Free Space Optical Communication has emerged as a promising technology for high-speed and secure data transmission between ground stations on Earth and orbiting satellites. However, this communication technology suffers from signal attenuation due to atmospheric turbulence and beam alignment precision. Low Earth Orbit satellites play a pivotal role in optical communication due to their low altitude over the Earth surface, which mitigates the atmospheric precipitation effects. This paper introduces a novel data path control law for satellite optical communication exploiting Artificial Intelligence-based predictive weather forecasting and a node selection mechanism based on Reinforcement Learning. Extensive simulations on three case studies demonstrate that the proposed control technique achieves remarkable gains in terms of link availability with respect to other state-of-the-art solutions.

To meet the increased bandwidth demand from remote locations, new satellites are being launched to allow networks to operate at higher frequencies (i.e., Ka-band – 27.0GHz - 40.0 GHz and V-band 40.0GHz – 75GHz). At these frequencies high availability is challenged by rain attenuation. Consequently, satellite network operators are offering hybrid solutions where, during a rain outage, traffic is moved to a lower frequency secondary heterogeneous satellite network not affected by rain.   This paper evaluates a mobile and network centric handover and admission strategy using a bespoke simulation model of a rain affected star topology hybrid satellite network with primary network operating on Ka band and secondary satellite network operating at a lower frequency.    Results show that a network centric prioritised queue-based admission strategy provides a reduction in handover failure probability of about 5% without extra capacity being made available for the handover of traffic to the secondary network. When queue-based admission is used in combination with adding extra capacity in the secondary satellite network a reduction in handover failure probability of nearly 30% was achieved.   However, when more capacity is made available in the secondary network, the advantages of the network centric queue-based admission strategy slowly disappear with added capacity. This means that operators have a clear choice based on whether they believe that their secondary network has sufficient spare capacity, or whether in fact this is unlikely and therefore the extra cost of a queue-based admission strategies should be weighed against loss of revenue due to lost connections.