A Kronecker product improvement to PCA for space time adaptive processing

Space time adaptive processing (STAP) is computationally demanding due to the large dimensions of the space-time covariance matrix. Covariance estimation is problematic for these dimensions, because a sufficient sample size is never available due to nonstationarity. One common method of addressing this issue is through principal components, in which only the principal interference subspace is retained. For problems arising in STAP, an additional structure is suggested; that the covariance has a dominant low-rank subspace with space-time separable residual. We apply a least square Kronecker fit to this residual covariance. Our results using the ONR UESA circular array data show that this considerably improves the performance, most notably when the sample support and reduced rank are small.