講演情報
[10p-E219-9]Reinforcement Learning for Adaptive Intensity Control in BB84 QKD
〇(D)MAHIMA KUMAR1 (1.NETAJI SUBHAS UNIVERSITY OF TECHNOLOGY)
キーワード：
Quantum key distribution、Reinforcement learning、BB84 protocol
Quantum key distribution (QKD) guarantees cryptographic security through quantum mechanics, but practical BB84 systems fix the mean photon number μ at design time, wasting power on good channels and risking security on poor ones. We propose an adaptive framework where a Proximal Policy Optimization (PPO) reinforcement learning agent adjusts μ in real time based on the observed channel state. The channel is modelled as a three-state Markov chain representing Good (η=0.60), Medium (η=0.20), and Poor (η=0.06) transmittance conditions. The reward function jointly optimizes the asymptotic GLLP secure key rate and optical power consumption. Simulation over 200 evaluation episodes shows the PPO agent achieves a 0.90% improvement in secure key rate over a tuned static baseline, while maintaining QBER well below the 11% security threshold with zero violations. Results confirm that adaptive intensity control via reinforcement learning is a viable strategy for power-efficient, secure QKD systems.