A Symmetric Orthogonal FastICA Algorithm and Applications in EEG

Extracting a number of independent components, FastICA algorithm can ensure that, each component extracted has never been extracted by adding orthogonalization steps. However, Gram-Schmidt orthogonal method implemented itself that estimation error of the first vector has been accumulated in the subsequent vectors. In this paper, the square root of the classical matrix method is used to achieve symmetric orthogonal that parallel to estimate the source of variables and avoid the Gram-Schmidt orthogonal method of estimation error accumulation problem. The symmetric orthogonal FastICA algorithm is apply in removal EEG artifacts (including eye movement, blinking, ECG, EMG, etc). The experiment and simulation results demonstrate that the symmetry orthogonal FastICA algorithm removes the artifacts and has a better separation efficiency and fast convergence.