Reinforcement Learning for QoE-Oriented Flexible Bandwidth Allocation in Satellite Communication Networks.

Optimizing the use of satellite bandwidth to achieve maximum return in a system where user demands are constantly changing, and application-specific Quality-of-Experience (QoE) requirements need to be met, presents a complex challenge for both satellite operators and service providers (SPs). The paper investigates the application of reinforcement learning (RL) algorithms for QoE-aware flexible bandwidth allocation, which enables satellite service providers to minimize the allocated band-width while meeting the QoE requirements of their customers. By employing a time-varying queuing model, we formulated a stochastic optimization problem and applied Q-learning and state-action-reward-state-action (SARSA) reinforcement learning algorithms to determine the optimal bandwidth allocation. The findings indicate that while the algorithms exhibit similar convergence speeds, Q-learning slightly outperforms SARSA due to its more efficient bandwidth selection to meet the requirements. This demonstrates the potential of reinforcement learning as a valuable tool for optimal bandwidth allocation in satellite communications, thereby contributing to the ongoing improvement of service quality in this domain.