Compressive radar clutter subspace estimation using dictionary learning

Space-Time Adaptive Processing (STAP) based on matched filter processing in the presence of additive clutter (modeled as colored noise) requires knowledge of the clutter covariance matrix. In practice, this is estimated via the sample covariance matrix using samples from the neighboring range bins around the reference bin. By applying compressive sensing, the number of training samples needed to estimate the covariance matrix can be significantly reduced, provided that the basis mismatch problem, inherent to compressive sensing can be mitigated. This paper presents an adaptive approach to choosing the best sparsifying basis, using dictionary learning to estimate the radar clutter subspace. Numerical results show that the proposed algorithm achieves the desired reduction in training samples, and is more accurate than previous reduced-rank algorithm baseline.