RL-Based Radio Resource Slicing Strategy for Software-Defined Satellite Networks

As an attractive infrastructure, satellite networks have obvious limitations such as high complexity and lack of flexibility. The introduction of software-defined networking (SDN) and network slicing into satellite networks is a promising solution. In satellite slicing networks, dynamic allocation of radio resource is a critical and challenging issue that should maximize system utility and take into account users' dynamic channel conditions and quality of service (QoS). In this paper, we propose a dynamic radio re-source slicing strategy called Sat-RRSlice which is based on reinforcement learning(RL) to serve the low-orbit satellite slicing networks. The strategy learns useful information from the wireless environment dynamically, then updates the number of radio resource units allocated to each slice based on the useful information dynamically. Secondly, we use the utility function as RL's reward function which considers users' dynamic channel conditions and QoS. Finally, this strategy avoids invalid allocation and resource waste by implementing access control. Simulation results demonstrate the effectiveness of Sat-RRSlice strategy.