Abstract
Device fingerprint can be utilized in optical communication system to strengthen the physical layer security for its uniqueness and unforgeability. In this letter, we propose and demonstrate a noise-model-assisted feature extraction method to reveal the device fingerprint hidden in the transmitted signal. Our scheme is verified in orthogonal frequency division multiplexing-passive optical network (OFDM-PON). First, the additive and multiplicative noise in normal data signal is extracted and two-dimensional feature matrix is formed. Then, a trained convolutional neural network (CNN) is used as a classifier to identify the fingerprint from the feature matrix. Experimental results show that our method achieves a high identification accuracy up to 99.25%. In the meanwhile, the loss function and training accuracy have an excellent performance. The ability of identifying rogue optical network unit (ONU) is also tested and the identification accuracy is 100%. With the noise-model-assisted CNN, the physical layer security of the system is adequately enhanced under the comprehensive consideration of the ability of identifying legal ONU and resisting illegal ONU.
Highlights
O PTICAL communication system has wide application scenarios in modern society, ranging from personal, commercial to military communications [1]
We propose and demonstrate a device fingerprint identification method based on a noise-model-assisted convolutional neural network (CNN) for enhancing the physical layer security
Four optical network unit (ONU) are constructed by an Alnair Labs TLG-200 narrow linewidth tunable laser, a Tektronix AWG7122C arbitrary waveform generator (AWG), four radio frequency amplifiers (CENTELLAX: OA4SMM4) and four Mach-Zehnder modulators (MZMs: three JDSU, 21067769-002 and one Photline, MX-LN-20)
Summary
O PTICAL communication system has wide application scenarios in modern society, ranging from personal, commercial to military communications [1]. The same type of devices cannot be identical because manufacturing tolerances exist in every component of the device, letting alone different types of devices If this kind of specificity can be extracted and identified by processing the signals transmitted through the device, the device fingerprint can be used as an identity of hardware identity and leveraged for securing the physical layer from illegal attacks. Targeted solutions are raised to deal with the different situation Methods such as CNN [15], wavelet transform [16] and statistical characteristics [17] have been studied and applied in the feature extraction stage. We propose and demonstrate a device fingerprint identification method based on a noise-model-assisted CNN for enhancing the physical layer security. The results show that, compared with our previous work [20], the proposed method achieves a higher legal ONUs identification accuracy up to 99.25%, and the identification accuracy of illegal ONU is 100%
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
