Multi-Armed Bandit for Contention Window Optimization

Future networks will require to support a tremendous number of communicating devices. Unfortunately, the basic access method employed by IEEE 802.11 networks does not scale well for an increasing number of stations. We prove that intelligent devices in those networks can use Multi-Armed Bandit (MAB) learning algorithms to improve the network throughput by choosing optimal contention window (CW) values. For this reason, we propose COMBAT, a contention window optimization approach via Multi-Armed Bandit, that can quickly and efficiently find the best contention window that maximizes the network throughput. COMBAT is defined upon the MAB learning algorithm UCB1 to handle the centralized decision-making of the contention window. The decided CW is broadcasted from the Access Point (AP) to all the connected devices. We show that using learning algorithms does help to fit more devices in such networks. We also demonstrate through a simulation campaign that COMBAT can learn the optimal policies faster than the current state-of-art methods while also requiring a low computational cost.