Kexin Liang

Continuous Variable Quantum Key Distribution (CVQKD) is a promising secure communication technology, whose system performance is limited by excess noise caused by phase/polarization drift. However, traditional compensation methods for suppressing excess noise face some challenges, such as dedicated hardware modules hindering system integration and channel fluctuations reducing the robustness of compensation algorithms. We introduce an ensemble learning based framework for excess noise suppression to address these issues. The framework realizes the collaborative suppression of the original excess noise in the system and the additional excess noise generated by channel fluctuations, achieving 90% of the overall suppression rate when the transmittance decreases by 35%. Meanwhile, the secret key rate is increased by an order of magnitude with the repetition frequency. Compared with other schemes, it substantially enhances the performance and robustness of CVQKD systems under dynamic channel conditions without increasing hardware complexity. This is a crucial step in promoting the practical and integrated development of future quantum networks.