Gain optimized cosine transform domain LMS algorithm for adaptive filtering of EEG

The most common adaptive filtering method is based on the least mean square (LMS) algorithm, which updates the filter coefficients by a gradient based method. The convergence properties of the LMS algorithm can be improved by updating the filter coefficients in the frequency domain. This work presents a new LMS algorithm, which updates the filter coefficients in the cosine transform domain. Instead of a constant gain factor in the coefficient updating the present method uses a time-varying optimized gain factor. This yields a considerably improved convergence performance. The algorithm was applied to the EEG activity analysis of freely behaving rats.