Research On Spontaneous EEG Filter Based On ARMA Model

An event-related potential (ERP) is a special brain evoked potentials which is caused by one or more stimulus (voices, light, images etc.). It is a record to the skull surface. It has several components, but the signal is very weak and susceptible to noise interference. The frequency range of the human brain is covered by spontaneous potential all the time and the signal amplitude of spontaneous potential is 2-5 times to the event-related potentials. Usually, people need to go through the 40 to 100 times superposition average to extract event-related potentials, which requires the same number of repetitive stimulation experiments. Therefore, extracting ERPs from EEG from a small count or single EEG(s) becomes the important and difficult problem for EEG signal processing.This paper is based on three basic assumptions:First:Electroencephalograph (EEG) signal is a short-time stationary. Second:EEG is not white noise but colored noise. Third:The spontaneous EEG before stimulation and the EEG after the stimulation has the same characteristics, so the Auto-Regressive and Moving Average Model (ARMA) model which is established by the EEG before the stimulation can also fit the EEG after the stimulation.This paper uses the ARMA model to filter spontaneous EEG. The main work is listed as followed:(l)Contrast a variety of ARMA model order selection methods, and model checking method by experiments. Determine optimal portfolio for EEG ARMA model modeling. Firstly, use the autocorrelation and partial autocorrelation graph to determine the initially model type, and then use the exhaustive order determination method to determine the specific order of the model, and then the fast Levinson-Durbin recursion method and the Newton-Raphson method were used to calculate the regression coefficient and the sliding average coefficient. Finally using χ2 test method for model checking.(2)Use AR model for modeling the EEG data, and AR whitening filter is constructed by AR coefficients. The EEG segments after the stimulation is filtered by each filter which is constructed according to each coefficients. According to the Wold decomposition theorem, those EEG data which are not suitable for AR model are modeled by high-order AR model which is a approximated result.(3) Use ARMA model for modeling the EEG data. Those EEG data which are not suitable for AR model are well solved by ARMA model. Design the whitening filter that is based on ARMA filter to complete EEG whitening work.(4) Experimental contrast, the proposed method and the results of superposition averaging were compared. The advantages of ARMA model is described in the experiment according to the result of AR modeling and ARMA modeling. The experiment provides a new idea of using ARMA model for extracting event-related potential from EEG.