FECG EXTRACTION USING VARIOUS LMS ALGORITHMS

Abstract

During pregnancy, it is very important to know the foetal development condition so that if there is any problem in development of foetal, it can be treated before creation of any critical condition. Foetal development & health status can be acknowledged through various methods such as ultrasound etc., but foetal ECG plays an important role in providing important information about the health status of the baby during labor condition. Doctors always perform foetal ECG extraction during labor to know if any disease is developing in the foetal & if there is any problem, then they try to diagnose the problem accordingly. Foetal ECG extraction is the process of separating the baby’s heartbeat signal from mother’s heartbeat signal. Various methods have been developed for the extraction of the foetal ECG signal such as Principal Component Analysis (PCA), Blind Source Separation (Independent Component Analysis (ICA)), Wavelet method etc. Adaptive filtering is one the most popular method used for the separation of foetal ECG signal. Adaptive filtering is the method which generates an error signal corresponding to the desired output signal. Adaptive filters are based upon adaptive algorithm. Adaptive algorithms are designed in such a way that it always tries to minimize the amplitude of error signal by changing the filter coefficient values in an iterative manner. Least Mean Square is the standard adaptive algorithm which tries to minimize its cost function value. The cost function for LMS algorithm is the square of the difference

v | P a g e

between the desired signal & the obtained output signal. In this project, we are taking pure foetal ECG signal which is our desired signal & the obtained foetal ECG signal. In this thesis, some improvements have been implemented in standard LMS algorithm. L1 norm penalty has been applied on the LMS cost function to generate a new algorithm named as Zero Attracting Least Mean Square (ZALMS). As the name suggests “Zero Attracting”, this algorithm tries to make the weights of the filter equal to zero as much as possible. Higher the number of the coefficients equal to zero, higher is the sparsity of the system & this higher sparsity helps in decreasing the error signal results in increasing the performance rate. Also an adaptive algorithm named as Normalized Least Mean Square (NLMS) is implemented for the extraction of the foetal ECG signal. Both ZALMS & NLMS provide better results in terms of signal to noise ratio and convergence speed in comparison to standard LMS algorithm.