Novel parameter tuned methodology for under-damped stochastic resonance applied to EEG signal enhancement

Abstract

In this article, motor imagery Electro-Encephalographic (EEG) signals for Brain-Computer interfaces (BCI) are processed under a weak signal detection (WSD) paradigm, with the goal to improve the low EEG Signal to Noise Ratio (SNR). Based on our previous results, Stochastic resonance (SR) is proposed, as a WSD method on EEG signals, for the first time. This novel methodology takes advantage of noise transitions on a non-linear, bi-stable, double well system. These transitions are synchronized with brain oscillations embedded in the EEG signal, achieving the desired brain waves enhancement. More over, our methodology modifies the bistable well separation and depth modulation on the Duffing system, having as target point the maximum SNR hence, recovering most of the input waveform shape. Dataset from the international BCI competition IV was used in this research, with an estimated SNR of −24dB. After stochastic resonance processing, the resulting signal enhancement on the desired μ − β band was observed, with an output SNR of 1.31dB during motor imagery and 1.41dB on resting states. In order to demonstrate the signal processing methodology applicability, a single EEG trial is reported from 17 channels processed. Also, for comparative purposes, dyadic wavelet transform (DWT) was applied to the same EEG signal. The results exhibit comparable, or even a better, signal enhancement with SR methodology than DWT processing.