Simple Reinforcement Learning based Contention Windows Adjustment for IEEE 802.11 Networks

This paper proposes a simple reinforcement learning-based CW adjustment for IEEE 802.11 Networks. In the proposed scheme, each node finds an optimal value of CWmin for networks from its transmission attempts. The proposed scheme applies a multi-armed bandit solution which is the most simple way among general reinforcement learning methods. In addition, we introduce a novel rewarding policy for the proposed scheme. This contributes to a simple and light implementation for even poor wireless devices. We demonstrate the proposed method provides almost the same network performance as the conventional method through the computer simulation results.