QoE based predictive handover mechanism in software-defined enterprise Wi-Fi networks

Abstract

In recent decades, service providers and enterprises have tried to fulfill the need of users to Wi-Fi connections inside residential buildings, campus, and public palaces with new centralized Wi-Fi frameworks. Although they have mostly solved the issue of Wi-Fi networks access and convergence by optimizing and softwarization of them using Software-Defined Networking (SDN), there is still room to make this optimization as intelligent as possible, specially with rapid growth of user’s demand and application on smartphones. Moreover, SDN will allow us to improve the performance of centralized systems. A guaranteed connection is a key feature for wireless network users so that they can continue to use their application even if they are moving from one side of a network to another side. Handover process makes this happen by steering user from one access point to another access point or from one network to another network. Deciding when to move the user from one interface to another interface can affect the QoS for users. In an enterprise Wi-Fi network, mobile users may be covered by multiple access points (APs). To optimize resource allocation, a soft handover is required in which the user’s device is seamlessly transferred from one AP to another, and this decision made centrally by a Wi-Fi network controller. Unfortunately, state-of-the-art soft handover mechanisms are often designed to optimize resources from the network provider’s point of view and do not take into account user’s real-time behaviors, which may affect user’s Quality of Experience (QoE). In this thesis, a new machine learning (ML)-based method presented to define an optimal handover mechanism. This method allows predicting whether the handover that is going to happen will maintain QoE when users are moving inside a building. Our first goal is to present a framework for handover prediction by introducing a continues score scaling based on user’s QoE. We study the behavior of tenant and effect of this behavior on the handover mechanism using a data-set obtained from a real case study on a university campus. Then we define a set of rules based on our prediction results and observation inside the network. Our framework for handover prediction is completed by feeding the handcrafted features to a Support vector regression (SVR). The proposed method applied to more than one year of collected data from access points of the mentioned campus. The evaluation of results proves the efficiency, generalization power, and robustness of our presented framework for predicting a time-independent handover mechanism. Our proposed method improves 34% of user throughput compared to state-of-the-art algorithms. In this work, our baseline is the XcellAir self-organization network, which is the service provider of the campus. We run and evaluate our experiment based on the results of their optimization system. The proposed framework is based on the hypothesis that handover happened when a user is moving between two stations, and steering will happen after the user faced performance degradation in the received service. This fact suggests the idea of a proactive method rather than using a threshold-based method for developing our framework. We proposed an approach by predicting a score that we defined based on QoE of users (From user perspective of view) and our handcrafted feature also from user feedback. Due to the bias that we can have in time-dependent predictors and the fact that moving inside a network can occur very quickly, we focused on using more important features and learning the behavior of users based on parameters rather than time. We test the introduced framework on our data-set, and the results confirm the efficiency of the proposed method in comparison to the baseline model.

Date	25 Nov 2019
Original language	American English
Awarding Institution	École de technologie supérieure
Supervisor	Mohamed Cheriet (Supervisor) & Charles Despins (Co-supervisor)