Learning to Bond in Dense WLANs With Random Traffic Demands

The Access Points (APs) in a Wireless Local Area Network (WLAN) must be assigned one or more channels to meet traffic demands from users. To date, prior works on channel assignment assume APs have a fixed traffic demand, meaning they do not consider or adapt to spatio-temporal changes in traffic demands. To this end, we leverage Deep Reinforcement Learning (DRL), where we equip APs with a DRL-based channel assignment solution, to maximize the average number of slots in which an AP has sufficient bandwidth to meet user demands. Our APs learn from their historical traffic loads and assign themselves partially overlapping channels with minimal interference. Simulation results show that our DRL solution leads to APs satisfying 60% more user demands as compared to fixed and greedy channel bonding algorithms.