Cognitive wireless access selection at client side: Performance study of a Q-learning approach

The high dynamics of mobile and wireless networks calls for intelligent mechanisms to select access networks and corresponding points of access for the clients and their active applications. However, one needs to be careful not to increase the number of handovers substantially as it may cause large communication overhead to the network. In this paper, we consider mechanisms located at the client-side where the greedy selfish behavior should be regulated by using algorithms which simultaneously improve the quality of experience (QoE) but do not disturb much or, in the best case, even improve the overall network performance. Specifically, we introduce a Q-learning based QoE-aware access selection algorithm which enables the clients to learn from past experiences in order to find the optimal actions. The statuses of the available points of access are described by a cascade fuzzy classifier. The Q-learning based solution is compared to the default mechanism and an opportunistic fuzzy inference algorithm by simulation. The results indicate that a Q-learning approach is able to keep the number of handovers reasonably low while still achieving a good QoE, thus providing a better approach both from the user and the network operator perspective.