基于CNN和图域理论的BPSK/QPSK信号调制识别方法研究

Research on Modulation Recognition of BPSK/QPSK Signals Based on CNN and Graph Domain Theory

为了增强对信号的调制识别性能,将图域理论与卷积神经网络相结合的方法应用于二进制相移键控/正交相移键控(BPSK/QPSK)信号的调制识别任务中。首先,计算待识别时域信号的平方谱,并加窗得到截断序列;之后对截断序列进行归一化—量化—图构建的图域变换,提取图的拉普拉斯矩阵和无符号拉普拉斯矩阵作为卷积神经网络的输入,训练得到分类识别结果。仿真结果表明,在低信噪比条件下,该方法具有较好的识别性能,具有一定的工程应用价值。

To enhance the modulation recognition performance of the signal,a method combining graph domain theory and convolutional neural network(CNN)was applied to the modulation recognition task of binary phase shift keying/quadrature phase shift keying(BPSK/QPSK)signals in this paper.Firstly,the square spectrum of the time domain signal to be identified was calculated,and the truncated sequence was obtained by windowing;then the truncated sequence was transformed into graph by normalization,quantization and graph construction.Accordingly,the Laplacian matrix and unsigned Laplacian matrix of the graph were extracted as the input of the CNN for training to obtain the recognition results.Simulation results show that this proposed algorithm still has good recognition performance under lower SNR conditions and has a certain value in engineering application.