A Distributed Link Scheduler for In-Band Full Duplex Wireless Networks

In a wireless network, a well designed link scheduler ensures links are activated frequently and experience minimal or no interference, meaning these links will have a high capacity. A challenging problem, however, is random or uncertain channel gains. Another challenge is determining whether a set of links can co-exist together, meaning they do not cause excessive interference to one another. These challenges have a direct impact on the number of activated links as well as their data rates. To this end, this paper proposes a reinforcement learning approach that enables each node to learn the best link to activate and data rate to use over random channel conditions. In addition, it allows nodes to learn the opportune time to use full-duplex or half-duplex transmissions. Our simulation results show our link scheduler results in an average throughput that is triple that of Carrier Sense Multiple Access (CSMA), and up to quadruple the average throughput of Time Division Multiple Access (TDMA). Moreover, our link scheduler remains superior when channel gains vary significantly from their average value.