Modeling and analysis for packet video

Abstract

The development of analytically simple, mathematically tractable video source models for the analysis of packet networks is described. As the bit-rate is not a constant, it is necessary to develop statistical models to measure the characteristics of the coded bit-rate sequence. Since model development and validation rely on the availability of appropriate experimental data, a complete system was set up for data collection. The bit rate profiles and the basic statistics of 30 independent video sources with various activity levels were collected. On a subframe basis, the autocovariance function of each of the 30 sequences shows a pseudoperiodic property. The actual shape of each function, however, depends on the scenes in that particular video segment. A finite-state machine representation was proposed to describe the probabilistic behavior of the aggregate packet-rate process when video sources are statistically multiplexed. With appropriate parameters, the model matches the empirical data very well and can be used for the development of protocols and control schemes for ATM (asynchronous transfer mode) networks.