Improving space-time adaptive processing (STAP) radar performance in nonhomogeneous clutter

Date of Award


Degree Type


Degree Name

Doctor of Philosophy (PhD)


Electrical Engineering and Computer Science


Hong Wang


signal detection, space-time adaptive processing (STAP)

Subject Categories

Electrical and Computer Engineering


We are primarily interested in radar signal detection, using STAP technique, in a nonhomogeneous noise background which has unknown covariance information. We should know that nonhomogeneous data, once joined to the covariance matrix estimation, will cause the degradation of STAP performance. To this end, the purpose of this dissertation is to find solutions to reduce the STAP performance degradation caused by the nonhomogeneous data.

We first discuss what nonhomogeneity is and its effects on STAP. Nonhomogeneity will cause the SCNR loss via the filtering, and the CFAR loss via the estimation of the threshold. These two losses are derived from a bad covariance matrix estimation because the weighting vector of STAP is ${\bf w}=u{\bf\ R}{\bf{\sp{-1}s}}.$ Thus, the covariance matrix estimation plays a decisive role regarding the reduction of the STAP performance degradation.

We introduce two methods, sample selection and data weighting, to handle nonhomogeneous data. Sample selection is a pre-STAP data processor in which we aim at screening the nonhomogeneous data before forming the covariance matrix. In other words, we choose only the likely homogeneous data to gain a better covariance matrix estimation. Definitely, sample selection is appropriate for discrete type nonhomogeneity.

The covariance matrix estimation via the maximum likelihood estimate (MLE) results in an equal weighting of all sample data, which is not an especially effective approach to control non i.i.d. nonhomogeneous data. Rather, we suggest a weighted average covariance matrix estimation, in which we weight the likely nonhomogeneous data with a smaller weighting than that of the likely homogeneous data. We thereby show that both the SCNR and the CFAR losses, under the data weighting situation, can be reduced.

Moreover, we must test all secondary data, and thus know their characteristics, before we can apply either a sample selection or a data weighting, or a combination of both to the data set. Indeed, choosing a proper test algorithm for nonhomogeneity detection is critical. Consequently, we also include a comparison of three CFAR embedded algorithms, GLR, MSMI and $T\sp2,$ in this dissertation.


Surface provides description only. Full text is available to ProQuest subscribers. Ask your Librarian for assistance.