Secondary data support in space-time adaptive processing

One of the primary problems with the application of space-time adaptive processing (STAP) techniques to radar is secondary data support for the interference plus noise covariance matrix estimate. Reed (1974) has shown the required secondary data support to achieve performance within 3 dB of optimal SINR is approximately twice the degrees of freedom (DOF). Reed proved this rule for sample matrix inversion (SMI) techniques. A concern arises when applying this rule to a newer class of reduced dimension STAP algorithms that do not fall under the SMI umbrella. This paper focuses on the cross spectral metric (CSM) algorithm (Goldstein and Reed, 1997). Through Monte Carlo simulations, Reed's rule for sample support is examined for this non-SMI technique. Optimum SINR performance for the CSM algorithm is obtained by choosing the number of DOF in the algorithm equal to the interference subspace dimension. With this choice, the required sample support for the covariance matrix estimate is approximately 2.5 times the interference subspace dimension. This relationship is not consistent.