Learning Wireless Network Association-Control with Gaussian Process Temporal Difference Methods

This paper deals with the problem of improving the performance of wireless networks through the use of association control, which is the activity of intelligently associating users with the network's access points (APs), taking advantage of overlaps in the coverage areas of the APs. The optimal solution to this problem is classified as NP-hard. We present an innovative association control method which utilizes a novel Reinforcement Learning (RL) algorithm – Gaussian Processes Temporal Differences (GPTD). GPTD, an algorithm which addresses the value function estimation in continuous state spaces, and GPSARSA, an algorithm which uses GPTD to compute a complete RL solution, were defined and presented by Engel et al GPTD has only been tested so far on simple and theoretical problems, and there was a desire to test its behavior under the conditions of real-life problems. In this study we attempt to accomplish these two symbiotic goals of (i) proposing a solution to the association control problem and also (ii) developing a realistic testing environment under OPNET for GPTD and for RL in general.