Long-term prediction of MPEG video traffic for broadband cable networks

In this paper, we evaluate the accuracy of a new long-term prediction model for MPEG video traffic. With this model, a video headend in a broadband cable network can forecast bandwidth usage well ahead of time and this can help minimize packet losses during periods of peak bandwidth demands or prefetching of video data. Although the model can be applied to a wide variety of compressed video traffic, we apply our model to several high-definition MPEG-4 AVC videos. First, we estimate the Hurst parameter values, which is used to investigate the long-range dependency or self-similarity of the videos. This leads us to design a new algorithm to predict the size of the I-, P- and B-frames, and the group of pictures (GOP) in the long term. Although long-term prediction may potentially suffer degraded prediction accuracy compared to short-term prediction, we will show that this is not always the case, especially when the prediction is applied to videos encoded with high quality resolution. In addition, we will demonstrate that the bit rate variability of multiplexed videos tends to smooth due a weaker long-range dependency.