An Efficient Greedy Algorithm for Sparse Recovery in Noisy Environment

摘要

Greedy algorithm are in widespread use for sparse recovery because of itsefficiency. But some evident flaws exists in most popular greedy algorithms,such as CoSaMP, which includes unreasonable demands on prior knowledge oftarget signal and excessive sensitivity to random noise. A new greedy algorithmcalled AMOP is proposed in this paper to overcome these obstacles. UnlikeCoSaMP, AMOP can extract necessary information of target signal from sampledata adaptively and operate normally with little prior knowledge. The recoveryerror of AMOP is well controlled when random noise is presented and fades awayalong with increase of SNR. Moreover, AMOP has good robustness on detailedsetting of target signal and less dependence on structure of measurementmatrix. The validity of AMOP is verified by theoretical derivation. Extensivesimulation experiment is performed to illustrate the advantages of AMOP overCoSaMP in many respects. AMOP is a good candidate of practical greedy algorithmin various applications of Compressed Sensing.