基于差值映射的压缩感知MUSIC算法

多快拍(MMV)问题旨在恢复具有相同稀疏结构的多列信号。在传统阵列信号处理中MMV问题的求解通常采用多重信号分类(MUSIC)等确定性方法实现，但当快拍数不足或存在相干源时该类方法失效；而在压缩感知(CS)的概率求解模型下，即使信源相干也能得到恢复结果，但现有算法普遍性能不足。近期Kim等人的研究表明，将CS与MUSIC相结合可得到比二者更加优秀的性能和更为宽泛的使用条件，该方法被称作压缩感知 MUSIC或CS-MUSIC算法。作为一种投影型非凸优化算法，差值映射(DM)最早用于解决X射线晶体学中的相位恢复问题，并逐渐在其他非凸及压缩感知问题的求解中展示出优良性能。该文提出一种基于差值映射的CS-MUSIC算法，仿真结果表明该算法在MMV问题求解中十分有效，相比经典CS-MUSIC具有更高的恢复成功率。

Compressive Sensing MUSIC Algorithm Based on Difference Map

The Multiple Measurement Vectors (MMV) problem addresses the recovery of unknown input vectors which share the same sparse support. The Compressed Sensing (CS) has the capability of estimating the sparse support even in coherent cases, where the traditional array processing approaches like MUltiple SIgnal Classification (MUSIC) often fail. However, CS guarantees the accurate recovery in a probabilistic manner, and often shows inferior performance in cases where the traditional ways succeed. Recently, a novel compressive MUSIC (or CS-MUSIC) algorithm is proposed by Kim et al., in which both the advantages of CS and traditional MUSIC-like methods are combined together. As an iterative projecting algorithm, Difference Map (DM) is first used to solve the phase retrieval problem in crystallography. Recent results show that it has excellent performance in solving a wide variety of non-convex problems like compressed sensing. In this paper, a DM-based CS-MUSIC algorithm is proposed. Experiments show that the proposed algorithm is very effective in MMV problem solving and the success rate of CS-MUSIC is dramatically improved.