Channel Characteristics Based Adjustable Fingerprint for Identity Authentication in WDM-PON With Deep Neural Networks

Abstract

Wavelength-division-multiplexed passive optical network (WDM-PON) has been widely deployed for the high-speed, reliable transmission and low-cost properties. The physical layer identity authentication in WDM-PON becomes increasingly prominent. Recently, many device fingerprint based identity authentication schemes are proposed. However, these schemes only realize constant fingerprint, which will be acquired by illegal ONU after optical spectrum analysis. Therefore, the higher-level security requirement cannot be satisfied. To solve the problem, we propose a physical layer identity authentication method in WDM-PON by exploiting the channel characteristics based adjustable fingerprint with deep neural networks (DNNs). By secretly negotiating with legal optical network units (ONUs) on the lengths of the local fibers applied by them, the optical line terminal (OLT) can acquire the unique channel characteristics fingerprint obtained by each legal ONU. It should be noted that the fingerprint can be adjusted by modifying the length of the local fiber. Moreover, the same number of DNNs as legal ONUs are trained for fingerprint identification. Simulation results show 100% identification accuracy for illegal ONU when the length deviation between two fibers applied by legal ONU and illegal ONU is greater than 1.5 km. Meanwhile, the identity of each legal ONU can be recognized with 100% accuracy.

Keywords

• Deep learning
• identity authentication
• channel characteristics fingerprint
• identity spoofing attack
• wavelength-division-multiplexed passive optical network (WDM-PON)