Adaptive multipath optical self-interference cancellation based on deep reinforcement learning

Abstract

An adaptive multipath optical self-interference cancellation scheme based on deep reinforcement learning is proposed and investigated. The simulation results demonstrate that the proposed scheme can adaptively achieve multipath self-interference cancellation using deep neural networks, where the multipath SI is successfully eliminated to the noise floor and a cancellation depth of 33.4 dB over 2 GHz bandwidth at a center frequency of 2 GHz is achieved within 5 steps. The proposed scheme may provide a promising solution for future in-band full-duplex systems.