A TW approach to MPEG traffic modelling using Markov chains

Abstract

Traffic modelling, as well as its specification, is one of the most important aspects in the study of communication network performance, and is a very active research field. In this paper we propose an MPEG video traffic model which considers that an MPEG movie can be modelled by means of a Markov chain with n states. Each state corresponds to a first-order autoregressive stochastic process, so that when the chain is in its k state, the sizes of the successive frames forming the MPEG movie are chosen according to the process AR(k). To construct this model we need to determine n, the number of states (AR processes) in the Markov chain, estimate the parameters of the AR processes corresponding to these states, and estimate the probability transitions in the Markov chain (i.e., the probability of moving from one state to another). In this paper, we deal with the first of the problems mentioned above, i.e., how to determine the n phases of the Markov chain in question. In fact, we show how several well known algorithms for detecting scenes can be used together with clustering techniques to determine the approximate number of different states or phases that can be found in a movie. We illustrate the methodology by analysing a case study.