Eye Blink Artefact Cancellation in EEG Signal Using Sign-Based Nonlinear Adaptive Filtering Techniques

Abstract

This paper presents the filtering of the noise from the Electroencephalogram (EEG) using an adaptive filtering approach known as Nonlinear LMS algorithm. The noise presence in the signals makes it difficult to analyze the EEG as well as to get the correct representation of the signal. It, therefore, becomes important to design the signal filters to minimize the noise in such EEG signals. Nonlinear adaptive filter is implemented to minimize the ocular artefact/eye blink from the EEG in which the parameters can be adjusted to maintain the input–output relationship. Further, within Nonlinear adaptive filter approach, different sign-based versions such as Nonlinear sign Least Mean Square (NSSLMS), Nonlinear sign Least Mean Square (NSLMS), Nonlinear sign regressor Least Mean Square (NSRLMS) are developed to minimize the eye blink noise along with reducing the computational complexity. After this, a comparison of the filtered output signal is done with the output from the conventional LMS filter. Based on the parameters such as signal-to-noise ratio (SNR), misadjustment ratio (Madj) and excess mean square error (EMSE), we noticed that the filtered output from the Nonlinear adaptive filter is considerably more efficient. The main benefit of preferring NLMS in our analysis is its superior computational speed with multiplier free weight update loops. Also, it eliminates the need to make assumptions regarding data distribution and its size. Finally, to test the performance of the proposed algorithm, the same is applied on EEG signals extracted from CHB-MIT database. On comparing the results of the proposed algorithms with the conventional LMS, we noticed that NSRLMS outperforms the current realizations in reducing the noise.