VBR MPEG Video Traffic Dynamic Prediction Based on the Modeling and Forecast of Time Series

The variable-bit-rate traffic characteristic brings a large complication to the utilization of network resources, especially bandwidth. To solve this problem, this paper proposes a dynamic prediction scheme of MPEG video traffic. We first advance an initial reversible transformation to the original video GoP sizes and then design an algorithm based on the ARMA process modeling to fit the transformed smooth sequence. At last we take advantage of the multi-step forecast of the series to perform a dynamic prediction for original GoP sizes. An experimental study has shown the good effect of the scheme in this paper. With the help of our prediction scheme, the waste of bandwidth will be limited effectively and the network bandwidth allocation and renegotiation can be developed to be very dynamic and intelligent.