基于CNN-BiGRU网络的CPM调制指数识别

CPM Modulation Index Recognition Based on CNN-BiGRU Network

连续相位调制(Continuous Phase Modulation,CPM)作为一类频带利用率高的非线性数字调制信号,在移动通信和卫星通信领域都有着广泛的应用前景。针对目前低信噪比下CPM信号的调制参数识别困难的问题,提出了一种卷积神经网络级联双向门控循环网络(Convolutional Neural Network Cascades Bidirectional Gated Recurrent Network,CNN-BiGRU)的方法来完成调制指数的识别。输入的二维信号首先经过卷积层(CNN)提取特征,再经过双向门控循环网络(BiGRU)提取信号的前后文信息,最后经过全连接层(DNN)进行特征整合,使用Softmax函数进行分类。实验结果表明,相较于传统方法,该方法在低信噪比下对不同调制指数CPM的识别效果得到了明显的提升。当信噪比大于4dB时,整体识别率在98%以上。即使针对调制指数相等的Single-h CPM和Multi-h CPM信号,该方法也能够进行很好的区分。

Continuous phase modulation(CPM),as a kind of nonlinear digital modulation signal with high frequency band utilization,has broad application prospects in the fields of mobile communication and satellite communication.Aiming at the current difficulty in identifying the modulation parameters of CPM signals under low signal-to-noise ratios,this paper proposes a Convolutional Neural Network cascades Bidirectional Gated Recurrent Network(CNN-BiGRU)method.Complete the identification of modulation index.The input two-dimensional signal first passes through the convolutional layer(CNN)to extract features,then passes through the bidirectional gated recurrent network(Bi GRU)to extract the context information of the signal,and finally through the fully connected layer(DNN)for feature integration,using the Softmax function for classification.