Speaker
Description
The deployment and operation of satellite quantum-key-distribution (QKD) networks require accurate knowledge of atmospheric turbulence, which determines optical coupling efficiency and therefore directly impacts the secure key rate. Conventional turbulence characterization is based on dedicated optical instrumentation and long-term monitoring campaigns, limiting both site assessment and operational forecasting. We investigate whether machine learning applied to routinely available meteorological observations can provide the turbulence information required to deploy and operate space-ground quantum networks.
We address two complementary tasks. First, for ground-station siting, we predict the turbulence-regime distribution of new locations using only weather data. A model trained on a desert observatory reproduces the turbulence distribution of an unseen coastal site with $94\%$ overlap and that of an unseen continental site with $88\%$ overlap, substantially outperforming climatological baselines. Second, for operational scheduling, we develop an intra-night forecasting framework that predicts the evolution of the atmospheric turbulence throughout the remainder of the night. Combining complementary forecasting models yields $80\%$ directional accuracy on the $69\%$ of nights for which the models agree, providing a practical confidence measure for scheduling decisions.
To connect these predictions to experimentally relevant quantum communication metrics, we map turbulence forecasts to secure key rates through an adaptive-optics-aware model of a satellite-QKD downlink. This enables candidate ground stations to be evaluated directly in terms of quantum-network performance. Our results demonstrate how machine learning can support both the planning and real-time operation of satellite-QKD infrastructure while expressing atmospheric predictions in the operational language of quantum information science.
| Topic | Machine learning techniques for experimental quantum information science |
|---|