LL Logo



Evaluation of STAP
Training Strategies with
Mountaintop Data

Daniel F. Marshall
MIT Lincoln Laboratory
244 Wood Street
Lexington, MA 02173-9108
email: dmars@ll.mit.edu

Abstract Most of the performance analysis to date of Space-Time Adaptive Processing (STAP) algorithms has been based on computer simulations. The Mountaintop database has been created to provide a source of actual field radar data which will allow more realistic assessments of STAP performance. But a major difference between STAP simulation studies and STAP studies based on real data is the necessity to implement robust covariance estimation (training) techniques in the latter case. Simulation studies may bypass the training process, or will typically involve a relatively well-behaved clutter model (e.g., sandpaper-earth). In contrast, spatially inhomogeneous real-world clutter presents a difficult covariance estimation problem. Thus, an examination of covariance training strategies is a necessary precursor to effective implementation and comparison of STAP algorithms with real data. In this presentation, Mountaintop data is used as a case study to evaluate several covariance training approaches. We will also examine related issues such as clutter rank and the available number of training samples. These results lend insight into the problem of covariance estimation in a real-world context and provide the necessary background for the study of STAP techniques based on real data.



LL Logo Disclaimer

Direct comments and questions to: webmaster@ll.mit.edu

MIT Lincoln Laboratory. All rights reserved.