Simulation Of Independent Component Analysis Algorithm On Computer And Application Of It To Multi-Channel EEG Signals Processing

The development of the technology of the computer and the information science has promoted the progress of the discipline of the Digital Signal Processing (DSP), and thus has produced a lot of new theories, technologies and algorithms. Independent Component Analysis (ICA) is such a new algorithm that came out with the development of Blind Source Separation in the past ten years. ICA has been applied in many fields, One of which is the application in the electroencephalogram (EEG) signals processing.At the beginning, the thesis summed up the current researches in areas of ICA and EEG signals processing, and then introduced the definition, standard mathematic model, hypothetical conditions and four estimation principles of ICA. The thesis also pointed out that ICA can recovery the source signals from the observed signals by certain learning rules.Three kinds of ICA algorithms, the ICA based on Kurtosis, the ICA based on Infomax and fast fixed point ICA (FastICA), were introduced in this article and realized through simulation on computer. The result of simulation showed that FastICA avoid the unconvergence problem which may appear in the methods of the ICA based on Kurtosis and the ICA based on Infomax for bad learning rate, because FastICA needn't choose the learning rate. Moreover this thesis verified the classical application of ICA is blind source separation in multi-channel signals.On the part of EEG signals acquisition, the thesis introduced the programmingframe and process of EEG signals acquisition through Non-NI analog-digital transform card, which were realized by the combination of LabVIEW, an excellent graphic programming language, and C language.On the part of EEG signals processing, FastICA algorithm was applied to the six-channel EEG signals, which contain 60Hz line noise and blink artifact. By adding two orthogonalized reference power signals to the observed signals that contain 6 channels EEG signals and 1 channel electrooculogram (EOG) signal, the problem of removing line noise and artifact was transformed to the problem of blind separation of multi-channel signals. When concerned with the map between the source signals and separated signals. Author tentatively used the matrix, which got by observed signals multiplied by the transform separated signals, to confirm the map between the observed signals and separated signals.