Denoising of single‐trial event‐related potentials using adaptive modelling

Abstract

In this study, the authors present a modelling method based on the adaptive linear combiner to denoise single-trial event-related potentials. The orthonormal Hermite basis functions act as inputs of the adaptive linear combiner. To estimate and to adjust the parameters of the adaptive filter, the authors use the variable step-size least mean square algorithm which is well suited to track rapid changes of non-stationary signals. The performance of the method is tested with simulated evoked potentials and with real visual event-related potentials. For simulated data, the adaptive Hermite model gave significant enhancement in latency and amplitude estimation as well as in the observation of single-trial event-related potentials, in comparison with wavelet techniques and with other models of adaptive filters. For the real data, the proposed method filters the ongoing electroencephalogram activity, thus allowing a better identification of single-trial visual event-related potentials. The results confirm that the Hermite adaptive linear combiner model provides a simple and fast tool that helps to study single-trial event-related potential responses.

Wavelets statistical denoising (WaSDe): individual evoked potential extraction by multi‐resolution wavelets decomposition and bootstrap

“…Fig. 3 shows the contour plots of noisy signals, noise‐free signals and the outputs of the WaSDe, Hermite [30], NZT [15] and Wang [13] methods. We can observe that the signals denoised by using the WaSDe method are similar to the noise‐free signals and clearly recognisable in the contour plots unlike the other methods.…”