Parameter estimation and optimization techniques for discrete-time semi-Markov models of H.264/AVC video traffic

Abstract

We investigate a variety of known and new approaches for the estimation of the parameters of discretetime semi-Markovian traffic models. We focus on modeling video traffic, since the accurate representation of its longterm autocorrelation is a challenge to the parameter estimation methods. The modeling techniques are applied to sample H.264/AVC-encoded video traces. We study their ability to reflect the autocorrelation and variability of the original traffic and also the delay probabilities of a resulting SMP/GI/1 queueing system. The delay probabilities are determined both by simulation and verified analysis of the queue.

FEC Recovery Performance for Video Streaming Services Based on H.264/SVC

“…In general, video traffic exhibits a long-term correlation nature, which is hardly modeled with traditional Markovian arriving processes. In (Kempken et al 2008), the authors considered discrete-time semi-Markov models of H.264/AVC video traffic. Focusing on the short term autocorrelation and the preservation of the mean value of the distribution of the size of group of pictures (GoP), the parameters of a discrete-time batch Markovian arrival process are optimized by simulated annealing approach.…”

Modeling self-similar traffic over multiple time scales based on hierarchical Markovian and L-System models

“…Several fitting procedures have been proposed in the literature for estimating the parameters of MMPPs from empirical data [7][8][9][10][11][12][13][14]. However, most procedures only apply to 2-MMPPs, which can capture traffic burstiness but have an insufficient number of states to reproduce variability over a wide range of time scales.…”