基于机器学习的分布式频谱监测关键技术研究

RESEARCH ON KEY TECHNOLOGIES OF DISTRIBUTED SPECTRUM MONITORING BASED ON MACHINE LEARNING

张猛

11649162

硕士

电子与通信工程

贡毅

2018-06-07

2018-07-09

哈尔滨工业大学

深圳

无线电通信业务的高速发展导致了对频谱资源的需求与日俱增，致使电磁频谱资源供需日益紧张，同时也对电磁频谱管理提出了巨大挑战。为了更好地管理电磁频谱资源，如何有效地进行频谱监测就是首先需要解决的问题。在频谱监测的众多任务中，调制识别技术起着关键性的作用。本文紧密结合通信理论以及机器学习算法，重点研究了调制识别技术。文章首先从频谱管理入手，对频谱监测的概念，监测内容等做了介绍。同时，从通信信号调制理论出发，文章介绍了常见模拟调制信号以及数字调制信号的基本调制原理和特性。其次，文章总结了调制识别算法的发展以及研究现状，着重介绍了信号的高阶统计量特征，同时也介绍了目前在分类识别任务中应用广泛的深度学习算法。然后，文章基于对调制信号特性的分析，将高阶统计量以及信号的原始In-phase and quadrature(I/Q)数据作为信号特征，将深度学习算法作为分类器对通信信号进行调制识别，通过设计仿真实验证明了该方法可以取得了良好的识别性能。最后，文章针对目前应用广泛的M-QAM信号，提出了基于Rihaczek分布和 Hough变换的 M-QAM调制识别方法， 将信号转化为时频特征图进行分类，在低信噪比 (-10dB)下依然能取得 90%的识别准确率 。

The rapid development of radio communication services has led to an increasing demand for spectrum resources, resulting in increasingly tight supply and demand of spectrum resources. At the same time, it also poses great challenges to the spectrum management. In order to better manage the spectrum resources, how to effectively monitor the spectrum is the first problem to solve. Modulation recognition technology plays a key role in many tasks of spectrum monitoring. Combining the communication theory and machine learning algorithm, the research of this paper focuses on the modulation recognition technology.Starting with spectrum management this paper introduces the concept and content of spectrum monitoring. Meanwhile, starting from the modulation theory of communication signals, the basic modulation principles and characteristics of common analog modulation signals and digital modulation signals are introduced.Then, the article summarizes the research status of the modulation recognition algorithm and the high-order statistical characteristics of the communication signal. Meanwhile, this work introduces the deep learning algorithm that has been widely applied in classification task.Based on the analysis of the characteristics of the modulated signal, the high-order statistics and the signal's raw In-phase and quadrature (I/Q) data are used as the signal features. Then deep learning methods are used to identify the modulation type of the communication signal. The simulation experiment shows that the method can achieve good recognition performance.Finally, for the widely used M-QAM signals, this paper proposes a M-QAM modulation recognition method based on Rihaczek distribution and Hough transform. The signals are converted to the time frequency characteristic images for further classification. This method can still obtain 90% recognition accuracy under low signal to noise ratio (-10dB).

频谱管理 频谱监测 机器学习 高阶统计量 I/Q数据 Rihaczek分布 Hough变换 调制识别

spectrum management spectrum monitoring machine learning higher order statistics I/Q data Rihaczek distribution Hough transform modulation recognition

中文

联合培养

南方科技大学

张猛. 基于机器学习的分布式频谱监测关键技术研究[D]. 深圳. 哈尔滨工业大学,2018.