Sparse Bayesian Learning for Long Coherent Integration Time in Passive Radar Systems

Maximising the radar coherent integration time is crucial when performing detection and parameter estimation on weak target echoes. The integration time is limited however by the migration of a target of interest out of a range and Doppler cell. To account for the range migration it is proposed to build here upon a Keystone transform and develop a joint sparse super-resolution target parameter estimation and target detection method using a super-resolution sparse Bayesian learning framework. The estimation scheme uses a variational version of the space-alternating generalized expectation maximization (VB-SAGE) algorithm, which permits reducing the numerical complexity of the scheme. Moreover, since the search space is not discretized, the parameter estimates are not restricted by the system resolution. Our simulation experiments demonstrate the effectiveness of the algorithm.