基于半监督生成对抗网络的通信信号调制识别算法

针对小样本条件下通信信号识别混淆的问题,提出了一种基于半监督生成式对抗网络的调制识别算法.首先结合半监督学习思想利用少量标签数据和大量未标签数据训练网络;其次在输出层添加辅助分类器进行结果判定,针对性设计了目标函数和损失函数,以满足网络生成虚假数据和实现信号分类的目的;最后使用不同的激活函数并用反卷积和Dropout代替池化操作,有效降低了算法复杂度并加快网络收敛速度.仿真实验表明:该算法适应性强、计算量小,较传统算法识别准确率提升了6％ ～13％,有效实现了小样本条件下的调制样式识别.

A Communication Signal Modulation Recognition Algorithm Based on Semi-Supervised Generative Adversarial Networks

In view of the confusion problem of few shot learning in communication signal recognition, a communication signal recognition algorithm is proposed based on semi supervised generative adversarial network in this paper. Firstly, this algorithm utilizes a small amount of labeled samples and a large amount of unlabeled samples for training the network with the idea of semi supervised learning. Secondly, an auxiliary classifier in the output layer is added to determine the result and a new loss function and objective optimization are designed to meet the purpose of generating fake data and realizing signal classification. Finally, in order to reduce the complexity of the algorithm and accelerate the convergence speed of the algorithm effectively, different activation functions are chosen by the network and the de convolution and Dropout are used to replace the pooling layer. The experimental results show that this algorithm is adaptable and computationally, the recognition accuracy is improved by 6%~13% compared with the traditional algorithm, effectively realizing the modulation pattern recognition in few shot learning.