面向通信信号调制识别的半监督生成对抗网络框架

针对只有少量标签数据的弱监督条件下现有调制信号识别模型准确率较低的问题，提出基于生成对抗网络的半监督学习框架。该方法通过对通信信号进行冗余空域变换，使其在适应生成对抗网络模型的同时保留丰富的信号相邻特征；通过梯度惩罚Wasserstein生成对抗网络的引入，构建适宜电磁信号处理的半监督学习框架，实现对无标签信号样本的有效利用。为了验证所提算法的有效性，在RADIOML 2016.04C数据集上进行测试。实验结果表明，该方法在半监督条件下能训练出高效的分类器，获得优异的调制识别结果。

Semi-supervised generative adversarial network framework for modulation recognition of communication signals

Due to overlapping frequency bands between frequency-consuming devices, the local transmitter will cause serious collocation interference to the receiver, causing the saturation of receiver antenna interface and failing to work normally. In response to the above problems, a collocation interference suppression method based on auxiliary array was proposed, a collocation interference suppression system model based on auxiliary array was established, the expression of minimum residual interference power was derived, channel conditions that need to be satisfied for collocation interference suppression was proposed. In addition, the location of the auxiliary was analyzed when main-transmitter array was non-cooperative, and the relationship of the locations of arrays was analyzed when the main-transmitter array was cooperative. The theory and simulation results show that the collocation interference suppression method based on auxiliary array can effectively suppress collocation interference at the receiver antenna interface without affecting the transmission efficiency of the local main-transmitter array.