Outlier Rejection Methods for Robust Kalman Filtering

In this paper we discuss efficient methods of the state estimation which are robust against unknown outlier measurements. Unlike existing Kalman filters, we relax the Gaussian assumption of noises to allow sparse outliers. By doing so spikes in channels, sensor failures, or intentional jamming can be effectively avoided in practical applications. Two approaches are suggested: median absolute deviation (MAD) and L1-norm regularized least squares (L1-LS). Through a numerical example two methods are tested and compared.