We propose key-sifting algorithms for a continuous-variable quantum key distribution (CV-QKD) protocol using a machine-learning framework based on the isolation forest and a traditional digital signal-processing scheme based on the Wiener filter. In previous works, the key-sifting procedure performs by discarding bits on different measurement basis, but a few useless portions are still retained, so that the transmission distance of CV-QKD systems is limited by excess noise. Here, we develop sifting schemes based on anomaly detection, which provide better feasibility in noisy environments. We compare the detection results for both approaches and estimate the excess noise to confirm the utility of our proposal in real-life applications. Results obtained over a 100-km fiber link indicate that the key-sifting algorithms can ensure low excess noise. Moreover, some of the techniques we develop are applicable to other QKD protocols.

• Received 18 May 2021
• Accepted 15 July 2021

DOI:https://doi.org/10.1103/PhysRevA.104.012616