Chaotic physical security strategy based on manifold learning-assisted GANs for SDM-OFDM-PONs.

Abstract

In this paper, we propose a high-security spatial division multiplexing orthogonal frequency division multiplexing passive optical network (SDM-OFDM-PON) encryption scheme based on manifold learning-assisted generative adversarial networks (MFGANs). The chaotic sequences generated by MFGANs are applied to produce the masking vectors to encrypt the constellation and frequency. With the help of manifold learning, the proposed scheme can learn the complex structures from various chaotic models and makes use of more parameters than a single traditional model to achieve the large key space of 1 × 10183. A 70 Gb/s encrypted OFDM signal transmission over 2 km 7-core fiber was experimentally demonstrated. In addition, due to the capacity of parallel computing of graphics processing units (GPUs), the encryption time by the proposed scheme is around 1.38% of the conventional encryption scheme. It is therefore shown that the proposed encryption scheme can ensure both efficiency and security in SDM-OFDM-PON systems.