In an increasingly connected world, wireless networks' monitoring and characterization are of vital importance. Service and application providers need to have a detailed understanding of network performance to offer new solutions tailored to the needs of today's society. In the context of mobility, in-vehicle infotainment services are expected to stand out among other popular connected vehicle services, so it is essential that communication networks are able to satisfy the Quality of Service (QoS) and Quality of Experience (QoE) requirements needed for these type of services. This paper investigates a multi-layer network performance monitoring architecture at the edge providing QoS, QoE, and localization information for vehicular video streaming applications in realtime over 5G networks. In order to conduct field trials and show test results, Mobile Network Operators (MNOs)' 5G Standalone (SA) network and Multi-access Edge Computing (MEC) infrastructure are used to provide connectivity and edge computing resources to a vehicle equipped with a 5G modem.

The increasing consumption of video streams and the demand for higher-quality content drive the evolution of telecommunication networks and the development of new network accelerators to boost media delivery while optimizing network usage. Multi-access Edge Computing (MEC) enables the possibility to enforce media delivery by deploying caching instances at the network edge, close to the Radio Access Network (RAN). Thus, the content can be prefetched and served from the MEC host, reducing network traffic and increasing the Quality of Service (QoS) and the Quality of Experience (QoE). This paper proposes a novel mechanism to prefetch Dynamic Adaptive Streaming over HTTP (DASH) streams at the MEC, employing a Machine Learning (ML) classification model to select the media segments to prefetch. The model is trained with media session metrics to improve the forecasts with application layer information. The proposal is tested with Mobile Network Operators (MNOs)' 5G MEC and RAN and compared with other strategies by assessing cache and player's performance metrics.

No abstract