Characterization of variable bit rate video traffic as vector autoregressive processes

In order to efficiently transmit the coded video infermation in asynchronous transfer mode (ATM) networks, we need to characterize the output bit-rate sequences of video coders. We characterize JPEG-like and H.261-like VBR video traffic sources without scene changes at the slice level by introducing vector autoregressive (VAR) processes. We show that slice-level video traffic sources generally exhibit periodic mean variations and that residual slice-level bit-rates also greatly vary depending on their positions in a picture. Empirical results show that the VAR models can fit the VBR video traffic sources very well in most cases.