OCULAR ARTIFACTS ELIMINATION AND FEATURE EXTRACTION IN MOTOR IMAGERY-BASED BCI USING NONLINEAR ADAPTIVE FILTER

Abstract

The paper proposes a novel methodology of de-noising raw electroencephalogram (EEG) data from ocular artifacts (OAs) and alpha waves extraction from motor imagery-based signals that could be further utilized for brain–computer interface (BCI)-based applications. An algorithm based on discrete wavelet transform (DWT) and nonlinear adaptive filtering for the removal of OA is advocated, with an aim of making the process computationally intelligent. This algorithm has been tested on pre-recorded EEG dataset for BCI (Dataset IIIa; obtained from the website of the BCI Competition III). To further validate the competence of the proposed method, synthetic EEG signals were created, which were fused with white Gaussian noise. A total of 20 EEG signals were generated, half of which had added noise with a signal-to-noise ratio (SNR) of 10[Formula: see text]dB and other half had added noise of 5 dBSNR. Each signal contained 1000 samples with a sampling frequency of 250[Formula: see text]Hz. An optimum bandpass filter (FIR and IIR) for extraction of alpha waves has been suggested. FIR Equiripple filter is found most appropriate for the task as it has highest SNR and computes the response faster when compared with other filters. Among different mother wavelets, Daubechies 4 wavelet obtained using statistical thresholding denoises the EEG data most successfully. Correlation and root mean square error (RMSE) parameters show that the performance of nonlinear adaptive filter developed using nonlinear Volterra series has an edge over conventional adaptive filters for the intended purpose.