A comparison of three adaptive algorithms used to improve evoked potential estimation

The objective of this work is to compare three adaptive algorithms used in a proposed noise canceller with an enhanced reference signal, to improve the Evoked Potential (EP) estimation. EEG averaging is a common method to obtain EP. A disadvantage of this method is the large number of stimuli needed to get a good visualization of one useful average EP. We use a two channel adaptive canceller to obtain single stimulus evoked potentials. The canceller reference signal is the quasi-periodic extension of an average EP over a few periods, aligned by the maximization of its cross-correlation with the primary signal. The Least Mean Square (LMS), Carlos Davila (CDA) and Recursive Modified Gram-Schmidt (RMGS) algorithms were used to update filter weights. To test the different algorithms, the described technique was applied to simulated signals based on actual registers obtained from visually stimulated subjects. The proposed adaptive cancellation technique provides, with fewer stimuli, individual estimates, allowing the analysis of the EPs variability between stimuli. The fastest convergence rate and best estimate were obtained with the CDA and RMGS algorithms. CDA stands out for its simplicity.