In this paper, we propose a strategy to estimate a Fetal Electrocardiogram (FECG) subspace from a set of mixed ECG recordings from the thoracic and abdominal electrodes attached on a pregnant woman. The ECGs from an expectant woman contain FECG that can provide valuable information for fetal health monitoring, such as the fetal heart rate (FHR). After applying blind source separation (BSS) methods to mixed ECG, a number of independent components are obtained. The main purpose of this paper is to classify an FECG group from all of these components which can be classified as FECG, MECG and noise according to the features of signals. This work is inspired by the concept of multidimensional independent component analysis (MICA). In order to automate the classification task, we propose a procedure based on cyclostationarity of FECGs; in particular, we propose an integrated Cyclic Coherence as a criterion to classify FECG subspace automatically. The method is validated on real world DaISy dataset and the results are promising.