Covariance matrix estimator performance in non-Gaussian clutter processes

This paper considers the sample covariance matrix estimator in spatially non-Gaussian airborne radar clutter modelled as a spherically invariant random process (SIRP). Analytic expressions are derived for the estimator variance. They reveal the variance increase for non-Gaussian clutter as well as the sample support size required to reduce the variance to that of the Gaussian case. Specific consideration is given to the special cases of Weibull and K-distributed processes. Finally, estimation of a signal with unknown constant amplitude in K-distributed white noise is considered.