Load-adjusted video quality prediction methods for missing data

Abstract

A polynomial fitting model for predicting the RTP packet rate of Video-on-Demand received by a client is presented. This approach is underpinned by a parametric statistical model for the client-server system. This model, namely the PQ-model, improves the robustness of the predictor in the presence of a time-varying load on the server. The advantage of our approach is that if we model the load on the server, we can then use this model, and any insights gained from it, to improve RTP packet rate predictions. A second advantage is that we can predict how the server will behave under previously unobserved loads - a tool which is particularly useful for network planning. For example, we can accurately predict how the system will behave when the load is increased to a previously unobserved value. Thirdly, the PQ-model provides accurate predictions of future RTP packet rates in scenarios where training data is unavailable.