基于深度学习的认知无线电调制参数估计

Deep Learning Based Cognitive Radio Modulation Parameter Estimation

自动调制分类是电磁空间感知的一个关键问题,目前传统的识别技术很难适应复杂的信号情况。现有的调制分类算法大多忽略了不同特征之间的互补性和特征融合的重要性。基于此,提出一种用于自动调制分类的图像特征融合方法。该方法充分利用了不同图像特征之间的互补性,通过格拉姆角场(Gramian Angular Field, GAF)方法将原始信号转换为图像,同时利用累积极坐标特征转换技术将接收到的信号从I-Q域转换为r-θ域,在r-θ域对原始信号进行特征编码然后转换为图像。使用深度学习对两种图像进行特征提取,将提取的特征融合后用作神经网络分类器的输入,以实现对多种类型信号的自动调制分类。实验结果表明,使用Swin-Transformer网络模型对转换后的图像进行分类,在信噪比大于4 dB的情况下,调制方法的识别率超过90%。

Automatic modulation classification is a key problem in electromagnetic space perception,and current traditional recognition techniques are difficult to adapt to complex signal situations.Most of existing modulation classification algorithms ignore the complementarity between different features and the importance of feature fusion.On this basis,an image feature fusion method for automatic modulation classification is proposed.The method makes full use of the complementarity between different image features by converting original signal into an image through a Gramian Angular Field(GAF)method,while converting received signal from the I-Q domain to the r-θdomain using a cumulative polar coordinate feature conversion technique,and feature-encoding the original signal from the r-θdomain and then converting it into an image.Deep learning approach is used to extract features from both images,and the extracted features are fused and used as input to a neural network classifier to achieve automatic modulated classification of multiple types of signals.Experimental results show that with a signal to noise ratio greater than 4 dB,the classification of the converted images using the Swin-Transformer network model achieves a recognition rate of over 90%for the modulation method.