AI-Based QOS/QOE Framework for Multimedia Systems

Abstract

Streaming applications have grown exponentially in recent years to be the most dominant contributor to global internet traffic, and this is because they satisfy a wide variety of customer needs such as video conferencing, video surveillance, and stored-video streaming. Recent studies show that the telecommunication industries lose millions of dollars due to poor QoE experienced by end users. Accenture also carried out a recent survey that showed that about 82% of customer defection was due to poor QoE. In this work, a resultant QoS framework, which delivered a QoE that can meet user’s expectation was developed. The system quantitatively measured QoE on Multimedia using different variables that affects (QoS). Deep learning algorithms and ground truth datasets were used for this work to map the QoS features, such as delay, bandwidth, packet loss rate and throughput, which serves as input, to the output QoE. Controlled experiment methodology and the active learning approach was used. Multilayer Perceptron (MLP) and Deep Belief Network (DBN) were used, to maintain, update and regulate the states of the network model. The incorporation of RBM on DBN was done using artificial features obtained at the Restricted Boltzmann Machine (RBM) stage. This research produced a deep learning hybrid approach to optimize QoS and QoE in multimedia applications. A novel predictive QoE model where relevant QoS parameters and how they influence QoE was also presented, and finally a rich QoS-QoE dataset was presented, which can further be used as a framework to ensure responsible AI in multimedia systems.