An accurate scene-based traffic model for MPEG video stream

In this paper, we propose an accurate scene-based traffic model for MPEG video stream. The autocorrelation structure of video stream is modeled by two Auto-Regress (AR) processes. We also take the correlation in GOP (Group of Picture) into account. According to the size of I frame, the size of B frame and P frame in the same GOP are determined by a lognormal distribution. As for the marginal distribution of the MPEG video stream, we use normal mixture distribution to fit it, and the parameters of normal mixture distribution are estimated through an Expecting Maximum (EM) algorithm. At last, the accuracy of the proposed model is validated with simulation experiments and the application of this model in network QoS (Quality of Service) evaluation.