Bayesian sparse estimation of a radar scene with weak and strong targets

We consider the problem of estimating a finite number of atoms from a dictionary embedded in white noise, using a sparse signal representation (SSR) approach, a problem which is relevant in many radar applications. In particular, the estimation of a radar scene consisting of targets with wide amplitude range can be challenging since the sidelobes of a strong target can disrupt the estimation of a weak one. In this paper, we present a Bayesian algorithm able to estimate weak targets possibly hidden by strong ones. The main strength of this algorithm lies in a novel sparse-promoting prior distribution which decorrelates sparsity level and target power and makes the estimation process span the whole target power range. This algorithm is implemented through a Monte-Carlo Markov chain. It is successfully evaluated on synthetic and semiexperimental radar data.