An improved subspace pursuit algorithm based on regularized multipath search

Compressive sensing (CS) is a novel signal sampling theory and it can recover sparse or compressive signals with lower rates than their Nyquist rates. Greedy pursuit algorithms are important recovery algorithms in CS. In this paper, we study the performance of subspace pursuit (SP) greedy algorithm and propose a modified SP termed as regularized multipath subspace pursuit (RMSP), which divides the test set into several subsets in each iteration by means of regularization, and gets several candidates of the support set by subsequent SP processing, then selects one candidate with the minimal residual as the estimated support set in the iteration. Finally simulation experiments are made to demonstrate that the performance of the RMSP is superior to that of the classical SP algorithm.