Blind source separation for extraction of maternal and fetal ECG

Electrocardiography is one of the most important tools to interpret the heart's electrical activity. It is externally recorded over skin electrodes and is studied over a period of time. It provides a lot of important clinical data that can be used to analyze the performance of the heart and hence ensure the health of a person. It is a non-invasive technique and is recorded by using an electrocardiograph device. The application discussed in this work of thesis is the study of Fetal ECG. It is desired to extract the fetal ECG that is most commonly corrupted by the maternal ECG. In order to ensure fetal health, it is important that we monitor the Fetus' heart beat while in the womb. But since it is a very feeble signal, it is usually submerged in all kinds of physiological and electrical noises. Thus, it is necessary to perform proper signal processing techniques in order to extract a clean Fetal ECG signal.;Blind source separation finds an important application in this field. ICA is a very important tool to separate unknown signals from observed matrices. The main objective of the thesis is to extract the Fetal ECG from a mixture of Maternal and Fetal ECG using the FastICA algorithm. This has to be performed as accurately as possible in order to preserve the clinical data in the signal. These signals are tested to make sure that they satisfy the criteria required to perform ICA. These include linear mixture of signals, Independence of signals, non-gaussianity, etc. Various signal processing techniques like Wavelet Transform, Cubic Spline Filtering, and Adaptive Filtering are employed over the ECG signals to obtain the cleansed signals.;Finally, Adaptive Noise Cancellation is performed to compare the performance and quality of output signals with FastICA. Numerous simulations are performed to understand properties and convergence of FastICA.