Evaluating Perceptual Video Quality for Mobile Clients in 802.11n WLAN

In this thesis, we characterize the performance of HD video streaming in 802.11n WLANs under user mobility. We conducted experiments in QuRiNet, a large-scale outdoor wireless testbed that experiences little electromagnetic interference. We observe the variation in video quality with the variance of both speed of a mobile user and his distance from an AP. Using subjective scores and objective video quality assessment metrics, we build a non-linear regression model to estimate video quality based on user speed and distance. An ensemble machine learning kernel, bagging, is used in conjunction with Reduced Error Pruning Decision Trees to build a non-linear prediction model that scores 69% correlation with video quality. Overall, we find that distance has larger impact on video quality than speed. However, the physical factors such as speed and distance cannot be used in isolation to estimate video quality accurately.