Denoising of EEG Signals in BrainComputer Interfaces Using Extended Kalman-Filtered Recurrent High-Order Neural Networks

Abstract

This paper presents a Recurrent High-Order Neural Network (RHONN) for EEG denoising in brain-computer interface (BCI) applications. Trained online using the Extended Kalman Filter (EKF), the proposed approach effectively suppresses EOG, EMG, and ECG artifacts under non-stationary conditions while preserving EEG temporal structure. Experimental results show that RHONN achieves performance comparable to or better than conventional filters and deep learning models, including MLPs, RNNs, GANs, and autoencoders. A key advantage of the RHONN--EKF framework is its very low computational cost. By modeling each EEG channel with a single high-order neuron, the method reduces computational load by more than 90% compared to state-of-the-art models, making it suitable for real-time, resource-constrained BCI systems and consistent with Green AI principles.