A remark on joint sparse recovery with OMP algorithm under restricted isometry property

The theory and algorithms for recovering a sparse representation of multiple measurement vector (MMV) are studied in compressed sensing community. The sparse representation of MMV aims to find the K-row sparse matrix X such that Y=AX, where A is a known measurement matrix. In this paper, we show that, if the restricted isometry property (RIP) constant δ_K+1 of the measurement matrix A satisfies δ_K+1<1K+1, then all K-row sparse matrices can be recovered exactly via the Orthogonal Matching Pursuit (OMP) algorithm in K iterations based on Y=AX. Moreover, a matrix with RIP constant δ_K+1=1K+0.086 is constructed such that the OMP algorithm fails to recover some K-row sparse matrix X in K iterations. Similar results also hold for K-sparse signals recovery. In addition, our main result further improves the proposed bound δ_K+1=1K by Mo and Shen [12] which can not guarantee OMP to exactly recover some K-sparse signals.