Abstract
In the best-effort IP network, quality of service (QoS) cannot be guaranteed, and thus packets could possibly be delayed or lost. Packet delay/loss will inevitably degrade the perceptual quality of real-time multimedia-over-IP service, such as Voice-over-IP (VoIP), Video-on-Demand (VoD), etc. In general, packet loss/delay exhibits temporal dependence. In order to efficiently conduct error recovery/concealment and improve the perceptual quality of the transmitted multimedia contents, packet loss/delay has to be precisely modeled. Different mathematical models, such as Bernoulli Model, Gilbert Model, Extended Gilbert Model, have been proposed to model network trace. However, none of them is able to precisely model the networked multimedia trace like the General Markov Model (GMM), which has rarely been applied in practice due to massive computational resource requirements. With the developments of modern computing hardware and parallel processing algorithms, GMM is becoming computationally feasible. In this paper, in order to obtain network traces for real-time VoD transmission, different connections have been setup to simulate a real VoD system. Data packets have been transmitted between the server and clients under RTP/UDP/IP protocol stack. Different models have been applied to analyze and model the obtained video transmission network traces. Specifically, a 6-state GMM has been applied to analyze the network trace, and the parameterized model has been obtained for further error recover/concealment. Compared to the other models, GMM offers the best modeling precision, in terms of loss-run distribution (LSD) and Forward Error Correction (FEC) performance prediction. The parameterized GMM is very useful to model and analyze network traces and further improve the QoS in multimedia-over-IP based on the modeling and analysis.