Unified and non-parameterized statistical modeling of temporal and spatial traffic heterogeneity in wireless cellular networks

Understanding and solving performance-related issues of current and future (5G+) networks requires the availability of realistic, yet simple and manageable, traffic models which capture and regenerate various properties of real traffic with sufficient accuracy and minimum number of parameters. Traffic in wireless cellular networks must be modeled in the space domain as well as the time domain. Modeling traffic in the time domain has been investigated well. However, for modeling the User Equipment (UE) distribution in the space domain, either the unrealistic uniform Poisson model, or some non-adjustable model, or specifc data from operators, is commonly used. In this paper, stochastic geometry is used to explain the similarities of traffic modeling in the time domain and the space domain. It is shown that traffic modeling in the time domain is a special one-dimensional case of traffic modeling in the space domain. Unified and non-parameterized metrics for characterizing the heterogeneity of traffic in the time domain and the space domain are proposed and their equivalence to the inter-arrival time, a well accepted metric in the time domain, is demonstrated. Coefficient of Variation (CoV), the normalized second-order statistic, is suggested as an appropriate statistical property of traffic to be measured. Simulation results show that the proposed metrics capture the properties of traffic more accurately than the existing metrics. Finally, the performance of LTE networks under modeled traffic using the new metrics is illustrated.