Adaptive Filtering for Interference Removal in FNIRS-Based BCl Using Empirical Wavelet Transform

Brain computer interface (BCI) is a communication device transmitting human brain activity signals to control external robots. Functional near-infrared spectroscopy (fNIRS), having advantages in excellent temporal resolution and greater portability, becomes a promising neuro-investigatory technique used for brain signal acquisition in BCI systems. However, instrumental and physiological interference induced by heartbeat and respiration adversely affect the controlling quality and accuracy of BCI. In our study, empirical wavelet transform (EWT) has been adopted for the interference suppression of fNIRS signals, and Monte Carlo simulation with five-layer human brain model are developed to evaluate the performance. The analysis results suggest that the series of components by EWT can effectively avoid the mode-mixing problem appeared in empirical mode decomposition (EMD). Therefore, the recovered hemodynamic response signals using EWT have better performance compared with low-pass filters and EMD with relatively low mean square error (MSE). By applying the EWT method to real fNIRS signals obtained through experiments and modify it to be better adaptable of fNIRS signal processing, it also demonstrates efficacy in removing contamination from functional brain response.