系统工程与电子技术 ›› 2023, Vol. 45 ›› Issue (3): 902-912.doi: 10.12305/j.issn.1001-506X.2023.03.32

• 通信与网络 • 上一篇    下一篇

基于联合特征参数和一维CNN的MIMO-OFDM系统调制识别算法

汪锐, 张天骐, 安泽亮, 王雪怡, 方竹   

  1. 重庆邮电大学通信与信息工程学院, 重庆 400065
  • 收稿日期:2021-11-25 出版日期:2023-02-25 发布日期:2023-03-09
  • 通讯作者: 汪锐
  • 作者简介:汪锐(1998—), 女, 硕士研究生, 主要研究方向为通信信号盲处理、深度学习
    张天骐(1971—), 男, 教授,博士研究生导师, 博士, 主要研究方向为通信信号的调制解调、盲处理、语音信号处理、神经网络实现, 以及现场可编程逻辑门阵列、超大规模集成电路实现
    安泽亮(1993—), 男, 博士研究生, 主要研究方向为调制识别、机器学习和深度学习
    王雪怡(1998—), 女, 硕士研究生, 主要研究方向为信道编码参数盲识别技术研究
    方竹(1996—), 男, 硕士研究生, 主要研究方向为卫星扩频信号捕获
  • 基金资助:
    国家自然科学基金(61671095);国家自然科学基金(61702065);国家自然科学基金(61701067);国家自然科学基金(61771085);信号与信息处理重庆市市级重点实验室建设项目(CSTC2009CA2003);重庆市自然基金(cstc2021jcyj-msxmX0836);重庆市教育委员会科研项目(KJ1600427);重庆市教育委员会科研项目(KJ1600429)

Modulation recognition algorithm for MIMO-OFDM system based on joint characteristic parameters and one-dimensional CNN

Rui WANG, Tianqi ZHANG, Zeliang AN, Xueyi WANG, Zhu FANG   

  1. School of Communication and Information Engineering, Chongqing University of Posts and Telecom-munications, Chongqing 400065, China
  • Received:2021-11-25 Online:2023-02-25 Published:2023-03-09
  • Contact: Rui WANG

摘要:

针对当前非协作通信中多输入多输出正交频分复用(multiple-input multiple-output orthogonal frequency division multiplexing, MIMO-OFDM)系统子载波的调制识别问题, 提出了一种基于一维卷积神经网络(one-dimensional convolutional neural network, 1D-CNN)的调制识别方法。首先, 利用特征矩阵的联合近似对角化(joint approximate diagonalization of eigenvalue matrix, JADE)算法从接收端的混合信号中恢复发送信号; 然后, 提取恢复信号的循环谱切片和四次方谱作为浅层特征; 最后, 利用1D-CNN对特征进行训练, 使用测试样本对所提出的调制识别方法进行仿真验证。仿真结果表明, 所提方法对MIMO-OFDM系统中的5种信号可以进行有效识别, 在信噪比为10 dB时的识别精度即可达到100%。

关键词: 多输入多输出正交频分复用, 调制识别, 循环谱, 四次方谱, 一维卷积神经网络

Abstract:

Aiming at the modulation identification of subcarriers in multiple-input multiple-output orthogonal frequency division multiplexing (MIMO-OFDM) system in non cooperative communication, a modulation identification method based on one-dimensional convolutional neural network (1D-CNN) is proposed. Firstly, the joint approximate diagonalization of eigenvalue matrix (JADE) algorithm is used to recover the transmission signal from the mixed signal at the receiver. Then, the cyclic spectrum slice and quartic spectrum of the recovery signal are extracted as shallow features. Finally, the features are trained by 1D-CNN, and the proposed modulation recognition method is simulated and verified by test samples. Simulation results show that the proposed method can effectively identify five signals in MIMO-OFDM system, and the recognition accuracy can reach 100% when the signal-to-noise ratio is 10 dB.

Key words: multiple-input multiple-output orthogonal frequency division multiplexing (MIMO-OFDM), modulation identification, cyclic spectrum, quartic spectrum, one-dimensional convolutional neural network (1D-CNN)

中图分类号: