【作者】 周凯；

【导师】 王平；

【作者基本信息】 西南交通大学 ， 信息与通信工程， 2022， 硕士

【摘要】 在非协作跳频通信系统中,对接收到的干扰信号与跳频信号进行有效检测识别,估计信号参数,是侦查接收机的重要能力。本文利用深度学习神经网络对跳频信号进行检测识别,获取跳频信号关键特征参数信息,同时利用数据聚类分析技术,对检测识别到的跳频信号数据样本进行有效分选。本文主要内容如下:1)介绍跳频信号模型及主要参数,并简要分析复杂电磁环境下存在的各类干扰信号;设计跳频信号检测识别系统,并分析其主要构成部分;针对数据预处理,分析常用的几种时频分析方法;分析复杂电磁环境下的干扰信号,将单音干扰、扫频干扰、突发干扰等信号用于跳频信号中的干扰信号识别研究,并采用Gabor特征变换获取跳频信号时频图做为检测识别系统目标检测部分数据输入。2)分析YOLOv5网络架构的输入端、Backbone层、Neck层、Prediction四层结构。通过调整网络锚定框、改进损失函数、增加网络Neck层操作等方式对干扰信号检测识别、跳频信号调制方式检测识别进行改进。实验结果表明,干扰信号检测识别概率在94%以上,均方误差在0.014以下;各调制信号检测识别概率为98%以上,至少提升4.5%,均方误差在0.01以下。3)提出自适应k值的Agnes层次聚类算法,对检测识别的输出参数进行聚类以实现信号分选,并通过分析ARI、AMI、V-调和平均、FMI以及SC值衡量算法性能。实验运行结果表明:采用改进后的Agnes算法进行聚类,结果与真实情况十分吻合,同时与原Agnes算法相比运行速度提升效果明显。更多还原

【Abstract】 In the non-cooperative frequency hopping communication system,it is an important ability of the detection receiver to effectively detect and identify the received interference signals and the frequency hopping signal,and to estimate the signal parameters.In this thesis,detection of frequency hopping signals based on deep learning neural networks can obtain the key feature parameter information,and then the data clustering and analysis technology is used to effectively classify the data samples.The main contents of this thesis are as follows:1)Frequency hopping signal model and main parameters are mainly introduced,and various interference signals in complex electromagnetic environment are briefly analyzed.Detection and identification system for frequency hopping signal is designed,which main framework are analyzed.The Time frequency characteristics of the interference signals in the complex electromagnetic environment are analyzed,including uni-tone interference,sweep interference and burst interference.Time-frequency transformation algorithms of signals are analyzed.The experimental results show that the performance of Gabor feature transform algorithm is better than the others.2)The structure of YOLOv5 network is analyzed.Detection and identification system of interference signal and the frequency hopping signal modulation mode are improved by adjusting parameter of the network anchor box,improving the loss function and increasing operation of the network Neck layer.The experimental results show that,the probability of detection and identification of interference signal is above 94 percent,and the mean square error is below 0.014;the probability of detection and identification of each modulated signal is above 98 percent,which increasing by at least 4.5 percent,and the mean square error is below 0.01.3)An adaptive k-value Agnes hierarchical clustering algorithm is proposed for the clustering of the parameters,which for signal classification.And the value of ARI,AMI,V-Harmonic Mean,FMI and SC is analyzed.Experimental results show that the clustering data which used the improved Agnes algorithm are consistent with the real situation,and compared with the original Agnes algorithm,the operating speed is improved obviously更多还原