Optimal Filtering of Single Channel EEG Data using Linear Filters

Electroencephalogram (EEG) is a flexible diagnostic tool used to study the underlying neural activities of the brain. However, these EEG datasets are often contaminated by artefacts that reduce the quality of these acquired bio-signals and are difficult to remove and cause misinterpretation of the underlying neural information. Moreover, no universally applicable denoising technique exists nor is there any set standard for comparison of denoising methods. A method to compare denoising methods based on Signal to Noise Ratio (SNR) and Cross-Correlation analysis using an ideally filtered signal was proposed in this paper. The ideally filtered signal was achieved using Blind Source Separation (BSS) technique. BSS techniques are computationally intensive, require greater number of EEG channels and require manual supervision for appreciable performance. Thus, in order to achieve comparable denoising performance on single-channel EEG data simple universally applicable linear filters were used. Filter order optimization was performed for various linear filters for maximal SNR performance. Signals filtered using linear filters were compared to BSS filtered signal using Cross-Correlation Analysis. It is seen that for most filters, their performances correlates to that of the optimal signal achieved by BSS, at their least orders. hereby overcoming the limitations of BSS.