Mobile real-time communication (RTC) applications, such as video calls, cloud gaming, and VR/AR, require consistently high throughput and low latency to ensure good quality of experience (QoE). Although advanced 5G networks are widely regarded as key ...
Mobile real-time communication (RTC) applications, such as video calls, cloud gaming, and VR/AR, require consistently high throughput and low latency to ensure good quality of experience (QoE). Although advanced 5G networks are widely regarded as key enablers of seamless mobile RTC, real-world deployments still suffer from frequent QoE degradation even when sufficient bandwidth is available. A fundamental reason is that today’s radio access networks (RANs) lack awareness of application QoE and primarily optimize radio resource efficiency.
In this dissertation, we present a QoE-driven RAN design for RTC applications that leverages the programmability of the Open RAN architecture. Our approach focuses on two goals: first, identifying the root causes of QoE degradation in existing RAN mechanisms, and second, designing RAN-side mechanisms that improve application QoE without compromising radio resource efficiency.
We first introduce PAVE, a RAN-side solution that breaks the long-standing trade-off between spectral efficiency and RTC latency. Despite abundant 5G bandwidth, video calls often experience poor quality and excessive latency during channel fluctuations, not due to congestion or limited capacity, but due to non-congestive delays within the RAN. These delays originate from general-purpose transmission procedures that prioritize reactive scheduling and static timer-based retransmissions over application latency. Existing solutions either target congestion-induced delays or reduce spectral efficiency to mitigate non-congestive delays. PAVE exploits two intrinsic properties of RTC traffic—periodic frame generation and deadline-driven urgency—to improve QoE while preserving efficiency. For practical deployment, PAVE infers these properties within the RAN, enables resource preallocation beyond strictly periodic traffic patterns, and incorporates selective retransmission and packet skipping that strategically leverages application-layer recovery. Implemented in an Open-RAN-compliant RAN Intelligent Controller, PAVE significantly improves tail frame rates and reduces video stalls in real-world experiments.
Next, we present QCON, a QoE-driven multi-connectivity framework for mitigating wireless bandwidth fluctuations. While existing multipath transport solutions can address link variability, they typically require multiple cellular subscriptions, which are impractical for most mobile users. QCON instead leverages 5G multi-connectivity, which allows simultaneous connections to multiple base stations, such as 5G and 4G, and is already deployed in commercial networks. Our measurements show that operator policies often restrict multi-connectivity to preserve backup links, leaving RTC applications vulnerable to single-link fluctuations. QCON optimizes application QoE while respecting such policies by introducing a RAN-side QoE monitor, multi-link scheduling that jointly optimizes QoE and radio resource efficiency, and priority-based re-injection that exploits RAN link recovery mechanisms. Experimental results from an Open-RAN testbed demonstrate substantial improvements in achieved bitrate and tail frame rates compared to existing multi-link scheduling schemes.
Finally, we propose Apollo, a semantic queue management system designed to prevent bursty stalls during short-term capacity shortages. In capacity-limited 5G networks, packet queues frequently build up, causing frames to miss playback deadlines or become undecodable due to lost dependencies. However, existing RANs enforce in-order delivery regardless of semantic validity, delaying even frames that could still be decoded on time. Apollo addresses this limitation by prioritizing semantically important packets and discarding those that no longer contribute to playback quality. We tackle two key challenges: inferring packet importance using only RAN-visible information by leveraging standard RTP header extensions that expose frame dependencies, and enforcing semantic-aware scheduling within rigid RAN protocol constraints. Evaluation on a Linux netem-based 5G emulator shows that Apollo significantly improves frame rates and reduces video freeze time compared to baseline approaches.