I. Introduction
Sparse representation and sparse modelling is a topic that has attracted many researchers in the last two decades [1], [2]. The sparse representation is a linear equation system which is underdetermined and its number of unknowns is larger than the number of equations [3]. The number of zero elements in unknown vector is assumed much larger than non-zero unknown that means the unknown vector is sparse. The sparse recovery is a method to solve this kind of equation systems based on solving an optimization problem. Sparse representation has many applications in various fields, including Compressed Sensing (CS) [1], [3], Blind Source Separation (BSS) [4], [5], communication and channel estimation [6], [7], radar imaging [8], [9], machine learning and machine vision [10], [11]. The compressed sensing might be considered as a turning point for sparse representation.