一种优化的卷积神经网络调制识别算法

针对非合作接收条件下信号的调制识别问题，提出了一种基于循环谱特征和深度卷积神经网络的自动调制分类算法。该算法首先利用二值化、形态学操作等技术对循环谱数据集预处理，提高网络泛化能力；然后将数据集输入到卷积神经网络模型中，经过网络的特征提取实现分类识别。在网络中添加残差块网络增大感受野，提高特征提取能力。采用Dropout、优化函数等技术优化网络结构，防止训练过拟合。仿真结果表示，与传统方法和现有的一些深度学习调制识别方法相比，该算法在低信噪比条件下有更高的准确率，具有明显的抗噪声优势，是一个有效的调制识别算法。

An optimized convolutional neural network modulation recognition algorithm

For the problem of modulation recognition of signals under non-cooperative reception conditions,an automatic modulation classification algorithm based on cyclic spectrum features and deep convolutional neural networks is proposed.The algorithm firstly uses the techniques of binarization and morphological operations to preprocess the cyclic spectrum dataset and improve the generalization ability of the network.Secondly,the dataset is input into the convolutional neural network model,and the feature extraction of the network is used to realize the classification and recognition.Adding a residual block network to the network increases the receptive field and improves the feature extraction capability.And Dropout,optimization functions and other techniques are used to optimize the network structure to prevent over-fitting.The simulation results show that compared with the traditional methods and some existing methods of deep recognition learning,the algorithm has higher accuracy under low signal-to-noise ratio(SNR) condition and has obvious anti-noise advantage.It is an effective modulation recognition algorithm.