Greedy Recovery of Sparse Signals with Dynamically Varying Support

In this paper, we propose a low-complexity greedy recovery algorithm which can recover sparse signals with time-varying support. We consider the scenario where the support of the signal (i.e., the indices of nonzero elements) varies smoothly with certain temporal correlation. We model the indices of support as discrete-state Markov random process. Then, we formulate the signal recovery problem as joint estimation of the set of the support indices and the amplitude of nonzero entries based on the multiple measurement vectors. We successively identify the element of the support based on the maximum a posteriori (MAP) criteria and subtract the reconstructed signal component for detection of the next element of the support. Our numerical evaluation shows that the proposed algorithm achieves satisfactory recovery performance at low computational complexity.