Real-Time state-dependent routing based on user perception

In order to successfully resolve the network infrastructure´s problems the network provider has to improve the service quality. However in traditional ways, maintaining and improving of the service quality are generally determined in terms of quality of service criteria, not in terms of satisfaction and perception to the end-user. The latter is represented by Quality of Experience (QoE) that becomes recently the most important tendency to guarantee the quality of network services. QoE represents the subjective perception of end-users using network services with network functions such as admission control, resource management, routing, traffic control, etc. In this paper, we focus on routing mechanism driven by QoE end-users. Today, NP-complete is one of the most routing algorithm problems when trying to satisfy multi QoS constraints criteria simultaneously. In order to avoid the classification problem of these multiple criteria reducing the complexity problem for the future Internet, we propose two protocols based on user QoE measurement in routing paradigm to construct an adaptive and evolutionary system. Our first approach is a routing driven by terminal QoE basing on a least squares reinforcement learning technique called Least Squares Policy Iteration. The second approach, namely QQAR (QoE Q-learning based Adaptive Routing), is a improvement of the first one. QQAR basing on Q-Learning, a Reinforcement Learning algorithm, uses Pseudo Subjective Quality Assessment (PSQA), a real-time QoE assessment tool based on Random Neural Network, to evaluate QoE. Experimental results showed a significant performance against over other traditional routing protocols.