Deep Learning-Based Optimization of Underwater QKD Systems With Time-Gated SPADs

In this paper, we consider underwater quantum key distribution (QKD) systems with time-gated single photon avalanche photodiodes (SPADs) and present a comprehensive performance analysis and optimization. We utilize a deep learning model for the system optimization. We first perform Monte Carlo simulations for a subset of possible scenarios and train the deep learning model on them. Then, we use this model to instantly extract the optimum values for possible underwater scenarios. Our study provides critical insights on determining the optimal bit time, field of view (FoV), and gate time in underwater QKD. Through a meticulous analysis of the propagation delay and angle of arrival results, we determine the bit time and FoV, respectively, by striking a balance between the average number of received photons and background noise. Afterwards, using the proper bit time and FoV from the previous steps, we determine the optimal gate times in the sense of minimizing the quantum bit error rate (QBER). Our findings reveal that the proposed method significantly improves QBER, with enhancements reaching up to two orders of magnitude across different link lengths.