【作者】 贺新颖；

【导师】 曾志民；

【作者基本信息】 北京邮电大学 ， 通信与信息系统， 2009， 博士

【摘要】 无线频谱资源紧缺与分配方式缺乏灵活性是限制当今无线通信发展的两大因素。认知无线电技术作为一种智能频谱共享技术,可显著提高频谱利用率,实现不可再生频谱资源的再利用,为解决如何在有限频谱资源条件下提高频谱使用率这一无线通信难题开辟了一条新的途径。研究认知无线电关键技术,对于推动认知无线电的发展和应用,意义重大。目前认知无线电关键技术的内容众多,研究方法繁杂,尚没有形成统一的理论研究体系。认知无线电技术本身强调通信设备的智能性,要求设备能通过对外界环境信息的感知和学习,实时改变其操作参数从而充分利用空闲频谱,提高频谱利用率,这使得利用人工智能领域中的机器学习理论研究认知无线电关键技术成为必要与可能。支持向量机是近年来比较流行的机器学习方法,具有相对优良的性能指标。本文从机器学习的角度,利用支持向量机方法,对认知无线电的几个关键技术进行了研究,本文的研究重点是认知无线电关键技术中的频谱感知和动态频谱接入,本文的主要工作包括：1.本文在第二章重新阐述了认知无线电关键技术、认知无线电网络、认知网络的定义,系统总结并详细介绍了认知无线电技术,尤其是关键技术的国内外研究、发展和应用现状；通过讨论与分析,明确了认知无线电关键技术研究中存在的问题和进一步的研究内容与方向,同时也充分体现了提出“基于支持向量机的认知无线电若干关键技术研究”这一课题的必要性和重要性。2.本文在第四章针对认知无线电频谱感知技术中信号调制方式识别的重要性,提出一种新型的认知无线电调制信号识别算法,依据调制信号的循环谱特性,结合支持向量机与隐马尔科夫模型与用于信号识别的各自优点,构建两级分类器,对调制信号进行识别。算法对信号调制方式的识别性能良好,且在低信噪比下仍具有较高的正确识别率。3.本文在第五章提出一种基于概率密度估计的动态频谱接入算法,通过由支持向量机拟合出的授权频段空闲时长的概率密度对信道状态进行预测评估,认知无线电用户根据信道状况选择接入。该算法可以进行自适应调整,具有良好的实用性与灵活性。仿真结果表明,所提出的算法可以有效降低信道的冲撞率,同时提高认知无线电用户的吞吐量和服务质量。4.本文在第六章提出了“无线通信网络中时域频谱利用概率密度拟合研究”这一子课题的重要意义和研究价值,对时域频谱利用概率分布建模的方法和步骤做了详细介绍,在仿真环境下按照传统统计学方法对信道空闲时长的概率密度进行拟合分析,同时也使用了基于支持向量机的密度估计方法。仿真结果说明提出的建模方法切实有效,并且在统计意义上得出了IEEE802.11g WLAN环境下传输UDP流时,信道的空闲时长服从Pareto分布的这一重要结论。更多还原

【Abstract】 The scarceness of wireless spectrum and the lack of flexibility of allocation methods are two major factors which restrict the development of current wireless communications. As a revolutionary smart spectrum sharing technology, Cognitive Radio (CR) can significantly improve the spectrum utilization and receives more and more interest within these years. It’s very important to research on key technologies of cognitive radio for promoting the development and application of cognitive radio. There are many different types of the key technologies in cognition radio at the moment, as well as research methods, which are still not formed the unified fundamental research system. Cognitive radio technology emphasize the intelligence of communications equipment, which make it possible to research the key technologies of cognitive radio based on machine learning theoretical. As a machine learning method, Support Vector Machine (SVM) is very popular in recent years, has a relatively good performance.In this dissertation, some key technologies of cognitive radio are researched based on support vector machine. The research focused on spectrum sensing and dynamic spectrum access (DSA), the dissertation’s main contents include:1. In Chapter 2, the key technology of cognitive radio, cognitive radio networks and cognitive networks are redefined, and a systematic summary and detailed information about cognitive radio technology are introduced, especially its key technology research, development and applications. By discussing and analyzing, the problems and further research direction of the key technology in cognitive radio are definable, which fully reflects the necessity and importance of the proposed subject that is research on some key technologies of cognitive radio based on support vector machine.2. In Chapter 4, a novel approach to signal classification is proposed for cognitive radio. Combining the spectral cyclostationary features, embed SVM into the framework of Hidden Markov Mode (HMM) to construct a hybrid HMM/SVM classifier for signal recognition. The simulation results show that the high performance and robustness of the proposed approach, even in low SNR. Compared to the conventional methods, the proposed approach is robust, has a rather higher recognition rate of signals.3. In Chapter 5, a dynamic spectrum access algorithm based on probability density estimation is proposed to estimate the probability density of spectrum idle duration with support vector machines and evaluates the channel states, and cognitive radio users access the channel according to the states. This practicable and flexible algorithm can be adjusted adaptively. Simulation shows that the proposed algorithm significantly reduces disruptions to primary users and improve the throughout as well as the quality of service of cognitive radio users.4. In Chapter 6, a very important and valuable sub-topic is proposed, namely research on fitting probability density of spectrum idle duration in wireless communication networks. And then modeling methods and steps are described in detail.In the simulation environment, the probability density of spectrum idle duration is estimated with conventional parameter estimation and support vector machines. Simulation results show that the proposed modeling methods are effective and robust.更多还原