Adaptive prediction and smoothing of MPEG video in ATM networks

An adaptive prediction and smoothing scheme for MPEG video is investigated. The scheme is based on the characteristic that video traffic depends on the mode in which the pictures are coded. Since we get a high correlation of the signal by picture decomposition, an autoregressive model is used to predict the bit rate of a signal and is shown to be very effective. Two approaches for smoothing video signals subject to a delay constraint are investigated. One is to delay the cell streams by one picture, and then use the exact bit rate for smoothing. The other is to get a prediction of the bit rate and use the prediction results when smoothing. The performance of these approaches is examined in terms of the burstiness of video signals vs. the delay constraint