基于神经网络的干扰识别算法研究

论文总字数：26028字

摘 要

随着通信技术的发展，无线通信技术的应用越来越广泛。而近年来，各种人为或非人为干扰问题日益突出，给军事和民用通信都带来极大的困扰。干扰信号多种多样，但目前没有一个有效的方法去抵御所有类型的干扰信号。可是我们如果能够有效地识别干扰信号类型就能有所针对进行干扰抑制，从而实现有效的抗干扰通信。而随着神经网络在信息识别领域的飞速发展，基于神经网络的识别算法相较于其他相关算法有着更为高效与高准确率的优势，因此选择利用神经网络来设计干扰识别算法。

首先介绍四种典型的干扰信号，并针对各种干扰信号的差别进行了研究，提取其典型特征信息。

接着，基于多层感知器和卷积神经网络，设计了两种基于神经网络的干扰信号识别模型，其中在模型训练与模型测试时，多层感知器的输入为人工提取的特征；卷积神经网络的输入为干扰信号复基带形式的实部与虚部。

最后，通过仿真实验，验证了本文设计算法的有效性。仿真主要包括在训练样本数目不同、干扰噪声比分布范围不同和对输入特征做不同的预处理时，观察干扰识别正确率的变化。结果表明，基于多层感知器的干扰识别算法和基于卷积神经网络的干扰识别算法的识别结果都较好，同时后者的性能要略优于前者。

关键词：干扰识别，多层感知器，卷积神经网络，反向传播

Abstract

With the development of communication technology, the application of wireless communication technology has become more and more extensive. In recent years, various artificial or non-human interference problems have become increasingly prominent, causing great troubles for both military and civilian communications. There are many different types of interfering signals, but there is currently no effective way to defend against all types of interfering signals. However, if we can effectively identify the type of interference signal, we can target interference suppression to achieve effective anti-jamming communication. With the rapid development of neural networks in the field of information recognition, neural network-based recognition algorithms have the advantages of higher efficiency and higher accuracy than other related algorithms. Therefore, neural networks are chosen to design interference recognition algorithms.

Firstly, four typical interference signals are introduced. The differences of various interference signals are studied and their typical characteristic information is extracted.

Then, based on multi-layer perceptron and convolutional neural network, two neural network-based interference signal recognition models are designed. In model training and model testing, the input of multi-layer perceptron is artificially extracted. The input to the convolutional neural network is the real and imaginary parts of the complex signal baseband.

Finally, the effectiveness of the proposed algorithm is verified by simulation experiments. The simulation mainly includes observing the change of the correct rate of interference recognition when the number of training samples is different, the jamming noise ratio distribution range is different, and the input features are differently preprocessed. The results show that the multi-layer perceptron-based interference recognition algorithm and the convolutional neural network-based interference recognition algorithm have better recognition results, while the latter performance is slightly better than the former.

KEY WORDS: Interference recognition, multi-layer perceptron, convolutional neural network, back propagation

目 录

第一章 绪论 1

1.1研究背景 1

1.2研究现状 1

1.3本文的大致结构 2

第二章 常见干扰信号及其特征提取 4

2.1常见干扰信号的复基带形式 4

2.1.1单音干扰 4

2.1.2多音干扰 5

2.1.3部分频带干扰 5

2.1.4线性扫频干扰 6

2.2干扰信号能量归一化 7

2.3干扰信号的特征提取 7

2.3.1单频能量聚集度 7

2.3.2频域矩峰度系数 9

2.3.3频域矩偏度系数 10

2.3.4频域参数 11

2.4本章小结 12

第三章 基于多层感知器的干扰识别算法 13

3.1多层感知器简介 13

3.1.1多层感知器结构 13

3.1.2反向传播算法 14

3.2 基于多层感知器的干扰识别算法 15

3.2.1 多层感知器构建 15

3.2.2仿真实验 16

3.3本章小结 19

第四章 基于卷积神经网络的干扰识别算法 20

4.1卷积神经网络简介 20

4.1.1 卷积 21

4.1.2 池化 21

4.1.3 全连接层 22

4.1.4 训练卷积神经网络模型 23

4.2 基于卷积神经网络的干扰识别算法 24

4.2.1卷积神经网络的输入 25

4.2.2卷积神经网络的基本组成 25

4.2.3 系统架构 26

4.3 仿真实验 29

4.3.1训练样本数目不同时的总体平均识别率 29

4.3.2训练样本分布不同时的总体平均识别率 29

4.4卷积神经网络与多层感知器比较 30

4.5本章小结 31

第五章 总结与展望 32

5.1论文总结 32

5.2对未来工作的展望 32

参考文献 33

致 谢 36

剩余内容已隐藏，请支付后下载全文，论文总字数：26028字