Keywords
Networks, Machine Learning, Simulation

Technologies
NS-3, Reinforcement Learning, Gymnasium

In this project, the idea of using machine learning on dense wifi networks is explored with the goal of dynamically allocating channels based on reward signals calculated based on changes made by our machine learning agents. This project was based on a prior research article with the goal of replicating its results. Reinforcement Learning (RL) lets agents observe environments and make changes based on reward signals. This project focuses on Proximal Policy Optimisation (PPO) and Deep Q Network (DQN) algorithms, using NS3 for network simulation.

With three distinct scenarios outlined, low, medium and high density with set numbers of stations, access points and users. The simulator runs with one of the density scenarios configured and the agent observes this space. After a set time interval the agent recieves a reward signal, which is a calculation using key performance indicators from the access points, and changes channel allocation based off results. The result shows how these algorithms can be implemented in this way and properly gather and calculate the necessary data and show real change in the simulation.