The spatial distribution of base stations (BSs) and traffic demands is essential for efficient network planning and BS sleeping, which are key elements of green cellular networking. This paper investigates their statistics, relation, and modeling, based on large-scale measurement data from commercial cellular networks. The spatial distribution of BSs shows not only high nonuniformity over a region but also diverse patterns in different regions, and thus the widely used homogeneous Poisson point process can only approximate the BS patterns in a specific small area. Therefore, the inhomogeneous PPP (IPPP), in particular, the Cox point process with spatially varying intensity is used to model the BS distribution over any spatial scale.
To model the intensity distribution of the IPPP, we exploit the relation, shown to be sublinear, between the BS distribution and the peak hour (PH) traffic density, based on the finding that the PH traffic density can be approximated by a log-normal distribution. Finally, we propose a spatial modeling framework for network simulations, and discuss potential applications of the proposed spatial distribution model of the BS patterns and the PH traffic density.