Enhanced Sage-Husa Adaptive Filtering for Discrete-time Systems with Unknown Non-Gaussian Noise

This paper addresses the challenge of unknown prior knowledge of non-Gaussian noise in the Kalman filter. Despite the exceptional performance of the Kalman filter in Gaussian noise scenarios, it struggles when confronted with unknown non-Gaussian noise. To remedy this, we propose an enhanced Sage-Husa adaptive filter and vector autoregressive models for estimating the covariance of non-Gaussian noise and identifying the noise coefficient matrix in real time. The effectiveness of this method is being validated through simulations, proving its robust performance under various initial parameter settings and different noise covariance scenarios.