基于贝叶斯理论的压缩感知恢复算法研究
发布时间：2018-11-15 10:32
【摘要】：随着人们对移动通信的需求不断增长，频谱资源的分配就变得越来越困难，而认知无线电技术可以解决这一难题。频谱感知作为认知无线电技术的关键，其目的是检测频谱空穴。传统的频谱感知只能对单个频段进行感知，为了提高检测效率，出现了宽带频谱感知技术。在对宽带信号进行感知时，极高的采样速率成为了限制这一技术的瓶颈，利用压缩感知方法可以解决这一难题。贝叶斯方法是近些年提出的一类压缩感知算法，它可以利用不同的先验概率灵活地构建稀疏信号的恢复模型，还能给出恢复信号的误差范围，具有优越的性能。因此本文的重点就是适用于宽带频谱感知的贝叶斯压缩感知恢复算法研究。 本文首先介绍了贝叶斯建模的过程，然后对使用拉普拉斯先验的压缩感知算法进行了改进，提出了一种快速算法——L-BSC算法，并给出详细流程。同时，本文将一种在贝叶斯框架下的自适应观测矩阵设计方法与提出的快速算法结合在一起，得到一种自适应的快速算法。仿真结果表明，这种自适应的快速贝叶斯算法不但在恢复一般信号时性能良好，应用于宽带频谱感知场景时也能获得很高的频谱重构精度，具有优良的频谱检测性能。因此认为该算法性能良好，适合应用于宽带压缩频谱感知中。 考虑到宽带频谱感知中频谱分配造成的频域块稀疏结构，本文引入了块稀疏贝叶斯学习（Block Sparse Bayesian Learning, BSBL）框架，并介绍了两种基于该框架提出的算法。在此基础上，本文提出了一种将BSBL算法和group lasso方法相结合的算法——BSBL-Group Lasso算法。该算法大大减少了迭代次数，提高了执行效率，并且保证良好的算法性能。为了解决某些时候不能获得信号块分布情况的问题，本文扩展了BSBL框架，得到一种应用于信号块分布未知情况下的模型，并在此基础上对算法进行了改进，得到了BSBL-EEM和BSBL-EBO算法。仿真结果显示，BSBL-Group Lasso算法在恢复块稀疏信号时可获得较高的恢复精度，，在宽带压缩频谱感知中可获得优良的检测性能。而BSBL-EEM和BSBL-EBO算法在信号的块分布未知的情况下，即使用户定义的分块情况与实际信号不一致也可获得很高的恢复精度，应用范围很广。
[Abstract]:With the increasing demand for mobile communication, the allocation of spectrum resources becomes more and more difficult, and cognitive radio technology can solve this problem. Spectrum sensing, as the key of cognitive radio technology, aims to detect spectral holes. Traditional spectrum sensing can only perceive a single frequency band. In order to improve detection efficiency, broadband spectrum sensing technology has emerged. The extremely high sampling rate becomes the bottleneck of this technique when we perceive the wideband signal. The compression sensing method can solve this problem. Bayesian method is a kind of compression sensing algorithm proposed in recent years. It can flexibly construct the sparse signal recovery model with different prior probabilities, and it can also give the error range of the recovery signal, so it has superior performance. Therefore, the emphasis of this paper is the research of Bayesian compressed sensing restoration algorithm suitable for wideband spectrum sensing. This paper first introduces the process of Bayesian modeling, then improves the compression perception algorithm using Laplace priori, proposes a fast algorithm, L-BSC algorithm, and gives a detailed flow chart. At the same time, an adaptive observation matrix design method based on Bayesian framework is combined with the proposed fast algorithm to obtain an adaptive fast algorithm. Simulation results show that the adaptive fast Bayesian algorithm not only performs well in recovering general signals, but also achieves high spectral reconstruction accuracy when it is applied to wideband spectrum sensing scene, and has excellent spectrum detection performance. Therefore, the proposed algorithm has good performance and is suitable for wideband spectrum sensing. Considering the frequency domain block sparse structure caused by spectrum allocation in wideband spectrum sensing, a block sparse Bayesian learning (Block Sparse Bayesian Learning, BSBL) framework is introduced, and two algorithms based on this framework are introduced. On this basis, this paper proposes a new algorithm, BSBL-Group Lasso algorithm, which combines BSBL algorithm with group lasso algorithm. The algorithm greatly reduces the number of iterations, improves the execution efficiency and ensures good performance. In order to solve the problem that the signal block distribution can not be obtained at some time, this paper extends the BSBL framework, and obtains a model which is applied to the unknown signal block distribution, and then improves the algorithm. BSBL-EEM and BSBL-EBO algorithms are obtained. Simulation results show that the BSBL-Group Lasso algorithm can achieve high recovery accuracy when restoring block sparse signals, and it can obtain excellent detection performance in wideband compressed spectrum sensing. The BSBL-EEM and BSBL-EBO algorithms can obtain high recovery accuracy even if the user-defined block is not consistent with the actual signal even if the block distribution of the signal is unknown and has a wide range of applications.
【学位授予单位】：哈尔滨工业大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN911.7
本文编号：2333057
[Abstract]:With the increasing demand for mobile communication, the allocation of spectrum resources becomes more and more difficult, and cognitive radio technology can solve this problem. Spectrum sensing, as the key of cognitive radio technology, aims to detect spectral holes. Traditional spectrum sensing can only perceive a single frequency band. In order to improve detection efficiency, broadband spectrum sensing technology has emerged. The extremely high sampling rate becomes the bottleneck of this technique when we perceive the wideband signal. The compression sensing method can solve this problem. Bayesian method is a kind of compression sensing algorithm proposed in recent years. It can flexibly construct the sparse signal recovery model with different prior probabilities, and it can also give the error range of the recovery signal, so it has superior performance. Therefore, the emphasis of this paper is the research of Bayesian compressed sensing restoration algorithm suitable for wideband spectrum sensing. This paper first introduces the process of Bayesian modeling, then improves the compression perception algorithm using Laplace priori, proposes a fast algorithm, L-BSC algorithm, and gives a detailed flow chart. At the same time, an adaptive observation matrix design method based on Bayesian framework is combined with the proposed fast algorithm to obtain an adaptive fast algorithm. Simulation results show that the adaptive fast Bayesian algorithm not only performs well in recovering general signals, but also achieves high spectral reconstruction accuracy when it is applied to wideband spectrum sensing scene, and has excellent spectrum detection performance. Therefore, the proposed algorithm has good performance and is suitable for wideband spectrum sensing. Considering the frequency domain block sparse structure caused by spectrum allocation in wideband spectrum sensing, a block sparse Bayesian learning (Block Sparse Bayesian Learning, BSBL) framework is introduced, and two algorithms based on this framework are introduced. On this basis, this paper proposes a new algorithm, BSBL-Group Lasso algorithm, which combines BSBL algorithm with group lasso algorithm. The algorithm greatly reduces the number of iterations, improves the execution efficiency and ensures good performance. In order to solve the problem that the signal block distribution can not be obtained at some time, this paper extends the BSBL framework, and obtains a model which is applied to the unknown signal block distribution, and then improves the algorithm. BSBL-EEM and BSBL-EBO algorithms are obtained. Simulation results show that the BSBL-Group Lasso algorithm can achieve high recovery accuracy when restoring block sparse signals, and it can obtain excellent detection performance in wideband compressed spectrum sensing. The BSBL-EEM and BSBL-EBO algorithms can obtain high recovery accuracy even if the user-defined block is not consistent with the actual signal even if the block distribution of the signal is unknown and has a wide range of applications.
【学位授予单位】：哈尔滨工业大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN911.7
【参考文献】
相关期刊论文 前6条
1 江晓林;顾学迈;何晨;;基于压缩感知的联合协作频谱感知算法[J];上海交通大学学报;2013年07期
2 甘伟;许录平;张华;苏哲;;一种贪婪自适应压缩感知重构[J];西安电子科技大学学报;2012年03期
3 付宁;曹离然;彭喜元;;基于子空间的块稀疏信号压缩感知重构算法[J];电子学报;2011年10期
4 杨俊杰;刘海林;;增广Lagrange函数优化算法在稀疏信号重构问题中的应用[J];计算机科学;2011年09期
5 王国富;张海如;张法全;徐婷;;基于改进遗传算法的正交匹配追踪信号重建方法[J];系统工程与电子技术;2011年05期
6 李树涛;魏丹;;压缩传感综述[J];自动化学报;2009年11期
相关硕士学位论文 前1条
1 刘亚新;用于压缩感知信号重建的算法研究[D];北京交通大学;2010年
本文编号：2333057
本文链接：https://www.wllwen.com/kejilunwen/wltx/2333057.html
