Characterization of MPEG scalable sources

Constant quality VBR video transmission causes a low utilization of network resources due to rare high-rate periods. Source scalability (i.e., the codec generates streams corresponding to several quality levels) is a promising approach to achieve a higher bandwidth utilization with a negligible reduction of the quality perceived by a human observer. To study bandwidth allocation for VBR scalable sources a characterization of the traffic generated by these sources is required. In this paper we develop and validate a Markovian model to characterize a scalable MPEG source. Our model is obtained by extending (without any additional state-space complexity) a Markov model previously developed for constant-quality MPEG sources. At any time instant, the model generates, with the same precision of the constant-quality model, the size of the highest quality pictures and approximates the size of the pictures corresponding to the other quality levels. By exploiting this model we show that the source scalability approach significantly increases the utilization while maintaining the quality of the video signal close to the highest quality for most of the time.