The Application of Independent Component Analysis in Removing the Noise of EEG Signal

In the process of collecting and processing ElectroEncephaloGrapgy signals, it is often interfered by various noises and artifacts such as Electrooculography and electrocardiogram. In order to remove these interferences, this paper applies Independent Component Analysis technology to the problem of EEG signal separation in the process of removing artifacts from the Brain computer interface. ICA can be executed by a variety of algorithms, from which we observe that there are four commonly used algorithms for brain signal separation. The four ICA algorithms are: Second Order Blind Identification (SOBI), Hyvarinen's Fixed Point Algorithm (FastICA), Infomax and Joint Approximation Diagonalization of Eigenmatrices (JADE). The ICA algorithm requires multiple iterations to obtain the separation matrix, which is computationally intensive and slow. Therefore, it is necessary to analyze which ICA algorithm is suitable for hardware when we need to consider design constraints and requirements. In this work, we use MATLAB to evaluate the four ICA algorithms based on some conditions (running time, allocated memory). The experimental results show that the MATLAB implementation of SOBI algorithm is the best among all analyzed ICA algorithms. Compared with the other three ICA algorithms, the SOBI algorithm can separate EEG signals more quickly and accurately.