基于频域稀疏特性的频谱感知算法研究

发布时间：2019-04-17 09:03

【摘要】：随着无线通信技术的迅速发展,静态的频谱分配策略和使用方式已不能满足其高速率、高质量的通信需求。而认知无线电系统为动态分配和使用频谱提供新的方案,自1999年提出以来,受到了广泛关注和研究。作为认知无线电的核心技术,频谱检测算法通过实时准确的识别“频谱空穴”,为非授权用户提供接入机会,进而提高频谱利用效率。信号在某些域具有的稀疏特性,一方面可以更加集中信号的能量,另一方面也可以有效降低采样速率和难度,分别对应频谱检测算法中对高检测概率和动态实时检测的要求。所以,本文尝试将稀疏特性和检测算法结合,着重研究基于频域稀疏特性的检测算法,并对算法的检测性能进行分析。首先,论文介绍了课题的研究背景和意义,分析现有检测算法的优势和缺陷,引出本文重点研究的基于功率谱估计最大值的检测算法和基于MWC-SBL系统的检测算法,接着总结了这两种算法所用到的基础理论,包括功率谱估计、随机带宽转换器、稀疏贝叶斯学习等,并给出了相应的仿真结果。其次,深入分析基于功率谱估计最大值的检测算法,定义统计判决变量为信号功率谱估计最大值。在分析统计判决变量的均值、方差、相关性等统计特性之后,利用卡方分布对H0、H1假设下的统计判决变量进行分布建模,推导检测概率、虚警概率、判决门限等理论表达式。仿真部分,验证卡方分布建模的正确性和选取估计最大值作为统计判决变量的优势,分析采样数据长度、功率谱分段数目对检测性能的影响,并与能量检测算法进行对比。随后的仿真表明,不同的窗函数选取对检测概率和卡方分布建模的准确性都有影响且二者呈现相反的趋势。尽管基于功率谱估计最大值的算法在稀疏窄带信号的检测中有着独特的优势,但由于前端采样的限制,对于稀疏多频带信号并不能达到及时检测的要求。最后,研究了基于MWC-SBL系统的检测算法,以压缩采样为基础,利用稀疏贝叶斯学习算法,充分挖掘观测数据的信息,为宽带检测提供另一种方案。定义支撑集为统计判决变量,通过支撑集的检测情况,完成每个频带信号存在与否的判断。在完成算法描述后,根据恢复得到的支撑集与原始支撑集之间的关系,定义检测概率、虚警概率的统计表达式。随后的仿真重点关注固定位置单信号和多信号的检测性能、随机位置单信号和多信号的检测性能、不同匹配原则对检测性能的提升以及均方误差等指标,并与基于MWC-OMP的检测算法进行对比。仿真表明,基于MWC-SBL检测算法在检测概率等性能指标上比MWC-OMP检测算法更具优势。
[Abstract]:With the rapid development of wireless communication technology, static spectrum allocation strategy and usage mode can no longer meet its high-speed, high-quality communication needs. Cognitive radio system provides a new scheme for dynamic allocation and use of spectrum. Since it was put forward in 1999, it has received extensive attention and research. As the core technology of cognitive radio, spectrum detection algorithm can identify "spectrum holes" in real-time and accurately, so as to provide access opportunities for unauthorized users and improve the efficiency of spectrum utilization. The sparse characteristics of the signal in some domains can concentrate the energy of the signal on the one hand, on the other hand, it can also effectively reduce the sampling rate and difficulty, corresponding to the requirements of high detection probability and dynamic real-time detection in the spectrum detection algorithm respectively. Therefore, this paper attempts to combine sparse characteristics with detection algorithms, focusing on the detection algorithm based on frequency domain sparse characteristics, and analyzes the detection performance of the algorithm. Firstly, the paper introduces the research background and significance of the subject, analyzes the advantages and disadvantages of the existing detection algorithms, and leads to the detection algorithm based on the maximum power spectrum estimation and the detection algorithm based on MWC-SBL system. Then the basic theories used in these two algorithms are summarized, including power spectrum estimation, random bandwidth converter, sparse Bayesian learning, and the corresponding simulation results are given. Secondly, the detection algorithm based on the maximum value of power spectrum estimation is deeply analyzed, and the statistical decision variable is defined as the maximum value of signal power spectrum estimation. After analyzing the statistical characteristics of statistical decision variables, such as mean value, variance and correlation, the statistical decision variables under H _ 0 and H _ 1 assumptions are modeled by chi-square distribution, and the theoretical expressions of detection probability, false alarm probability and decision threshold are derived. In the simulation part, the correctness of the chi-square distribution modeling and the advantage of selecting the estimated maximum value as the statistical decision variable are verified. The effects of the length of sampling data and the number of power spectrum segments on the detection performance are analyzed, and compared with the energy detection algorithm. The simulation results show that different window functions have an effect on the accuracy of detection probability and chi-square distribution modeling, and they show opposite trend. Although the algorithm based on power spectrum estimation has unique advantages in the detection of sparse narrow-band signals, due to the limitation of front-end sampling, sparse multi-band signals can not meet the requirements of timely detection. Finally, the detection algorithm based on MWC-SBL system is studied. On the basis of compressed sampling, sparse Bayesian learning algorithm is used to fully mine the information of observed data, which provides another scheme for broadband detection. The support set is defined as a statistical decision variable. Through the detection of the support set, the existence of each frequency band signal is judged. After the algorithm is described, the statistical expressions of detection probability and false alarm probability are defined according to the relationship between the restored support set and the original support set. The following simulation focuses on the detection performance of fixed position single signal and multi-signal, the detection performance of random position single signal and multi-signal, the improvement of detection performance by different matching principles and the mean square error (MSE), and so on. And compared with the detection algorithm based on MWC-OMP. Simulation results show that the MWC-SBL-based detection algorithm has more advantages than the MWC-OMP detection algorithm in terms of detection probability and other performance indicators.
【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】：2017
【分类号】：TN925

【参考文献】