利用卷积神经网络和协方差的协作频谱感知算法  

Cooperative Spectrum Sensing Algorithm Using Convolutional Neural Networks and Covariance

作　　者：鲁华超 赵知劲[1,2] 尚俊娜[1] 戴绍港[1] Lu Huachao;Zhao Zhijin;Shang Junna;Dai Shaogang(School of Communication Engineering,Hangzhou Dianzi University,Hangzhou,Zhejiang 310018,China;State Key Lab of Information Control Technology in Communication System,The 36 th Research Institute of China Electronics Technology roup Corporation,Jiaxing,Zhejiang 314001,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 [2]中国电子科技集团第36研究所通信系统信息控制技术国家级重点实验室,浙江嘉兴314001

出　　处：《信号处理》2019年第10期1700-1707,共8页Journal of Signal Processing

摘　　要：针对基于信号协方差矩阵的频谱感知算法门限难于准确得到及没有充分利用原始信号信息等问题,提出了基于卷积神经网络和协方差矩阵的协作频谱感知算法。首先将接收的I、Q两路正交信号的归一化协方差矩阵组成双通道输入矩阵,然后使用卷积神经网络直接提取协方差矩阵的特征信息,并进行训练得到分类器,最后使用训练好的模型进行频谱感知。仿真结果表明,本文所提出的频谱感知算法优于对比算法,在信噪比为-13 dB、40个次用户协作感知时,本文算法虚警概率低于0.1,检测概率达到0.9以上。A cooperative spectrum sensing algorithm based on convolutional neural network and covariance matrix is proposed to solve the problems that the threshold value of spectrum sensing algorithm based on signal covariance matrix is difficult to get accurately and the raw signal information is not fully utilized.Firstly,the normalized covariance matrix of the received I and Q orthogonal signals is formed into a two-channel input matrix.Then,the characteristic information of the covariance matrix is extracted by using convolutional neural network to train the classifier.Finally,the trained model is used for spectrum sensing.The simulation results show that the proposed spectrum sensing algorithm is superior to the contrast algorithms.The false alarm probability of this algorithm is less than 0.1 and the detection probability is above 0.9,when the SNR is -13 dB and 40 secondary users cooperate in sensing.

关 键 词：卷积神经网络 协方差 协作频谱感知 

分 类 号：TN911[电子电信—通信与信息系统]