Message Board

Dear readers, authors and reviewers,you can add a message on this page. We will reply to you as soon as possible!

2020 Volume 45 Issue 7
Article Contents

Chun-yan LIU, Chuan LI, Jing QI. A Perturbation Analysis of Block Sparse Signal Reconstruction with Block Orthogonal Matching Pursuit[J]. Journal of Southwest China Normal University(Natural Science Edition), 2020, 45(7): 144-149. doi: 10.13718/j.cnki.xsxb.2020.07.019
Citation: Chun-yan LIU, Chuan LI, Jing QI. A Perturbation Analysis of Block Sparse Signal Reconstruction with Block Orthogonal Matching Pursuit[J]. Journal of Southwest China Normal University(Natural Science Edition), 2020, 45(7): 144-149. doi: 10.13718/j.cnki.xsxb.2020.07.019

A Perturbation Analysis of Block Sparse Signal Reconstruction with Block Orthogonal Matching Pursuit

More Information
  • Received Date: 02/09/2018
    Available Online: 20/07/2020
  • MSC: TN911

  • In this paper, the block orthogonal matching pursuit(BOMP) algorithm has been presented when the measurement matrix is perturbed. The simulation results show that when the disturbance level is lower, the number of partial disturbance elements is less, the number of blocks is smaller or the number of samples is more, the relative error of reconstructed signal is smaller, that is, the reconstruction performance of disturbance BOMP algorithm is better. Compared with the traditional perturbation OMP algorithm, the experimental results show that the perturbation BOMP algorithm can deal with the block sparse signal more effectively, which shows that the signal structure is very important for signal recovery.
  • 加载中
  • [1] DONOHO D L. Compressed Sensing [J]. IEEE Transactions on Information Theory, 2006, 52(4): 1289-1306.

    Google Scholar

    [2] CANDES E J, WAKIN M B. An Introduction to Compressive Sampling [J]. IEEE Signal Processing Magazine, 2008, 25(2): 21-30.

    Google Scholar

    [3] LUSTIG M, DONOHO D, PAULY J M. Sparse MRI: The Application of Compressed Sensing for Rapid MR Imaging [J]. Magnetic Resonance in Medicine, 2007, 58(6): 1182-1195.

    Google Scholar

    [4] WRIGHT J, YANG A Y, GANESH A, et al. Robust Face Recognition via Sparse Representation [J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2009, 31(2): 210-227.

    Google Scholar

    [5] DUARTE M F, DAVENPORT M A, TAKHAR D, et al. Single-Pixel Imaging via Compressive Sampling [J]. IEEE Signal Processing Magazine, 2008, 25(2): 83-91.

    Google Scholar

    [6] 刘春燕, 张静, 王建军.冗余字典的扰动压缩数据分离[J].西南大学学报(自然科学版), 2015(9): 150-156.

    Google Scholar

    [7] 刘春燕, 王建军, 王文东, 等.基于非凸极小化的扰动压缩数据分离[J].电子学报, 2017, 45(1): 37-45.

    Google Scholar

    [8] AMALDI E, KANN V. On the Approximability of Minimizing Nonzero Variables or Unsatisfied Relations in Linear Systems [J]. Theoretical Computer Science, 1998, 209(1-2): 237-260.

    Google Scholar

    [9] HERMAN M A, STROHMER T. General Deviants: an Analysis of Perturbations in Compressed Sensing [J]. IEEE Journal of Selected Topics in Signal Processing, 2010, 4(2): 342-349.

    Google Scholar

    [10] LIU C Y, WANG J J, WANG W D, et al. Non-Convex Block-Sparse Compressed Sensing with Redundant Dictionaries [J]. IET Signal Processing, 2017, 11(2): 171-180.

    Google Scholar

    [11] PARVARESH F, VIKALO H, MISRA S, et al. Recovering Sparse Signals Using Sparse Measurement Matrices in Compressed DNA Microarrays [J]. IEEE Journal of Selected Topics in Signal Processing, 2008, 2(3): 275-285.

    Google Scholar

    [12] MAJUMDAR A, WARD R K. Compressed Sensing of Color Images [J]. Signal Processing, 2010, 90(12): 3122-3127.

    Google Scholar

    [13] MALIOUTOV D, CETIN M, WILLSKY A S. A Sparse Signal Reconstruction Perspective for Source Localization with Sensor Arrays [J]. IEEE Transactions on Signal Processing, 2005, 53(8): 3010-3022.

    Google Scholar

    [14] 刘春燕, 李川, 邱伟.基于冗余字典的扰动OMP算法研究[J].河南教育学院学报(自然科学版), 2019, 28(4): 12-16, 39.

    Google Scholar

    [15] WANG J J, ZHANG J, WANG W D, et al. A Perturbation Analysis of Nonconvex Block-Sparse Compressed Sensing [J]. Communications in Nonlinear Science and Numerical Simulation, 2015, 29(1-3): 416-426.

    Google Scholar

    [16] FOUCART S, LAI M J. Sparsest Solutions of Underdetermined Linear Systems via $\ell $ q-Minimization for 0 <q≤1[J]. Applied and Computational Harmonic Analysis, 2009, 26(3): 395-407.

    Google Scholar

    [17] FOUCART S. A Note on Guaranteed Sparse Recovery via $\ell $1-minimization [J]. Applied and Computational Harmonic Analysis, 2010, 29(1): 97-103.

    Google Scholar

    [18] CAI T T, ZHANG A R. Sparse Representation of a Polytope and Recovery of Sparse Signals and Low-Rank Matrices [J]. IEEE Transactions on Information Theory, 2014, 60(1): 122-132.

    Google Scholar

    [19] 张贤达.矩阵分析与应用[M].北京:清华大学出版社, 2004.

    Google Scholar

    [20] CHANG L H, WU J Y. An Improved RIP-Based Performance Guarantee for Sparse Signal Recovery via Orthogonal Matching Pursuit[J]. IEEE Transactions on Information Theory, 2014, 60(9): 5702-5715.

    Google Scholar

    [21] MO Q. A Sharp Restricted Isometry Constant Bound of Orthogonal Matching Pursuit[EB/OL]. (2015-01-08)[2018-05-06]. https://arxiv.org/abs/1501.01708.

    Google Scholar

    [22] WEN J M, ZHOU Z C, LIU Z L, et al. Sharp Sufficient Conditions for Stable Recovery of Block Sparse Signals by Block Orthogonal Matching Pursuit [J]. Applied and Computational Harmonic Analysis, 2019, 47(3): 948-974.

    Google Scholar

  • 加载中
通讯作者: 陈斌, bchen63@163.com
  • 1. 

    沈阳化工大学材料科学与工程学院 沈阳 110142

  1. 本站搜索
  2. 百度学术搜索
  3. 万方数据库搜索
  4. CNKI搜索

Figures(5)  /  Tables(1)

Article Metrics

Article views(644) PDF downloads(8) Cited by(0)

Access History

Other Articles By Authors

A Perturbation Analysis of Block Sparse Signal Reconstruction with Block Orthogonal Matching Pursuit

Abstract: In this paper, the block orthogonal matching pursuit(BOMP) algorithm has been presented when the measurement matrix is perturbed. The simulation results show that when the disturbance level is lower, the number of partial disturbance elements is less, the number of blocks is smaller or the number of samples is more, the relative error of reconstructed signal is smaller, that is, the reconstruction performance of disturbance BOMP algorithm is better. Compared with the traditional perturbation OMP algorithm, the experimental results show that the perturbation BOMP algorithm can deal with the block sparse signal more effectively, which shows that the signal structure is very important for signal recovery.

  • 压缩感知理论是建立在矩阵分析、统计概率、优化与运筹学、泛函分析等理论基础上的信息采样与重构的理论[1-2].压缩感知是利用信号的稀疏表示,对信号进行有效压缩采样,通过压缩采样与非线性重构算法完美地重构信号.这一理论为信号恢复带来了新的处理方法,被广泛应用于医学成像、雷达探测、无线通信等诸多领域[3-5].在信号x$\mathbb{R}$n稀疏或近似稀疏的条件下,压缩感知理论转化为如下欠定线性模型:

    其中:y$\mathbb{R}$m为观测信号,A$\mathbb{R}$m×n(m < < n)为测量矩阵,x$\mathbb{R}$n为待重构的信号.实际观测中,加性噪音v时常影响信号y,上述压缩感知模型常转化为下述模型:

    此外,实际应用中测量矩阵常常受外界测量或其他因素的影响,导致测量矩阵A受扰动矩阵E的影响,即实际测量矩阵$\mathit{\boldsymbol{\widehat A}} = \mathit{\boldsymbol{A}} + \mathit{\boldsymbol{E}}$,其中A称为理想测量矩阵.本文将扰动矩阵E与加性噪音v同时存在的压缩感知问题称之为完全扰动问题[6-7],该问题对应的模型为:

    从欠定线性模型中求解一个最稀疏的解是压缩感知的核心问题,该问题实际上是一个l0问题.由于l0问题在多项式时间内不能有效求解[8],文献[9]研究了如下l1完全扰动问题:

    其中εAKy为全噪音参数.文献[9]中研究了信号鲁棒重构的限制等容条件(restricted isometrics property,RIP).信号重构条件为压缩感知的发展提供了重要研究意义.

    在实际应用中,自然信号的结构丰富多彩.非零元素以块的结构方式出现的信号称为块稀疏信号[10],例如DNA-微阵列[11]、彩色图像处理[12]和源定位[13]等.块稀疏压缩感知实际是单独处理重构信号的每一小块.块稀疏压缩感知使块信号重构与采样的速度更快,且占用存储空间更少.若块稀疏信号用数学模型来表示,则在分块$\mathscr{I}$ ={d1,…,dN}给定情况下,可以将任意向量x$\mathbb{R}$n描述为如下形式

    相应地,测量矩阵A$\mathbb{R}$n×N也可以描述为如下形式

    其中向量x的第i个子块为x[i],第i块的大小为di.如果在给定分块$\mathscr{I}$下,信号x至多有K个非零块,即$||\mathit{\boldsymbol{x}}|{|_{2, 0}} = \sum\limits_{i = 1}^N I \left({||\mathit{\boldsymbol{x}}[i]|{|_2}} \right) \le K$,则称x为块K稀疏信号,其中I(·)表示指示函数[14].实际上,当di=1(i=1,…,N)时,块稀疏信号x为传统稀疏信号.此外,信号x$\frac{{{\mathit{\boldsymbol{l}}_2}}}{{{\mathit{\boldsymbol{l}}_1}}}$范数$||\mathit{\boldsymbol{x}}|{|_{2, 1}} = \sum\limits_{i = 1}^N I ||\mathit{\boldsymbol{x}}[i]|{|_2}$;信号x$\frac{{{\mathit{\boldsymbol{l}}_2}}}{{{\mathit{\boldsymbol{l}}_\infty }}}$范数‖x2,∞=max{‖x[1]2,…,‖x[N]‖2};规定信号x的范数‖x2,2=‖x2.

    实际上,在稀疏信号的重构中,测量矩阵要求满足限制性等容条件或相干性等重构条件.特别地,在不引起混淆的情况下,本文记矩阵A的块限制等容常数δBK=δK.为了重构块稀疏信号,文献[15]引入了$\mathit{\boldsymbol{\widehat A}}$的块限制性等容条件(BRIP).设测量矩阵A的块限制性等容常数为δK(K=1,2,…),εA(K)为相对误差上限,且$\frac{{||\mathit{\boldsymbol{E}}||_2^{(K)}}}{{||\mathit{\boldsymbol{A}}||_2^{(K)}}} \le \varepsilon _A^{(K)}$,若${\hat \delta _{K, \max }} = \left({1 + {\delta _K}} \right){\left({1 + \varepsilon _\mathit{\boldsymbol{A}}^{(K)}} \right)^2} - 1$,对于任意块K稀疏信号x$\mathbb{R}$n,矩阵$\mathit{\boldsymbol{\widehat A}}$=A+E的块限制性等容常数(BRIC)为满足

    的最小常数${\hat \delta _K} \le {\hat \delta _K}, \max $,则称矩阵$\mathit{\boldsymbol{\widehat A}}$满足块限制性等容条件(BRIP).类似于文献[9],有如下不等式成立:

    通过研究l1问题,学者获得了大量的充分条件:文献[16]得到了信号重构的RIPδ2k < 0.453 1;文献[17]将RIP重构条件放松为δ2k < 0.465 2;文献[18]将RIP改进为${\delta _{2k}} < \frac{{\sqrt 2 }}{2}$.

    压缩感知的重构理论与重构算法具有重要的应用价值[19-22].

    本文借鉴文献[22]延展了一类完全扰动的块压缩感知问题.令噪音上界为l2范数与$\frac{{{\mathit{\boldsymbol{l}}_2}}}{{{\mathit{\boldsymbol{l}}_\infty }}}$范数,其问题具体形式可分别表示为如下形式:

    其中εAKy=‖A2(1+εA)εAKy.近年来压缩感知基本是以RIP为理论基础,但是验证RIP理论是一个在多项式时间内不能有效解决的问题,因此,利用RIP理论获得的结论,在实际应用中很难实行.本文针对问题(5),建立了处理块稀疏信号的扰动BOMP算法,根据扰动BOMP算法建立了各项仿真实验,验证了扰动BOMP算法的鲁棒性和稳定性.

1.   扰动的块正交匹配追踪(BOMP)算法
  • 本文考虑了理想测量矩阵A受扰动矩阵E的影响,即实际测量矩阵$\mathit{\boldsymbol{\widehat A}}$ =A+E.对于扰动矩阵E与加性噪音v同时存在的块压缩感知问题,称为完全扰动的块压缩感知问题.此外,本文给出了完全扰动的块压缩感知问题的完全扰动块正交匹配追踪(BOMP)算法.对于完全扰动的块压缩感知问题(5),求解测量矩阵$\mathit{\boldsymbol{\widehat A}}$与残差向量rk-1的最强相关列λk

    其中Λk表示第k次标签集.通过最小二乘法估计最小化问题的解

    下面介绍完全扰动的BOMP算法具体迭代过程:

2.   数值实验与分析
  • 重构算法是压缩感知理论中的关键环节,本文从不同的角度建立了一系列数值仿真实验,研究了完全扰动的BOMP算法的重构性与有效性.为了保证算法的有效性,模拟实验对块稀疏信号采用均匀分块,即di=d(i=1,…,N).实验采用随机高斯矩阵作为测量矩阵A(A$\mathbb{R}$128×256),设定采样数m=128,信号维度n=256,扰动矩阵‖E2=εAA2E中的元素服从正态分布,测量矩阵$\mathit{\boldsymbol{\widehat A}}$ =A+E,测量$\mathit{\boldsymbol{\widehat y}}$ =$\mathit{\boldsymbol{\widehat A}}$ x+v.假设块稀疏信号x中的元素服从标准正态分布,噪音v为高斯噪音.实验中扰动BOMP算法的重构性能运用相对误差$\frac{{{{\left\| {\mathit{\boldsymbol{x}} - {\mathit{\boldsymbol{x}}^*}} \right\|}_2}}}{{{{\left\| \mathit{\boldsymbol{x}} \right\|}_2}}}$ (x*表示重构信号,x表示原始信号)以及信噪比SNR=20 log10$\left({\frac{{{{\left\| \mathit{\boldsymbol{x}} \right\|}_2}}}{{{{\left\| {\mathit{\boldsymbol{x}} - {\mathit{\boldsymbol{x}}^*}} \right\|}_2}}}} \right)$来刻画.数值仿真实验在CPU为Intel奔腾i5-4300U(2.49 GHz),内存为4GB的电脑上运行,运行软件为MATLAB 7.14(R2012a).为了让实验结果更加准确,所有实验以独立重复200次的平均值作为最终统计结果.

    为了说明扰动BOMP算法的重构效果,在扰动水平分别为εA=0与εA=0.1下,图 1呈现了扰动BOMP算法的重构信号x*和原始信号x. 图 1描述了扰动BOMP算法的重构效果,该实验说明原始信号x的重构效果理想.此外,图 1也表明了扰动水平εA=0.1比εA=0重构效果略差,说明扰动BOMP算法的重构性能受扰动水平影响.因此,为了更为详细说明扰动水平对扰动BOMP算法的影响情况,首先将信号均分为128块,扰动水平分别取0,0.05,0.1,0.15,0.2. 图 2分别对块稀疏度从15到35变化的信号的信噪比以及相对误差进行了重构效果比较.从图 2可以看出:扰动水平越大,重构效果越差.此外,图 3表明运用扰动BOMP算法处理块稀疏信号的重构误差比扰动OMP算法更低,说明信号结构对于信号恢复至关重要.

    与扰动OMP算法的重构效果比较接下来,考虑块稀疏度与分块数对扰动BOMP算法的影响.在扰动水平εA=0.1时,信号的块大小d分别取1,2,4,8,图 4分别对信号的信噪比以及相对误差进行了重构效果比较. 图 4表明当块稀疏度从8到48变化时,块大小d均为2,4,8,曲线变化趋势不明显,即重构误差受块稀疏度影响较小.显然块大小d越大,重构效果越好.另外,在扰动水平εA=0.1时,信号的块大小d分别取1,2,4,8,图 5(a)中描述了信号采样数从64到128变化的相对误差变化趋势.显然,不管块大小d取1,2,4,8中的哪个值,采样越多,信号重构的误差越低,即采样数能有效降低信号重构误差.本文随机选取了矩阵A的部分或全部元素进行了扰动,其中测量矩阵A的元素部分扰动比例分别为0%,10%,20%,…,100%,探讨了部分扰动对扰动BOMP算法的重构效果的影响. 图 5(b)是部分扰动与相对误差的关系图,其中ξ表示矩阵A中元素部分扰动的比例.当εA=0时,矩阵A中元素扰动的比例ξ越小,信号重构误差越小.显然,扰动水平、部分扰动、分块数或采样数影响扰动BOMP算法的信号重构效果,且当扰动水平越低、部分扰动元素越少、分块数越小或采样数越多时,扰动BOMP算法重构信号的相对误差越小,即块稀疏信号重构效果越好.

3.   总结
  • 本文利用完全扰动块压缩感知重构理论,运用数值仿真实验,研究了扰动水平、部分扰动、分块数以及采样数对扰动BOMP算法的重构性能的影响.实验表明:当扰动水平越低、部分扰动元素越少、分块数越小或采样数越多时,重构信号的相对误差越小,扰动BOMP算法能更加有效处理块稀疏信号.

Figure (5)  Table (1) Reference (22)

Catalog

    /

    DownLoad:  Full-Size Img  PowerPoint
    Return
    Return