Understanding spatial relationships in resource usage in cellular data networks

We conduct a detailed measurement analysis to investigate the spatial characteristics of network resource usage using a large-scale data set collected in situ in a nationwide 3G cellular data network. The data set spans over thousands of base stations. We first characterize the spatial correlation in radio resource usage using different statistical techniques. The analysis shows existence of significant spatial correlation that varies during the day, peaking during the middle of the day and waning in the middle of the night. We also use the notion of spectral clustering to show how base stations can be clustered based on how correlated they are in terms of radio resource usage. We show that this produces spatially connected clusters. We also show that only a few clusters exist when clustered optimally. Finally, we use the concept of Granger causality to understand the underlying functional connectivity and flow of influence in the network. We show that roughly one-third of neighboring base station pairs exhibit statistically significant Granger causality, and long causal paths exist in the network. Our observations can lead to development of new techniques for network monitoring and resource management in future cellular data networks.