Embedded Radio Frequency Fingerprint Recognition Based on A Lightweight Network

The recent research on individual identification of Internet of Things (IoT) communication devices based on radio frequency feature have promoted autonomous deployments of wireless communication. But the identification of radio frequency signals is faced with the defect of high complexity, low degree of integration and high cost of calculation. We propose a lightweight model. Convolutional neural memory network (CNMN), which reduces the complexity of network model and accelerates calculation speed, has been installed on small computing devices to meet the identification needs of different scenarios. In the experiment, we will carry the universal software radio peripheral (USRP) on the low-power embedded platform Jetson Nano to form a miniaturized RF signal collection platform. The experimental results show that our RF identification model usesless computational costs. The proposed method can achieve better classification identification accuracy than traditional RF identification methods under the multi-classification target. The experiments of the signal collection are carried out in the real environment. Therefore, the networks we proposed are very reliable in real communication environment.