基于深度学习的无线通信信号检测与识别研究  

Research on Wireless Communication Signal Detection and Recognition Based on Deep Learning

作　　者：梁晶 杨晶晶[1,2] 黄铭[1,2] LIANG Jing;YANG Jingjing;HUANG Ming(School of Information Science&Engineering,Yunnan University,Kunming 650091,China;Wireless Innovation Laboratory,Yunnan University,Kunming 650091,China)

机构地区：[1]云南大学信息学院,云南昆明650091 [2]云南大学无线创新实验室,云南昆明650091

出　　处：《无线电工程》2023年第3期611-618,共8页Radio Engineering

基　　金：国家自然科学基金(61863035,61963037)。

摘　　要：现有的无线电监测系统中信号检测和信号识别是分开研究的,缺乏将二者结合起来实现智能监测的文献报道。基于此,实现了分别将卷积注意力模块(Convolution Block Attention Module,CBAM)与长短期记忆(Long Short-Term Memory,LSTM)网络和时间卷积网络(Temporal Convolutional Network,TCN)结合起来通过二分类进行信号检测,以及基于多分类进行信号识别的级联方法,分类模型为CBAMLSTM和CBAMTCN。对于RadioML2016.10a数据集,仿真结果表明,与能量检测(Energy Detection,ED)、深度神经网络(Deep Neural Network,DNN)、卷积神经网络(Convolutional Neural Network,CNN)、LSTM和TCN比较,二分类模型在低信噪比下的检测性能有所改善;在高信噪比下多分类模型比基线模型的识别准确度提高5%~9%。表明分类算法在信号类型和噪声特性已知的前提下,可用于无线信号检测和识别。Signal detection and signal identification are studied separately in existing radio monitoring systems,and there is a lack of literature reporting on combining the two to achieve intelligent monitoring.Based on this,a cascade method of combining Convolutional Block Attention Module(CBAM)with Long Short-Term Memory(LSTM)and Temporal Convolutional Network(TCN)for signal detection through binary classification and multi-classification for signal recognition,respectively,is implemented.Among them,the classification models are CBAMLSTM and CBAMTCN.For the RadioML2016.10a dataset,simulation results show improved detection performance of the binary classification model at low signal-to-noise ratios when compared with Energy Detection(ED),Deep Neural Network(DNN),Convolutional Neural Network(CNN),LSTM and TCN.The multi-classification model improves the recognition accuracy by 5%~9%over the baseline model at high signal-to-noise ratios.This shows that the classification algorithm can be used for wireless signal detection and identification provided that the signal type and noise characteristics are known.

关 键 词：信号检测 信号识别 卷积注意力模块 长短期记忆网络 时间卷积网络 

分 类 号：TN911.7[电子电信—通信与信息系统]