Attention is needed for RF fingerprinting

Radio Frequency (RF) fingerprinting is a novel solution for identifying unique radio from large pool of devices by analyzing the subtle characteristics that are inherent in waveform. Deep convolutional neural networks have been widely used to handle RF task because their exceptional capacity representation learning. However, there still challenges employing deep networks, such as how enable model learn more robust and discriminative fingerprints. This paper aims explore new architectures Hence we proposes Dual Attention Convolution module simultaneously learns channel attention spatial tune fingerprints, enhancing layers’ potential Our proposed lightweight plug-and-play. A number can be equipped with our module, which enables them extract approach has extensively tested through experimental trials, results demonstrated its effectiveness. It shown performance on improved 1.5% average, DAConv-ResNet50 combined ResNet50 achieve 95.6% recognition accuracy 10 USRP X310. source code available at https://github.com/zhangweifeng1218/Adaptive_RF_Fingerprinting.