Random Matrix Theory Based Radio Frequency Fingerprinting Identification of WiFi Signal

This paper introduces a method of Radio Frequency Fingerprint (RFF) recognition based on Random Matrix Theory (RMT). RFF identification is a physical layer method to identify devices that exchanging messages in the network. The extracted subtle features of the transmitters working as the biometric fingerprint can uniquely represent the identity of the devices. However, the differences in RFF become extremely subtle when transmitters are the same type of industrial standard equipment and it is very difficult to identify the devices under this scenario. In the proposed method, a random matrix of the I/Q (In-phase/Quadrature) data of bit-similar Universal Software Radio Peripheral (USRP) devices is constructed firstly, and the extracted features are sent to the Convolutional Neural Networks (CNN) network for recognition. Experiments are carried out under Signal-to-Noise Ratio (SNR) $\in[-10,20] \mathrm{dB}$ between different datasets. It demonstrates a 0.98 accuracy in identification, outweighting the traditional classification method.