Operating the Cross Spectral Metric Algorithm with Limited Secondary Data Support

摘要

One of the primary problems with the application of Space-Time Adaptive Processing (STAP) techniques is secondary data support for the covariance matrix estimate. A newer class of reduced dimension STAP algorithms uses a decomposition of the sample covariance matrix, thereby deviating from the SMI algorithm class. This report focuses on the performance of the Cross Spectral Metric (CSM) Algorithm with varying secondary data support sizes. The algorithm is shown to be highly susceptible to poor estimation of eigenvalues in the noise subspace. This susceptibility is manifested through large drops in output SINR with secondary data set sizes near the space-time product. To determine the cause of these performance drops, the CSM algorithm is recast in the structure of the eigenbeam model for SMI techniques presented by Gabriel. This new form of the CSM weight vector illustrates that eigenvalue spread in the noise subspace, a result of insufficient sample support, has a direct negative impact on the overall weight vector. Furthermore, eigenvalue spread is mitigated through the use of a diagonally loaded sample covariance matrix. Reducing the eigenvalue spread reduces the impact of cross spectral eigenbeams lying in the noise subspace. Monte Carlo simulations show the SINR performance drop mentioned previously is alleviated through the use of an appropriate diagonal load factor.