A Reinforcement Learning Approach to Wi-Fi Rate Adaptation Using the REINFORCE Algorithm

In this paper, we propose a novel approach for rate adaptation in 802.11 wireless networks based on the re-inforcement learning algorithm, REINFORCE. Our approach leverages a more comprehensive set of observations, including received signal strength, contention window size, current MCS and throughput, enabling a nuanced response to varying network conditions and hence leads to optimal network throughput. The unique application of REINFORCE allows our methodology to iteratively learn optimal actions under different states, resulting in enhanced rate adaptation in dynamically changing environments. We utilize ns-3, a discrete-event network simulator, in conjunction with ns3-ai, a reinforcement learning gym, to implement and evaluate our approach (named ReinRate) under varying network scenarios, with and without interference from competing transmitters. The extensive simulations conducted under static and dynamic scenarios clearly demonstrate our method's superiority over traditional strategies like the Minstrel and Ideal rate adaptation algorithms, with up to 102.5% and 30.6% higher network throughput in network scenarios without interference, and by up to 35.1% and 66.6% in network scenarios with interference.