Functional link network with genetic algorithm for evoked potentials

In this study, a functional link network with genetic algorithm is employed for extracting evoked potentials. Evoked potentials are time-varying signals typically buried in relatively large noises of electroencephalogram. Recently, to extract evoked potentials more effectively from noises, adaptive filtering techniques are widely employed for evoked potentials. In general, least-mean-square algorithm is used to adapt filter weights. However, it is well-known that the selection of step-sizes is a trade-off on the convergence rate and steady state performance. In practice, an inappropriate step-size always causes deficiency. Therefore, the proposed method employs genetic algorithm to improve this issue. Genetic algorithm is basically a kind of evolutionary strategies. The step-size candidates are regularly generated and evaluated, and the fittest candidate is selected for subsequent adaptation. Results show that the proposed method is insensitive to the selection of step sizes and reference inputs. It is applicable for extracting EP in different noise levels.