H.264 video traffic modeling via Hidden Markov Process

The H.264 video coding standard includes new syntactic structures that allow efficient drift-free switching among precoded sequences at different bit-rates, making H.264 standard suitable for video streaming in time-varying channels. In network design, a good modeling of the video source is desirable to achieve a good dimensioning. A model of the traffic generated by a bitstream switching source requires considering not only variations of the video sequence activity, as it occurs in modeling classic VBR sources but also global variations of the average bit-rate. In this work we study a synthetic source constituted by a Hidden Markov Process modeling a real bit-rate switching video source. Parameter estimation is performed by the Expectation-Maximization algorithm. Model accuracy is assessed by a comparison of the frame loss rate of a fixed size buffer filled with the synthetic source and with a real H.264 video source.