Observable-Level Framework for Finite-Key Security in Coherent-One-Way Quantum Key Distribution

Abstract

Practical security analysis of Coherent-One-Way (COW) quantum key distribution requires optical-device modeling, finite experimental statistics, and security inference to be treated consistently without conflating assumptions. We present a modular observable-level simulation framework organized through the pipeline θ_phys -> p -> O -> S_j(O), in which physical source, channel, receiver, and detector parameters determine detector-resolved event probabilities, a finite-block layer applies an effective dead-time treatment and generates aggregate detector counts, and interchangeable security evaluators act on the same observable record. The framework is verified against analytical optical, virtual-state, and finite-key relations and evaluated against reported experimental COW-QKD operating points. The Kato finite-key calculation reproduces the refined experimentally reported secure-key rates most closely at intermediate distances, with deviations of -1.5% at 50 km and -10.3% at 75 km. Monitoring bright-port decoy counts agree within 0.9-2.4%, while remaining distance-dependent dark-port residual identifies the principal model-sensitive observable of the present physical description. Applying Kato and Azuma-Hoeffding concentration treatments to identical detector records isolates the effect of finite-key statistical inference and shows substantially tighter Kato phase-error bounds for rare-event monitoring records studied here. Holding the Kato evaluator fixed, joint optimization of coherent-state mean photon number mu and receiver splitting fraction t_B increases model-predicted secure-key rate by 7.4-20.2% relative to optimization of mu alone with t_B = 0.3. The framework therefore provides a common computational interface for experimental reproduction, controlled security-model comparison, model-based protocol optimization, and modular adversarial stress testing in COW-QKD.

Abstract: arXiv descriptive metadata (CC0). Thank you to arXiv for use of its open access interoperability. This service was not reviewed or approved by, nor does it necessarily express or reflect the policies or opinions of, arXiv.