摘要
针对现有的基于深度学习的调制识别算法训练速度慢、识别率不高和识别调制类型少的问题,提出了一种基于循环谱和局部感受野超限学习机(ELM-LRF)的调制识别算法。首先,提取调制信号的循环谱图并对其进行归一化预处理,然后将处理后的循环谱分成训练集和测试集,我们用训练集训练ELM-LRF,最后用测试集对网络进行测试。对11种数字调制和模拟调制信号进行分类识别,实验结果表明,在信噪比大于0 dB时,本文算法的总体识别率超过了95%,同时相比于基于传统深度学习的调制识别算法,训练时间大大减少,验证了ELM-LRF是一种高效快速深度学习方法,具有较大的研究价值。
In order to solve the problems such as slow training speed,low recognition rate,few types of current modulation recognition algorithm based on deep learning,this paper proposes a modulation recognition algorithm based on cyclic spectrum and local receptive fields based extreme learning machine(ELM-LRF).Firstly,the cyclic spectrum of modulation signal is extracted and pre-normalized,then the processed cyclic spectrums are divided into training set and testing set.We use training set to train ELM-LRF and finally use the training set to test the network.The algorithm identifies 11 types of digital modulation and analog moduation singnals.The experiment results show that the overall recognition rate is over 95%when the signal-to-noise ratio is more than 0 dB.At the same time,compared with the modulation recognition algorithms based on traditional deep learning,the training time is greatly reduced,which proves that ELM-LRF is an efficient and fast deep learning method and has great research value.
作者
李晨
杨俊安
刘辉
LI Chen;YANG Jun-an;LIU Hui(National University of Defense Technology,Hefei 230037,China)
出处
《舰船电子对抗》
2020年第1期52-57,95,共7页
Shipboard Electronic Countermeasure
关键词
循环谱
极限学习机
局部感受野
调制识别
cyclic spectrum
extreme learning machine
local receptive field
modulation recognition