基于半监督联合神经网络的调制识别算法

针对基于有监督学习通信信号分类算法需要大量有标签训练样本，而在实际场合大多无法满足数量要求的问题，提出利用数据驱动模型的半监督学习方法，通过对比预测编码无监督算法预训练和有监督学习相结合，利用LSTM (long short term memory)和ResNet (residual network)联合神经网络实现小样本自动提取特征，提高小样本条件下信号识别准确率。在真实通信调制信号集上实验表明，半监督联合神经网络结构较以往方法，识别准确率提升3%-20%，小样本条件下性能提高60%，同时在低信噪比条件下识别能力突出，0dB时对11种调制信号平均识别正确率达到92%，具有明显优势。 

Semi-supervised Joint Neural Network Based Recognition Algorithm of Modulation Signal

Due to the presence of demanding a large number of labeled samples of communication signal based supervised learning classification algorithm and considering that most of the actual situations can ’ t meet the requirements of large samples.Proposing a data driven model method based on semi-supervised learning,which combined Contrastive Predictive Coding with unsupervised pre-training algorithm and supervised learning algorithm,using a joint neural network composed by LSTM(Long Short-term Memory) and ResNet(Residual Network) to realize automatic extraction feature and improving signal recognition accuracy on small sample conditions.Experimental results show,on real communication modulation signal data sets,joint neural network structures compared with previous methods have obvious advantages that recognition accuracy increased by 3%~20%,60% performance improvement under small sample conditions and the average recognition rate of 11 kinds of modulation type is 92% at low SNR condition with 0 dB.