Application of Riemannian mean of covariance matrices to space-time adaptive processing

The application of space-time adaptive processing to airborne GMTI radar data requires estimation of the covariance matrix. Among many novel results in the radar signal processing context, it has been shown that the usual sample covariance matrix based STAP approaches are suboptimal in that they are in the Euclidean space, rather than a natural Riemannian space of symmetric cones. In this paper, the algorithms of Riemannian mean based on the Karcher barycenter is shown to provide dramatically improved performance over the classic sample matrix inversion (SMI) technique.