Performance of Various ICA Algorithms for an Electrocardiogram Signal

Abstract

De-noising of electrocardiogram signals is of utmost importance to diagnose the cardiovascular diseases. Many techniques have been utilized for filtering and removing noise from the raw electrocardiogram signal, such as traditional Wiener methods, neural networks and wavelet decomposition methods. These methods may lead to reduction in the amplitude of the QRS complex. The recorded multichannel electrocardiogram signal is correlated and is difficult for analysis. A suitable technique to overcome these problems is the appropriate use of Independent Component Analysis to maximize the required statistical parameters to make the output to be correlated. This work relates the performance of three of the Independent Component Analysis Algorithms such as JADE (Joint Approximate Diagonalisation of Eigen Matrices), Fixed Point ICA (Fast ICA) and AMUSE (Algorithm for Multiple Unknown signals Extraction) by which the three channel signal was made uncorrelated so that the best signal among the three different channels can be identified for further processing based on the values of the Signal to Interference Ratio(SIR) obtained in each of the algorithm