Susanto Rahardja / Northwestern Polytechnical University
Lifan Zhao / Nanyang Technological University
We consider the problem of recovering clustered sparse signals with unknown cluster sizes and locations. We propose an alternative extended block sparse Bayesian learning algorithm (AEBSBL) for clustered sparse signal recovery. By analyzing the graphic models of extended block sparse Bayesian
learning algorithm (EBSBL), a cluster structured prior for sparse coefficients is obtained, which encourages dependencies among neighboring coefficients by properly manipulating the hyperparameters of the neighborhood. With the sparse prior, other necessary probabilistic modelings are constructed
and Expectation and Maximization (EM) is applied to infer all the hidden variable and unknown parameters. The alternative algorithm reduces the unknowns of EBSBL and is more effective than EBSBL. Numerical results of comprehensive simulations demonstrate that the proposed algorithm outperforms other recently reported clustered sparse signal recovery algorithms particularly under noisy and low sampling scenarios.