基于一维CNN的多入多出OSTBC信号协作调制识别  

Cooperative modulation recognition based on one-dimensional convolutional neural network for MIMO-OSTBC signal

作　　者：安泽亮 张天骐[1] 马宝泽 邓盼 徐雨晴 AN Zeliang;ZHANG Tianqi;MA Baoze;DENG Pan;XU Yuqing(School of Communication and Information Engineering,Chongqing University of Posts and Telecommunications,Chongqing 400065,China;School of Computer Science and Technology,Chongqing University of Posts and Telecommunications,Chongqing 400065,China)

机构地区：[1]重庆邮电大学通信与信息工程学院,重庆400065 [2]重庆邮电大学计算机科学与技术学院,重庆400065

出　　处：《通信学报》2021年第7期84-94,共11页Journal on Communications

基　　金：国家自然科学基金资助项目(No.61671095,No.61702065,No.61701067)。

摘　　要：为识别多入多出正交空时分组码(MIMO-OSTBC)系统所采用的调制样式,提出了一种基于一维卷积神经网络(1D-CNN)的协作调制识别算法。首先,采用迫零盲均衡来提升不同调制信号间区分度,并选用天然无损的同相正交(I/Q)信号作为浅层特征;然后,设计并训练基于1D-CNN的识别模型,从浅层特征中提取深层特征;最后,采用投票决策和置信度决策融合策略,提升多天线接收端协作识别精度。实验结果表明,所提算法能有效识别{BPSK,4PSK,8PSK,16QAM,4PAM}5种调制方式,当信噪比大于或等于−2 dB时,识别精度可达100%。To recognize the modulation style adopted in multiple-input-multiple-output orthogonal space-time block code(MIMO-OSTBC) systems, a cooperative modulation recognition algorithm based on the one-dimensional convolutional neural network(1 D-CNN) was proposed. With the lossless I/Q signal selected as shallow features, the zero-forcing blind equalization was first leveraged to improve the discrimination of different modulation signals. Then the 1 D-CNN recognition model was devised and trained to extract deep features from shallow ones. Later, two decision fusion strategies of voting-based and confidence-based were leveraged in the multiple-antenna receiver to improve recognition accuracy. Experimental results show that the proposed algorithm can effectively recognize five modulation types {BPSK, 4 PSK,8 PSK,16 QAM,4 PAM}, with a 100% recognition accuracy when the signal-to-noise is equal or greater than-2 dB.

关 键 词：调制识别 多入多出正交空时分组码 迫零盲均衡 一维卷积神经网络 决策融合 

分 类 号：TN911.7[电子电信—通信与信息系统]