基于深度学习的无线通信调制方式识别技术研究

论文总字数：22621字

摘 要

本文基于深度学习网络，以差分星座轨迹图作为特征信息可视化的媒介，实现了对六种无线通信的调制方式的识别。

本文介绍了无线通信中最具代表性的六种调制方式，并在MATLAB中对其进行了不同信噪比下的仿真；分析了差分星座轨迹图的生成原理，生成了不同的频偏及信噪比下的差分星座轨迹图，并进一步通过聚类算法提取出了特征信息矩阵；介绍了深度学习网络的普遍原理及本课题采用的BP分类算法。

课题基于多组频偏和信噪比下的特征矩阵训练集，实现了24组数据的测试集1与30组数据的测试集2的较高识别率，并分析了测试集中差分间隔、接收端频偏与信噪比的取值不同所导致的对分类性能的不同影响。

关键词：无线通信、数字调制、差分星座轨迹图、深度学习、模式识别、卷积神经网络、BP神经网络

Abstract

In this theis, based on deep learning network, differential constellation trajectory map is used as the medium of feature information visualization to realize the recognition of six modulation modes of wireless communication.

This paper introduces six most representative modulation modes in wireless communication, and simulates them with different SNR in MATLAB. The generation principle of differential constellation trajectory map is analyzed, and the differential constellation trajectory map with different frequency deviation and signal-to-noise ratio is generated, and the feature information matrix is further extracted by clustering algorithm. This paper introduces the general principle of deep learning network and BP classification algorithm.

Based on the training set of characteristic matrix under multi-group frequency deviation and signal-to-noise ratio, the training set with a total of 30 sets of data has a high recognition rate in the range of 10dB to 30dB and every 5dB is an interval of signal-to-noise ratio.

KEY WORDS: wireless communication, digital modulation, differential constellation trajectory, deep learning, pattern recognition,Convolutional neural network, BP neural

目 录

摘要.....….….….….….….….….….….….….….….….….….….….….….….….…...….........Ⅰ

Abstract.….….….….….….….….….….….….….….….….….….….….….….….….......…...Ⅰ

第一章 绪论.….….….….….….….….….….….….….….….….….….….….….….….…...…1

1.1 课题背景.….….….….….….….….….….….….….….….….….….…...….….…..........1

1.2 数字通信系统简介.….….….….….….….….….….….….….….….…...….….….........1

1.3 深度学习的发展简介.….….….….….….….….….….….….….….….…...….…......…2

1.4 本文的研究目的和主要研究内容.….….….….….….….….….….….….….........……3

第二章 多种数字通信调制方式的仿真与实验.….….….….….….….….….….….….....…..5

2.1 数字通信的调制方式原理简介.….….….….….….….….….….….….….......….....….5

2.1.1 BPSK调制、QPSK调制与8PSK调制.….….….….….….….…....……............5

2.1.2 16QAM调制与64QAM调制.….….….….….….….….….….….…..……..........6

2.1.3 MSK调制.….….….….….….….….….….….….….….….….….….………..…...7

2.2 基于MATLAB的数字通信系统仿真.….….….….….….….….….….….….…......….8

2.2.1 信源模块.….….….….….….….….….….….….….….….….….….….……….....8

2.2.2 调制器模块.….….….….….….….….….….….….….….….….….………...…....8

2.2.3 信道模块与噪声源模块.….….….….….….….….….….….….….………...…....8

2.2.4 解调器模块.….….….….….….….….….….….….….….….….….………...…....8

2.2.5 差分星座轨迹图生成模块.….….….….….….….….….….….….………...….....8

2.3 六种调制方式的星座图.….….….….….….….….….….….….….….….….…..............8

1. 差分星座轨迹图的仿真与实验生成.….….….….….….….….….….….….….….....11

3.1差分星座轨迹图的生成原理.….….….….….….….….….….….….….….….….......…11

3.2差分星座轨迹图的实例分析....….….….….….….…….…….…….…….…….......…..12

3.2.1不同频偏下的差分星座轨迹图….….….….….….….….….….…..….……........12

3.2.2不同信噪比下的差分星座轨迹图.….….….….….….….….….….….….……....14

3.3差分星座轨迹图的差分星座轨迹图的特征矩阵计算….….….….…....….……...…...15

第四章 基于深度学习的差分星座轨迹图分类识别.….….….….….….….….….….….…...18

4.1神经网络简介.….….….….….….….….….….….….….….….….….….….….…..........18

4.2 BP神经网络.….….….….….….….….….….….….….….….….….….….….….......….19

4.2.1 BP神经网络简介.….….….….….….………...….….….….….….….….….…....19

4.2.2正向传播.….….….….….….….….….….….….….….….….….….….……..…...20

4.2.3反向传播.….….….….….….….….….………...….….….….….….….….…...….20

4.2.4损失函数.….….….….….….….….….….………...….….….….….….….…..…..20

4.2.5激活函数.….….….….….….….….….….….………...….….….….….….…..…..21

4.2.6优化器.….….….….….….….….….….….….………...….….….….….….…..….21

4.3 BP网络分类识别的结果.….….….….….….….….….….….….….….…..........…...….21

4.3.1 不同差分间隔和频偏值下的分类性能.….….…..…....….….…….….….….….22

4.3.2 不同信噪比下的分类性能.….….….….….….….….….….….….……….….….23

第五章 总结与展望.….….….….….….….….….….….….….….….….….….….….…...…...26

5.1成果总结.….….….….….….….….….….….….….….….….….….….….….….............26

5.2成果展望….….….….….….….….….….….….….….….….….….….….….…..............27

参考文献（References）.….….….….….….….….….….….….….….….….….….….…..….28

致谢.….….….….….….….….….….….….….….….….….….….….….….….….….….….….30

剩余内容已隐藏，请支付后下载全文，论文总字数：22621字