Downlink Performance Approximation of Cellular Networks via Stochastic Geometry and Machine Learning

Stochastic geometry facilitates to comprehend correlations between operation variables, but has a limit that it is applicable only to limited environment because of simplified modeling of real network operation. On the other hand, simulation can analyze performance in various environments, but it is difficult to comprehend correlation between operation variables. This paper parameterizes the downlink SINR (Signal to Interference plus Noise Ratio) performance of cellular networks, and proposes the method to learn the performance according to path loss exponent, shadowing, and thermal noise, via stochastic geometry and machine learning. In addition, it is demonstrated that the proposed performance can be applied to the design of base station (BS) density and transmit power.