Knowledge-Aided Bayesian Space-Time Adaptive Processing

Ground moving target indicator (GMTI) radar processing attempts to distinguish between radar returns emanating from moving targets and stationary ground clutter. The task is confounded by the relative motion between the radar platform and the scene, as well as by the strength of clutter returns. Techniques such as space-time adaptive processing require an unknown interference covariance describing clutter, jammers, and thermal noise. The covariance is estimated from training data not under test, but, heterogeneous, contaminated, or limited training data degrade the covariance estimate and reduce the detection performance. State-of-the-art techniques for interference covariance estimation reduce the required amount of training data by imposing assumed structure on the covariance matrix. Here, a Bayesian signal model is adopted for jointly estimating targets and clutter in a single cell under test, allowing GMTI processing without training data. The approach incorporates the knowledge of an approximate digital elevation map, platform kinematics (velocity, crab angle, and antenna spacings), and the belief that moving targets are sparse in the scene. Low-complexity computation with the Bayesian model is enabled by recent algorithm developments for fast inference on linear mixing models. Results from the KASSPER I dataset show improved detection performance compared to existing techniques using scores or even hundreds of training bins.