Cloud gaming (CG) enables the ubiquity of gaming by allowing users to access games on demand, anywhere and anytime, across heterogeneous devices connected to remote CG servers. As a result, the same AAA game can be played on an inexpensive smartphone at home, while commuting, or in public spaces through heterogeneous access networks (HANs) such as WiFi, 4G, and 5G. However, while CG broadens service accessibility, it does not guarantee sufficient delivery quality. In practice, HANs are often affected by stochastic conditions that degrade Quality of Service (QoS) factors such as round-trip time (RTT), packet loss (PL), and jitter. These degradations can immediately impair the CG service and, consequently, the user’s gaming session through reduced responsiveness, lower streaming quality, and disrupted interaction when the player is actively engaged in gameplay and expects uninterrupted performance. 

This thesis investigates the effect of QoS factors on CG services through Quality of Experience (QoE). QoE is a user-centred metric that reflects users' perceived quality and service acceptance. In the context of CG, it provides a way to understand how degradations in factors such as RTT, packet loss, and jitter are reflected in the gaming experience. Among the devices through which users may access CG services, this thesis focuses on two relevant cases, namely mobile cloud gaming (MCG), which is challenging to support over HANs due to the broader network variability of smartphone mobility and portability, and virtual reality cloud gaming (VRCG), which is challenging due to its stricter requirements on responsiveness, frame stability, and synchronization.

To address these challenges, this thesis proposes and validates methods for QoE measurement and prediction for MCG and VRCG under representative HANs conditions. First, controlled laboratory environments were established to emulate degradations for RTT, PL, random jitter, bursty jitter, and their combinations. To ensure standardised and repeatable experimentation, the thesis introduces ALTRUIST, a multi-platform orchestration tool that automates test execution, questionnaire control, traffic capture, data labelling, and control of experimental conditions across heterogeneous devices. Using these testbeds and ALTRUIST, four subjective studies were conducted, comprising two MCG and two VRCG datasets, with 116 participants recruited. The studies covered 37 network conditions for MCG and 28 for VRCG, and collected users’ ratings of overall QoE, video quality, audio quality, interactivity, and service acceptability, together with traffic traces for throughput analysis. Building on these measurements, the thesis uses Bayesian Networks to analyse probabilistic relationships among interactivity, video quality, audio quality, and overall QoE, providing further insight into how network degradation affects the gaming experience in both MCG and VRCG.

Lastly, this thesis proposes QoE objective models for MCG and VRCG using accessible network-level QoS factors. For MCG, three regression models are developed using two subjective datasets, showing that QoE can be estimated in real time from RTT, PL, random jitter, bursty jitter, and their combinations, and that these models remain valid across 60-120 FPS. The best MCG model, a spline regression, achieved MAE=0.186 and cross-validated MAE=0.281. For VRCG, three regression models are developed and validated using a combined dataset from two game types, showing that QoE can be estimated from network-level factors while remaining valid for shooter and casual games. The best VRCG model, a non-linear regression, achieved MAE=0.22 and cross-validated MAE=0.26. Together, these models provide practical methods for real-time QoE estimation, service monitoring, and QoS-threshold identification for CG provisioning over HANs.

The thesis research resulted in 10 publications and led to additional outcomes, including ITU-T contributions, a patent, ITU-T Recommendations, and conference papers -- all built on the datasets, models, and ALTRUIST tool developed in this thesis.