Spatio-temporal Mobile Data Traffic Modeling Using Fourier Transform Techniques

The exponential growth in mobile data traffic is forcing telecom operators to invest on new infrastructures. But, devising techniques for optimum network utilization, which can be provided by traffic modeling, has the potential to reduce investment cost. Modeling traffic variation in different service areas and time is also applicable for energy efficient network planning, understanding customers’ data traffic usage behavior and dynamic resource allocation. In this paper, based on a week data collected from 734 Base Stations of an operator in Addis Ababa, Ethiopia, we model the real traffic in space and time with a tunable accuracy. Firstly, a rectangle that can inscribe the geographical area of the city is selected and divided equally into N by N smaller groups. Then, to understand the temporal behavior of the data, the time-series data traffic of each group is transformed to spectral domain by using Fast Fourier Transform, where it is observed that all groups have the same four major frequency components but with different magnitude of coefficients and phases. Secondly, matrices corresponding to coefficient and phase values are transformed from spatial to spectral domain by applying Two Dimensional Discrete Cosine Transform. In the spectral domain higher frequency components that contain less information are removed out and the remaining are used for the reverse transform. The different application areas of the model for the operator require different level of accuracy which, in turn, is dependent on the level of frequency components truncation. As a result, we have developed a relation between model performance and truncation level making the model to be tunable around accuracy.