Modeling MPEG Coded Video Traffic by Markov-Modulated Self-Similar Processes

Markov modulated self-similar processes are proposed to model MPEG video sequences that can capture the LRD (Long Range Dependency) characteristics of video ACF (Auto-Correlation Function). The basic idea is to decompose an MPEG compressed video sequence into three parts according to different motion/content complexity such that each part can individually be described by a self-similar process. Beta distribution is used to characterize the marginal cumulative distribution (CDF) of the self-similar processes. To model the whole data set, Markov chain is used to govern the transitions among these three self-similar processes. In addition to the analytical derivation, initial simulations have demonstrated that our new model can capture the LRD of ACF and the marginal CDF very well. Network cell loss rate using our proposed synthesized traffic is found to be comparable with that using empirical data as the source traffic.