Reinforcement Learning Based Congestion Control in Satellite Internet of Things
Zhou Wang, Jiaxin Zhang, Xing Zhang, Wenbo Wang · 2019
Satellite Internet of Things (SIoT) is widely used to support various IoT applications where a reliable terrestrial backhaul is hard to be established. In order to solve the problem of unavoidable high transmission delay and interrupt probability in satellite IoT, Delay/Interrupt Tolerant Network (DTN) is further concerned in recent years. However, DTN requires enough storage resources to ensure store-and-forward process, which are limited in satellite network, resulting in network congestion and reduction of Quality of Service (QoS). In this paper, the Limited Greedy Fast Congestion Control (LGFCC) algorithm is proposed based on incomplete information game mechanism and reinforcement learning, which integrates the usage of multi-hops packet switching links into DTN opportunistic routing. It offloads the scheduling tasks into the accessing IoT nodes, which realizes fast congestion control and improves QoS in the SIoT scenario. Numerous results show that the algorithm can reduce the bundle abort rate by approximately 30% and accelerate the convergence rate by approximately 60%.