Wireless Device Identification Based on Improved Convolutional Neural Network Model

In this paper, we propose an improved convolutional neural network (CNN) model based on Lenet-5 for radio frequency fingerprint (RFF) identification. Our model can recognize wireless devices using the raw received samples without synchronization. We evaluate the 54 ZigBee devices classification problem with our model. The highest identification accuracy is 95.6% under 30dB SNR. This identification accuracy is superior to most of traditional CNN methods. We also demonstrate the robustness of our model over a wide range of signal-to-noise ratios (SNR) and receiver sampling rates. These experimental results validate the robustness of our model. Finally, in order to save computing resources, we discuss the minimum sampling points required to achieve an acceptable identification accuracy. Our experiments show that the 800 to 1000 sampling length is a suitable choice for our CNN model in this 54 ZigBee devices classification problem with the trade-off between performance and complexity.