| 摘要: |
| 在实际调制过程中，无线电波传输多径及衰落引起的符号间干扰和信号接收端的载波频偏会造成星座图难以识别。针对这一问题，提出了一种基于星座图恢复和卷积神经网络的多进制相位调制信号识别算法。首先，设定相邻采样点距离和相位角的阈值以筛除发生符号间干扰时的采样点，保留剩余的有效采样点并形成聚类组；然后，通过旋转相邻聚类组抵消载波频偏带来的影响，实现星座图的恢复；最后，利用卷积神经网络对星座图进行特征自动提取和调制识别。实验结果表明，对于实测信号，所提算法能够较好地恢复星座图并实现BPSK、QPSK和8PSK的准确识别。最终的识别准确率达到了99.9%，较星座图恢复前提高了24.2%。 |
| 关键词: 调制信号识别 星座图恢复 聚类算法 卷积神经网络 |
| DOI： |
|
| 基金项目:成都市科技项目(2015-HM01-00050-SF)；浙江大学CAD&CG国家重点实验室开放课题(A1923) |
|
| Multi-phase modulation recognition based on constellation recovery |
| WU Peijun,HOU Jin,LYU Zhiliang,GUI Meishu,ZHANG Xiaoyu,CHEN Zeng |
| (School of Information Science and Technology,Southwest Jiaotong University,Chengdu 611756,China;Chengdu Huari Communication Technology Co.,Ltd.,Chengdu 610041,China) |
| Abstract: |
| In the actual modulation process,the constellation shape will be distorted by the inter-symbol interference(ISI) caused by multipath and fading of radio wave transmission and the carrier frequency offset(CFO),so that it becomes difficult to be identified.To overcome this problem,a blind recognition algorithm of multiple phase shift keying(MPSK) signals based on constellation recovery and convolutional neural network(CNN) is proposed.First,the thresholds of the distance and phase angle of adjacent sampling points are set in order to remove the points caused by ISI,and the rest points are clustered.Then,the rotation operation of adjacent cluster groups is designed to eliminate the effects caused by the CFO and recover the constellation diagram.Finally,the CNN is used for automatic feature extraction and modulation recognition.The experiments with measured signals show the effectiveness of the proposed algorithm both in constellation diagram recovery and modulation recognition for BPSK,QPSK and 8PSK.The eventual recognition accuracy reaches %,which is 24.2% higher than the classification result without constellation recovery. |
| Key words: modulation signal recognition constellation recovery clustering algorithm convolutional neural network |
