Burg estimation of radar covariance matrix for mixtures of Gaussian stationary distributions

We propose three estimators for the covariance matrix of a continuous mixture of Gaussian stationary autoregressive vectors. For Gaussian autoregressive processes, Burg methods are often used in case of stationarity for their efficiency when few samples are available. Unfortunately, if we directly apply these methods to estimate the common covariance matrix of N vectors coming from a non-Gaussian distribution, the efficiency will strongly decrease. We propose then to adapt these methods to mixtures of Gaussian vectors by changing the energy functional to minimize in the Burg algorithm.