A Markovian model of coded video traffic which exhibits long-range dependence in statistical analysis

The purpose of this paper is to construct a Markovian model generating a sequence having almost the same statistical characteristics as a real video traffic process. We deal with measured traffic data from a certain video source. Taking scene changes into account we analyze the data and construct a model, called Markov–autoregressive (AR) model, composed of three submodels; a Markov transition model for scene changes, an AR model for spikes of scenes, and an AR model with random parameters for bit rate sequences of individual scenes. A simulation study shows that statistical characteristics of a sequence generated by this model are very similar to the actual video traffic. In particular, in the variance-time analysis and in the R/S analysis, these two sequences give similar estimates for the Hurst parameter and exhibit long-range dependence even though the model consists of only Markovian type, short-range dependent processes.