Macromodelling analysis of VBR video in an ATM environment

A single video source with low-medium activity can be modeled as an N-state Markov chain. Each state of the model represents a quantized number of cells/frame. The transitional probabilities between the states are determined by fitting the mean and autocorrelation function of the model with those of the real data derived from the coded video sequence. For sequences with scene changes, an additional state is added into the model. Any state, with probability proportional to the percentage of the frames with scene change, can make a transition towards the scene cut state. The multiplex model of M such sources is then a linear Markov chain with N×M states, and the state probability functions derived from the generating function of a single source. The transitional probabilities are represented by a truncated discrete normal distribution. The authors show that both the multiplexed model and the multiplex of M single source model give similar performance using real video data, at both cell and burst scale levels