Kalman filter for dynamic imaging based on complex empirical covariances

Kalman filter (KF) is a priori unsuitable for the estimation from sample covariance matrices as they cannot be formulated analytically as a function of state parameters to be estimated. In this work, we propose a novel KF adapted to sample covariance matrices under the unconditional signal model. It is evaluated on simulated data representative of a dynamic radio astronomy framework, considering multiple uncorrelated sources and Gaussian noise. The results show that our method is capable of effectively tracking moving sources in complex scenes with greater accuracy than a KF regularized in a standard way, i.e., without proper formalization of the noise model.