Traffic characterisation and modelling of VBR coded MPEG sources

This paper intends to study the characteristics of VBR MPEG sources and performs models for their traffic behaviour. Our analysis focuses on long-time MPEG video films, approximately 27 min. Known probability density functions (pdf) are presented which fit well with real the pdf of I, P, B and Group of pictures (GOP) frames. In order to approximate the traffic behaviour, we study the autocovariance fuctions of the previous frames and we propose two different models so as to achieve the necessary fit. The first model is based on GOP frame layer which is able to approximate the cell probability loss at utilisation about 0.7-0.85 but it slightly overestimates the frame loss probability. The other model correlates the error of I, P and B AR models without spoiling the i.i.d. (independent identical distributed) property. Our results show that the latter model approximates well the cell as well as the frame loss probabilities.