Multipath Subspace Pursuit for Compressive Sensing Signal Reconstruction

This paper proposes a greedy reconstruction algorithm to recover sparse signal from compressed measurements, called multipath subspace pursuit (MSP). Different from the subspace pursuit (SP), the MSP creates several candidates of the support set when iteration ends. We can select one best candidate as the final support set. At the beginning of each iteration, MSP uses a nuanced method to add the atoms into the support set, it makes two paths for next iteration. The processing to find the minimum residual becomes a tree search problem. Simulation results show that the proposed MSP is better than SP in reconstructing sparse signal.