Physics-informed cross-layer adaptive optimization for continuous-variable quantum key distribution

Abstract In practical fiber optic communication networks, continuous-variable quantum key distribution (CV-QKD) needs to be transmitted along the same fiber as classical service signals, which introduces complex noise and crosstalk, leading to parameter estimation bias and modulation mismatch, affecting the reliability of secure communication. Existing methods lack physical constraints and security boundary determination. To address this, this paper constructs a deep semantic model (DSM) that integrates a physics-informed neural network and a Transformer, embedding channel physical laws into the process to achieve adaptive adjustment of modulation variance under complex environments. Furthermore, a semantic credibility index is proposed for the first time to constrain physical consistency and security boundaries. Under the GG02 protocol, the proposed method achieves a key rate gain of approximately 0.082 bits/pulse (a 21% improvement). The results show that this method can effectively improve the security, robustness, and deployment feasibility of CV-QKD in practical fiber optic communication networks.