基于CNN-SVM的调制方式识别优化算法

信号调制方式识别在通信领域是一个研究热点,针对目前已调信号分类识别率受噪声的影响较大的问题,提出一种基于CNN-SVM的调制方式识别算法.该算法对不同已调信号做循环谱估计,生成相应的循环谱图,并截取等高截面图作为特征图,然后借助卷积神经网络提取相应的特征,并采用t分布邻域嵌入算法对特征值降维处理,最后输入到支持向量机对已调信号进行分类识别.经实验仿真,当信噪比高于-2 dB时,算法识别率高于96%,证实了该算法具有很好的识别效果.

Optimization algorithm of modulation recognition based on CNN-SVM

Modulation recognition is a research hotspot in the field of communication. Aiming at the problem that the classification recognition rate of modulated signal is greatly affected by noise, this paper proposes a modulation recognition algorithm based on cnn-svm. The algorithm estimates the cyclic spectrum of different modulated signals, generates the corresponding cyclic spectra, and intercepts the contour section as the feature map, then extracts the corresponding features with the help of convolutional neural network (CNN), and uses the t-Distributed Stochastic Neighbor Embedding (t-SNE) algorithm to reduce the dimension of the eigenvalues. Finally, we input the eigenvalues into the support vector machine (SVM) to classify and recognize the modulated signals. The experimental results show that when the SNR is higher than - 2dB, the recognition rate of the algorithm is higher than 96%, which proves that the algorithm has good recognition effect.