Which Statistical Distribution Best Characterizes Modern Cellular Traffic and What Factors Could Predict Its Spatiotemporal Variability?

Abstract

Spatiotemporal characterization of user traffic is the first step in designing, optimizing, and automating a mobile cellular network. While it is well known that voice telephony follows Poisson distribution, the distribution of SMS and Internet data usage along with voice calls and the factors influencing the distribution is still an open question. We characterize the distribution of multi-faceted cellular traffic while identifying the factors influencing the parameterization of the distribution. Eight latent features that play a statistically significant role to characterize the traffic distribution variations over time and space are determined by leveraging a large real dataset. The features used to characterize the dynamism of the traffic distribution are points of interest, day of the week, special events, and region. The results show that generalized extreme value distribution best describes SMS, call, and Internet activity and it does not change with spatiotemporal features. Also, traffic distribution is not stationary. Insights gained from this analysis can pave the way toward more precise and resource efficient planning, designing, and optimization of future cellular networks.