Prediction of MPEG-coded video source traffic using neural networks

Due to the character of the original source materials and the nature of batch digitization, quality control issues may be present in this document. Please report any quality issues you encounter to digital@library.tamu.edu, referencing the URI of the item.Includes bibliographical references (leaves 202-207).Issued also on microfiche from Lange Micrographics.One of the most popular standards used for storing, delivering and viewing real-time streaming media over the Internet is MPEG. The time-series representing frame/VOP sizes of an MPEG sequence is extremely noisy and has very long-range time dependency. Present research aims to provide a generalized solution to the problem of predicting the frames/VOP sizes of MPEG-coded real-time video streams for single and multiple steps ahead. In order to incorporate the nonlinear behavior of time-series representing frame or VOP sizes of MPEG-encoded streams in mathematical models, Artificial Neural Networks (ANNs) are used extensively in this research. The final model for single-step-ahead prediction consists of one RMLP network for prediction of I-VOPs and two FMLP networks for prediction of P- and B-VOPs, respectively. Each of these networks has external indicators derived from the time-series. Multi-step-ahead prediction of individual frame/VOP sizes is found to be very inaccurate. A moving average of the sizes of frames/VOPs is generated from the individual frame sizes. A multi-step-ahead prediction scheme based on RMLP networks is developed and used to perform two-step-ahead and four-step-ahead prediction of moving average of the frame/VOP sizes with some success. This research demonstrates that video source traffic encoded using MPEG standards can be predicted single-step-ahead with reasonable amount of accuracy using neural networks as predictors. Use of two-step-ahead and four-step-ahead prediction schemes to predict the time series of the moving average of size of frame/VOPs is a novel concept introduced in this research. This predictive capability should be useful in the more effective bandwidth allocation mechanism and in the development of a control scheme to control the multimedia traffic over wide area networks