Self-Similarity and modeling of LTE/LTE-A data traffic

Large- and Small-Scale Modeling of User Traffic in 5G Networks

“…3, we can observe how closely the two proposed models approach the desired Hurst parameter H, provided by the large-scale component. The blue line shows the evolution of H with the average load µ X , as observed in previous research [8]. We depict an inverted exponential increase of H as µ X increases, in order to feed the models with a smooth trend.…”

“…The average load E{X(t)} = µ X of a typical mobile base station exhibits a strong daily pattern, with valleys in the night along with midday and afternoon peaks. Several models have been proposed to foresee the average hourly load based on these observed patterns [4], [8]. The shape of this pattern is similar for most base stations, although different variants exist for residential, office, transport, or entertainment areas [4].…”

“…Although this parameter actually indicates the independence of the traffic phenomena on the time scale, it has been observed to also vary according to a daily pattern. In fact, a direct relation between the average load and the Hurst parameter has been suggested, as it is observed that mobile traffic is highly self-similar at the peak hours (H ≈ 0.9), whereas at the valley hours the self-similarity is less noticeable (H ≈ 0.65) [8].…”

“…The small-scale component W S (t, µ X , M, H) of X(t) models the traffic variation in the order of a few scheduling intervals, in contrast to the slow evolution of the large-scale components. Owing to the self-similar nature of the mobile data traffic [8], we compare two self-similar models for the small-scale component: a naïve superposition model and our suggested synthetic model. The superposition model is simple and intuitively fits the source of the traffic, but it fails at representing the traffic when combined with the large-scale component.…”

“…However, Poisson models are known to ignore the long-range dependency of the traffic fluctuation that is present in many communication networks [7]. Models based on self-similar processes can be used to capture this long-range dependency, which has been successfully applied to 4G traffic [8].…”