Quantum secured 10-Gbit/s ethernet passive optical access network and its machine-learning-assisted implementation

The extension from point-to-point quantum key distribution (QKD) to multipoint communication is an inevitable trend toward the large-scale development of QKD networks. At present, it is a relatively simple and cost-effective implementation to integrate quantum access networks (QANs) into existing Ethernet passive optical access networks (EPONs). This work proposes a quantum secured 10 Gbit/s EPON (10G-EPON), in which a plug-and-play twin-field QKD (PnP-TF-QKD) architecture is developed, requiring a single untrusted laser and a pair of shared detectors. The PnP-TF-QKD implementation is compatible with standard 10G-EPON and can support up to 64 users. In addition, prior to the deployment of QKD on EPON, it is crucial to predict in advance whether the attainable QAN indicators, such as the secure key rate, the maximum feeder fiber length, etc., meet the requirements of a certain circumstance. However, this task is often complex and time consuming. Therefore, we present a machine-learning-based user-demand-oriented prediction model to facilitate the evaluation of QAN parameters. The plug-and-play twin-field quantum access network (PnP-TF-QAN) and its machine-learning-assisted implementation method provide guidance for further experiments and actual deployments of large-scale QANs.