Deep Learning for VBR Traffic Prediction-Based Proactive MBSFN Resource Allocation Approach
Abdennaceur Ghandri, Houssem Eddine Nouri, Maher Ben Jemaa
Abstract
To cope with the growing demand for multimedia services, the 3rd Generation Partnership Project (3GPP) has introduced the Multimedia Broadcast/Multicast Service (MBMS) to better distribute multimedia content over mobile network, particularly in the context of the Multicast Single Frequency Network (MBSFN). The conventional approach specified in the 3GPP standard adopts a semi-static resource allocation mechanism for MBMS services based on their Quality of Service (QoS) parameters rather than real-time traffic behavior. This approach is inefficient and unsuitable for Variable Bit Rate (VBR) video traffic, such as live video streaming. In this paper, we propose a proactive resource allocation approach that dynamically adjusts the allocation of subframes in an MBSFN area to keep pace with fluctuating VBR traffic. Our goal is to maximize the overall system utility by striking a balance between fairness in resource sharing and throughput maximization while reducing the MBSFN resource waste. The main idea of the proposed scheme is to periodically reallocate MBSFN subframes based on a Deep Learning (DL) prediction model of VBR traffic behavior. Simulation results show that our developed approach, in comparison to other sophisticated scheme, can effectively improve MBSFN resource allocation while considering QoS requirements and fairness constraints between unicast and multicast traffic.