Independent Component Analysis

ICA is a computational method for separating a multivariate (multi-channel) signal into independent additive subcomponents. ICA separation of mixed signals give very good results, providing two assumptions are met: Version implemented in this GUI is based on FastICA by Aapo Hyvärinen.

Computing ICA

In the configuration dialog, you can choose After selecting “OK” computation starts and, in a while, a new (temporary) signal will appear, consisting of ICA components calculated from the input signal. The number of components will be the same as the number of selected channels. Due to the randomization in FastICA implementation, the components will be presented in random order.

Analysing components

After the computation is finished, you can use “Describe components” to preview the spatial structure of the ICA components. After selecting the component from the displayed list, its topography will be displayed: electrodes' color will correspond to the associations between signal channels and ICA components. Blue colour means the channel adds to the coefficient with the positive sign, red colour denote the negative sign. Absolute value of the coefficient is represented by intensity of the colour.

Zeroing components

By using “Zero selected components” method, it is possible to perform an inverse transform, representing signal channels as a superposition of ICA components. In addition, selected components may be suppressed in this process. In this way, one can remove artifacts and/or noise, if they're represented as separate ICA components.

Literature

  1. Comon P.: Independent component analysis, A new concept?, Signal Processing 36 (1994), pp 287-314
  2. Hyvärinen A.: Fast and Robust Fixed-Point Algorithms for Independent Components Analysis, IEEE Transactions on Neural Networks 10(3) 1999, pp 626-634