Filter Out High Frequency Noise in EEG Data Using The Method of Maximum Entropy

Abstract

We propose a maximum entropy (ME) based approach to smooth noise not only in data but also to noise amplified by second order derivative calculation of the data especially for electroencephalography (EEG) studies. The approach includes two steps, applying method of ME to generate a family of filters and minimizing noise variance after applying these filters on data selects the preferred one within the family. We examine performance of the ME filter through frequency and noise variance analysis and compare it with other well known filters developed in the EEG studies. The results show the ME filters to outperform others. Although we only demonstrate a filter design especially for second order derivative of EEG data, these studies still shed an informatic approach of systematically designing a filter for specific purposes.