A Universal Sparse Signal Reconstruction Algorithm via Backtracking and Belief Propagation

Abstract

The belief propagation (BP) algorithm under the Bayesian framework can accelerate Compressed Sensing (CS) encoding and decoding by using the sparse encoder matrix. To improve the reconstruction performance we consider a backtracking-based belief propagation algorithm (CS-BBP) for the sparse signal reconstruction. The backtracking is added after performing BP and minimum mean square error (MMSE) estimate in every iteration. Simulation results show that the CS-BBP is a universal reconstruction algorithm which has a good performance for both 1-D Gaussian and 2-D image signal reconstructions.