面向射频指纹识别的高效IQ卷积网络结构

现有应用于射频指纹识别的卷积网络对时序同相正交(in-phase and quadrature-phase, IQ)信号的处理都是将其简单视为图像进行的，存在识别准确率低和计算量大的问题。针对以上问题，提出了一种基于IQ相关特征的卷积神经网络结构。该网络分步提取了IQ相关特征及时域特征，通过自适应平均池化获得了各通道特征均值，并用单个全连接层进行分类。实验结果表明，较传统卷积网络结构，所提网络在多种场景下的识别准确率更高，并且计算量更小。

High-efficiency IQ convolutional network structure for radio frequency fingerprint identification

Existing convolutional neural networks, which are used for radio frequency fingerprints recognition, process time-sequenced IQ (in-phase and quadrature) signals as images directly, resulting in low recognition accuracy and high computation complexity. IQCNet(convolutional neural network structure based on IQ correlation features), an efficient convolutional network structure, was proposed. IQCNet firstly extracted IQ correlation features and time domain features, then obtained the average value of each channel features through adaptive average pooling, and finally used only one fully connected layer for classification. Experimental results under a variety of channel conditions show that IQCNet improves recognition accuracy greatly with lower computation complexity compared with traditional convolutional neural networks.