Adaptive Prediction of Variable Bit Rate MPEG Video Traffic for RCBR Network
Yong Li · Chinese Journal of Computers · 2002
Recently researchers have brought forward some models to describe the characteristics of video traffic. For instance, neural network, non-adaptive time-domain linear filtering, adaptive time-domain linear filtering, and wavelet-domain methods for video traffic prediction and resource allocation were reported. Among them, adaptive time-domain prediction using the least-mean-square (LMS) algorithm is of particular interest because of its simplicity and relatively good performance. Although it improves the utilization of networks, its prediction delay and slow convergence degrade its performance of prediction and further processing. In this paper authors exploit the periodical dependence characteristics of MPEG video and develop a novel coding model based LMS algorithm to predict the bandwidth required by the future frame and group of pictures (GOP). It integrates the fixed structural information of coding model into the prediction. For any kind of frames (I, P, B frames), the algorithm can modify its parameters automatically using the coding information without considering different frames separately. Compared to the LMS, the modified LMS can predict the bit rate of frames fast and accurately without any delay. Through the analysis of queuing of video traffic in buffer of a renegotiate constant bit rate network, prediction for dynamic allocating bandwidth exhibits good performance than fixed bandwidth allocation.