Enhanced authentication technology based on deep neural network in the physical layer of optical communication

Abstract

To solve the problem of device identity security management in Passive Optical Networks (PON), this study proposes an enhanced authentication technique based on Deep Neural Network (DNN) in the physical layer of optical communication. By negotiating the device power and transmission distance used with legitimate Optical Network Units (ONUs), the Optical Line Terminal (OLT) can obtain a unique channel feature fingerprint for each legitimate device. These features are then identified and classified using a trained Deep Neural Network. Also, the method takes into account the substitution and fiber-splitting interception attacks. Simulation results show that the legitimate devices can be accurately identified using 64 neurons and 20 iterations, with a recognition accuracy of 100 %. Therefore, the method effectively improves the ability of the PON system to resist spoofing attacks at the physical layer and significantly improves the security level of the system.