Super-resolution on Edge Computing for Improved Adaptive HTTP Live Streaming Delivery

Abstract

According to a Cisco report, mobile network speeds will more than triple by 2023, from 13.2 Mbps in 2018 to 43.9 Mbps in 2023. The average 5G connection speed is forecasted to reach 575 Mbps by 2023. This increase in bandwidth on mobile networks, along with the growing demand for streaming video content, has imposed unprecedented challenges on the backhaul networks that interconnect mobile networks to the Internet core. A trend to mitigate this problem has been to bring the source of content closer to the users, bringing it from the cloud to multi-access edge computing (MEC), therefore shifting the traffic pattern from the Internet core to the edge. In this article, we propose a framework called live streaming with super-resolution (LiveSR) that uses deep neural network-based super-resolution. In the LiveSR, live video moves in low resolution down to MEC and upscales to high resolution before being delivered to viewers over high-bandwidth mobile networks. We evaluate the proposed framework in scenarios with real 5G network traces. When we compare the proposed framework and a cloud-based video delivery system in a network defined by congested backhaul links, results show that the LiveSR framework can increase the quality of experience (QoE) in adaptive live videos by 49%, 51%, and 58% for the LoL+, BOLA, and L2A-LL adaptive algorithms, respectively. A considerable reduction in traffic in the backhaul is also recorded, ranging from 97.36% to 98.18%.