西安电子科技大学学报 ›› 2016, Vol. 43 ›› Issue (2): 1-5+28.doi: 10.3969/j.issn.1001-2400.2016.02.001

• 研究论文 •    下一篇

支撑驱动的非凸压缩感知恢复算法

王峰1,2;向新2;易克初1;熊磊2   

  1. (1. 西安电子科技大学 综合业务网理论及关键技术国家重点实验室,陕西 西安  710071;
    2. 空军工程大学 航空航天工程学院,陕西 西安  710038)
  • 收稿日期:2014-10-04 出版日期:2016-04-20 发布日期:2016-05-27
  • 通讯作者: 王峰
  • 作者简介:王峰(1978-),男,西安电子科技大学博士研究生,E-mail: wangfengisn@163.com.
  • 基金资助:

    国家自然科学基金资助项目(61379104);陕西省自然科学基金资助项目(2014JM2-6106)

Support driven recovery algorithm for non-convex compressed sensing

WANG Feng1,2;XIANG Xin2;YI Kechu1;XIONG Lei2   

  1. (1. State Key Lab. of Integrated Service Networks, Xidian Univ., Xi'an  710071, China;
    2. Aeronautics and Astronautics Engineering College, Air Force Engineering Univ., Xi'an  710038, China)
  • Received:2014-10-04 Online:2016-04-20 Published:2016-05-27
  • Contact: WANG Feng

摘要:

为解决带噪压缩感知信号恢复的难题,提出一种基于支撑驱动的恢复算法,分2步完成稀疏信号的恢复.①使用阈值基追踪方法获取信号支撑信息,并生成权值矩阵与所需其他参数.②使用迭代重加权算法求解非凸目标函数.在理论分析的基础上,与现有7种有竞争力的算法(含oracle 估计器)进行了数值仿真比较,结果证明,文中算法以较低的运算量实现了高概率恢复.

关键词: 压缩感知, 基追踪, 迭代重加权最小p范数

Abstract:

A novel method is presented for the purpose of recovering sparse high dimensional signals from few linear measurements, especially in the noisy case. The proposed method works in the following two steps: ①The support of signal is approximately identified via Thresholded Basis Pursuit(TBP), the weighting matrix and parameters needed for the next step are also computed; ②The Iteratively Reweighted Lp Minimization(IRLp) procedure is used to solve the non-convex objective function. As theoretic interpretation and simulation results show, lower computational complexity is required for the proposed Support Driven IRLp(SD_IRLp) algorithm for high probability recovery, in comparison to 7 analogous methods(including an oracle estimator).

Key words: compressed sensing, basis pursuit, iteratively reweighted Lp minimization