Learning Based Media Access Control for UAV Ad-Hoc Networks

With the rapid development of unmanned aerial vehicle (UAV) technology and the drastic growth of multi-UAV collaborative applications, research on UAV Ad-Hoc Network (UAVNET) has received extensive attention in recent years. Due to the decentralization nature and being prone to failure and high dynamics, the performance of traditional contention-based carrier sense multiple access (CSMA) protocols and variants is far from satisfaction for UAVNETs. In this paper, we propose an intelligent access mechanism based on actor-critic (AC) algorithm, called AC-CSMA, with aim of achieving low access collision rate and high throughput for UAVNETs in dynamic environments. The UAVs are modeled as decision-making agents without priori information about the network (e.g., the number of nodes, link conditions, and other agents’ strategies). Individual UAV agents learn their optimal strategy by using the historical sensory information including the number of collisions or successful transmissions. Numerical results show that the proposed AC-CSMA mechanism significantly outperforms traditional access mechanism in terms of collision rate and throughput for UAVNETs without compromising access fairness.