A survey of statistical source models for variable-bit-rate compressed video

Abstract.

It is predicted that, in the near future, the transport of compressed video will pervade computer networks. Variable-bit-rate (VBR) encoded video is expected to become a significant source of network traffic, due to its advantages in statistical multiplexing gain and consistent video quality. Both systems analysts and developers need to assess and study the impact these sources will have on their networks and networking products. To this end, suitable statistical source models are required to analyze performance metrics such as packet loss, delay and jitter. This paper provides a survey of VBR source models which can be used to drive network simulations. The models are categorized into four groups: Markov chain/linear regression, TES, self-similar and i.i.d/analytical. We present models which have been used for VBR sources containing moderate-to-significant scene changes and moderate-to-full motion. A description of each model is given along with corresponding advantages and shortcomings. Comparisons are made based on the complexity of each model.