Adaptive neural network filter for visual evoked potential estimation

Abstract

The authors describe a new approach to enhance the signal-to-noise-ratio (SNR) of visual evoked potential (VEP) based on an adaptive neural network filter. Neural networks are usually used in an nonadaptive way. The weights in the neural network are adjusted during training but remain constant in actual use. Here, the authors use an adaptive neural network filter with adaptation capabilities similar to those of the traditional linear adaptive filter and suitable training scheme is also examined. In contrast with linear adaptive filters, adaptive neural network filters possess nonlinear characteristics which can better match the nonlinear behaviour of evoked potential signals. Simulations employing VEP signals obtained experimentally confirm the superior performance of the adaptive neural network filter against traditional linear adaptive filter.