to be obtained, because it is affected by various signals such as mother electrocardiogram (mECG), electromyogram (EMG) derived from the uterus and muscle contractions, the respiration signal, electronic noise, etc. Inspired by recent work estimating fECG subspace from a mixed ECG signals using multidimensional independent component analysis (MICA) along with the cyclic coherence (CC), we propose here an approach to separate and classificaty ECG signals recorded from abdominal and thoracic electrodes of pregnant women. The first step, blind source separation (BSS) is done by applying the joint approximate of Eigen matrices (JADE) algorithm to obtain independent components (ICs). Then continuous wavelet transform (CWT) is adopted for classifying the independent components previously obtained into three subspace components: fetal ECG signals, the mother ECG signals, and the noise. Our experimental results have corroborated the proposed approach using the database DaISy.