FAST BLOCK SPARSE BAYESIAN LEARNING FOR RECOVERY OF BLOCK-SPARSE SIGNAL

Sparse Bayesian learning (SBL) has shown a superior capability for sparse signal recovery. In practice, block-sparse signals without knowledge of block structure are often encountered. To recover arbitrary block-sparse signals in a low-complexity manner, we extend the existing efficient SBL algorithm and propose a fast block SBL algorithm for block-sparse signal recovery. It adopts the pattern-coupled priors of entries to encourage block sparse structure and its update rules of the hyperparameters are derived by employing a joint sparsity assumption to deal with the hyperparameter coupling. Finally, numerical experimental results show that the proposed algorithm can efficiently recover block-sparse signal, especially in the case of low SNR when compared to some closely related SBL algorithms.