A Q-Learning Driven Energy-Aware Multipath Transmission Solution for 5G Media Services

Supported by the latest evolution of the 5G technologies, Augmented Reality (AR) & Virtual Reality (VR) video streaming services are experiencing an unprecedented growth. However, the transmission issues caused by heterogeneous access and dynamic traffic are still challenging 5G communications. The Internet Engineering Task Force (IETF)'s Multipath Transmission Control Protocol (MPTCP) can aggregate bandwidth and balance traffic across multiple subflows in a heterogeneous network environment. However, in order to support delivery of high quality 5G media services, researchers should also address MPTCP's inefficient data scheduling to heterogenous sub-paths, consideration of multiple criteria, including energy consumption and its inconsistent behavior when employed along with the Dynamic Adaptive Streaming over HTTP (DASH) adaptive application layer protocol. To address these issues, we propose a Q-Learning driven Energy-aware Data Scheduling (QLE-DS) mechanism for MPTCP-based media services. QLE-DS models the multipath scheduling as a Q-learning process and employs a novel quantum clustering approach to discretize the high dimensional continuous Q-table. An asynchronous framework is designed to improve the learning efficiency of QLE-DS. The simulation results show that QLE-DS performs better than other MPTCP scheduling algorithms in terms of flow completion time (FCT), retransmission rate, and energy consumption.