摘要
为解决传统信号识别方法对直接序列扩频信号识别率低、智能化程度不高等问题,提出一种基于循环谱特征提取的直扩信号智能识别算法。在深入分析直扩信号循环谱特征基础上,以直扩信号循环谱独特的稀疏特性为依据,设计一种稀疏滤波-卷积神经网络模型对提取的循环谱等高线图进行识别,采用无监督预训练和有监督训练微调的方式对网络参数进行更新,提升网络对信号整体特征的表达能力和对小数据量信号样本的学习能力。仿真结果表明:本文算法能够有效识别直接序列扩频信号,在不低于-10 dB高斯噪声条件下对采用正交相移键控调制的直扩信号识别准确率达到98%以上;在混合其他调制信号条件下,相较于常见的几种深度学习算法,本文提出的算法具有更高的识别准确率和鲁棒性。
As for the problems of low recognition accuracy,poor intelligence and confusion with ordinary modulation signals in traditional signal recognition methods,an algorithm of intelligent recognition of direct sequence spread spectrum signals extracted by cyclic spectrum feature is proposed.On the basis of in-depth analysis of the cyclic spectrum characteristics of DSSS signals,and based on the unique sparse characteristic of the cyclic spectrum of the direct sequence spread spectrum signal,a sparse filtering convolution neural network model is designed to identify the extracted cyclic spectrum contour map,the unsupervised pre-training and supervised training fine-tuning methods are used to update the network parameters,so as to improve the network′s ability to express the overall characteristics of the signal and its ability to learn small amount of signal samples.The simulation results show that the algorithm can effectively recognize the direct sequence spread spectrum signal,and the recognition accuracy of the direct sequence spread spectrum signal using quadrature phase shift keying modulation can reach 98%or above under the condition of no less than-10 dB Gaussian noise.Compared with the common deep learning algorithm,the proposed algorithm has higher recognition accuracy and robustness when mixed with other modulated signals.
作者
王源
冯永新
钱博
WANG Yuan;FENG Yongxin;QIAN Bo(Shenyang Ligong University,Shenyang 110159,China)
出处
《沈阳理工大学学报》
CAS
2023年第4期31-39,共9页
Journal of Shenyang Ligong University
基金
国家自然科学基金项目(61971291)
中央引导地方科技发展项目(2022020128-JH6/1001)
辽宁省教育厅科学研究项目(LJKZ0242)。
关键词
信号识别
直接序列扩频
循环谱
稀疏滤波-卷积神经网络
深度学习
signal recognition
direct sequence spread spectrum
cyclic spectrum
sparse fil-ter-convolution neural network
deep learning
