Learning-based space-time adaptive processing

Space-time adaptive processing (STAP) has been used for many years in moving target indicator (MTI) radars to solve the problem of target detection in presence of an interfering environment. Over the years, different versions of STAP have been introduced to enhance its performance and overcome its practical difficulties. In this work, we introduce a new method for target detection and localization in which the need for a large homogenous target-free set of training range bins - which is traditionally used to estimate the interference covariance matrix - is reduced by the use of regression methods and pattern classification techniques to train over the 2D spatial-temporal space. It is shown that the proposed Learning-Based Space-Time Adaptive Processing (LBSTAP) technique not only reduces the need for a large, usually unavailable, homogenous target-free set of range bins, but also provides better performance in terms of Doppler side-lobe-level reduction.