基于GRU-CNN并联神经网络的自动调制识别

Automatic Modulation Recognition Based on GRU-CNN Parallel Neural Network

为提高非合作通信系统的调制方式识别准确率,提出了一种基于并联门控循环单元(Gated Cycle Unit,GRU)神经网络和卷积神经网络(Convolutional Neural Network,CNN)的数字通信信号识别方法。根据调制信号的特性,将笛卡尔坐标下的原始数据转换到极坐标下,同时求原始数据的自相关序列,作为输入数据分别送入GRU和CNN网络中。对含BPSK、QPSK、8PSK、π/4-DQPSK以及四类QAM调制信号集合进行的实测信号实验结果表明,所提方法在低信噪比下能取得较好的识别性能,在0 dB时平均识别率接近90%。

In order to improve the accuracy of modulation recognition in non-cooperative communication systems,a digital communication signal recognition method based on parallel gated cycle unit(GRU)neural network and convolutional neural network(CNN)is proposed.According to the characteristics of the modulated signal,the original data in Cartesian coordinates is converted to polar coordinates,and the autocorrelation sequence of the original data is obtained and sent to the GRU and CNN networks as input data.The measured signal experiments on the sets of modulated signals including BPSK,QPSK,8PSK,π/4-DQPSK and four types of QAM show that the proposed method can achieve good recognition performance at low signal-to-noise ratio(SNR),and the average recognition rate is close to 90%when SNR is 0 dB.