利用稀疏自编码器的调制样式识别算法  

Modulation Recognition Algorithm Based on Sparse Auto-Encoder

作　　者：杨安锋 赵知劲[1] 陈颖[1] YANG An-feng;ZHAO Zhi-jin;CHEN Ying(School of Communication Engineering,Hangzhou Dianzi University,Hangzhou,Zhejiang 310018,China)

机构地区：[1]杭州电子科技大学通信工程学院,浙江杭州310018

出　　处：《信号处理》2018年第7期833-842,共10页Journal of Signal Processing

基　　金："十二五"国防预研项目(41001010401)

摘　　要：针对传统调制样式识别算法对复杂幅相信号识别率低,所需特征参数多的问题,提出一种利用稀疏自编码器的调制样式识别算法。将稀疏自编码器和Softmax分类器级联构成识别系统,将两个高阶累积量特征参数的格雷码编码构成系统输入矢量,利用稀疏自编码器提取的深度特征作为Softmax分类器输入。在系统训练阶段,先训练稀疏自编码器,然后利用有监督算法训练分类层,接着利用有监督算法进行整体优化。对BPSK、QPSK、8PSK、16QAM、32QAM、16APSK和32APSK等7种信号识别的仿真结果表明,在低信噪比时本文算法的平均正确识别率比对比算法高。Aiming at the problem that traditional recognition algorithms for complex amplitude-phase modulation signals have low identification rate and large number of required feature parameters,a modulation type recognition algorithm using sparse auto-encoder is proposed.The sparse auto-encoder and Softmax classifier are cascaded to form a recognition system.The Gray code encoding of two high-order cumulant feature parameters constitutes the system input vector,and the depth feature extracted by the sparse auto-encoder is used as the input of the Softmax classifier.In the system training phase,sparse auto-encoder is trained first,then supervised algorithm is used to train the classification layer,and then supervised algorithm is used for overall optimization.The simulation results of seven kinds of signal recognition such as BPSK,QPSK,8PSK,16QAM,32QAM,16APSK and 32APSK show that the average correct recognition rate of this algorithm is higher than that of the comparison algorithm at low SNR.

关 键 词：调制样式识别 高阶累积量 格雷码 深度学习 稀疏自编码器 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构] TN713[自动化与计算机技术—计算机科学与技术]