A generalized TES model for periodical traffic

The transform-expand-sample (TES) is a young, but already well-established traffic modeling technique. It creates a random-governed modulo-1 sequence which is smoothed and adapted to a given marginal distribution. The smoothing is realized by the so-called stitching function, which is demonstrated to have a crucial influence on the shape of the resulting simulated time series and its autocorrelation function. The main aim of this paper is to show how the choice of suitable (generalized) stitching functions provides a very attractive way to simulate sample paths that are visually indistinguishable from the given time series. The rather intuitive procedure is illustrated for the cases of periodic network utilization and the modeling of MPEG frame sizes.