基于多端特征融合模型的MIMO-OFDM系统盲调制识别

自动调制分类（Automatic Modulation Classification， AMC）在认知无线电中起着提高频谱利用率的重要作用，然而，现有的大多数工作都集中在单输入单输出系统中的单载波通信。针对当前非协作通信中多输入多输出正交频分多路复用（Multiple-Input Multiple-Output Orthogonal Frequency Division Multiplexing， MIMO-OFDM）系统子载波的盲调制识别问题，本文提出了一种基于多端特征融合模型的盲调制识别方法。首先，利用特征矩阵的联合近似对角化算法（Joint Approximate Diagonalization of Eigenvalue Matrix， JADE）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剖面和同向正交分量作为浅层特征。最后，搭建多端特征融合模型，利用一维卷积网络（One-Dimensional Convolutional Neural Network， 1D-CNN）与通道注意力模块（Channel Attention Module， CAM）的串联模型完成对浅层特征的提取和映射，并使用测试样本对所提出的调制识别算法进行仿真验证。仿真结果表明，本文方法在不需要先验信息的情况下对MIMO-OFDM系统的调制方式可以进行有效识别，在信噪比为4 dB时的识别精度可达到90%。 

Blind Modulation Recognition of MIMO-OFDM System Based on Multi-terminal Feature Fusion Model

? ?Automatic Modulation Classification （AMC） plays an important role in improving spectrum efficiency in cognitive radio. However， most of the existing work focuses on single carrier communication in single input single output system. Aiming at the problem of Blind Modulation Recognition of subcarriers in Multiple-Input Multiple-Output Orthogonal Frequency Division Multiplexing （MIMO-OFDM） system in non-cooperative communication， a blind modulation recognition method based on multi terminal feature fusion model is proposed in this paper. Firstly， the joint approximate diagonalization algorithm （JADE） of the characteristic matrix is used to recover the transmitted signal from the mixed signal at the receiver. Then， the cyclic spectral profile and co directional orthogonal component of the recovered signal are extracted as shallow features. Finally， the multi terminal feature fusion model is built， the shallow feature extraction and mapping are completed by using the series model of one-dimensional convolutional network （1D-CNN） and channel attention module （CAM）， and the proposed modulation recognition algorithm is simulated and verified by test samples. Simulation results show that this method can effectively identify the modulation mode of MIMO-OFDM system without priori informations， and the recognition accuracy can reach 90% when the signal-to-noise ratio is 4 dB.