Distribution of loss volume and estimation of loss for aggregated video traffic

Estimation of the packet loss for a video transmission over a communication network is an important step for assessing the QoS perceived by the users. The loss probability is particularly important, but also the characteristics of the loss periods have a significant effect, since bursty losses influence the perceived QoS in a different way than uniformly distributed losses. In this paper, the video traffic is modeled as a discrete multivariate Gaussian process, taking the correlation between consecutive frames into account. Exceedances of frames sizes over a threshold constitute loss periods, and the distribution of the length of a loss period is found in an earlier work. Here, the distribution of the loss volume in a loss period is pursued using a numerical approach, giving the packet loss in the bufferless case. It is then shown how the loss volume can be employed to estimate the packet loss in a bottleneck node with small buffers. Furthermore, relations between loss period characteristics for discrete and continuous Gaussian processes are investigated. In particular, it is found that a relation between the length and volume of excursions over high thresholds for a continuous process is valid for a discrete process as well.