A Machine Learning Approach to Predicting Coverage in Random Wireless Networks

There is a rich literature on the prediction of coverage in random wireless networks using stochastic geometry. Though valuable, the existing stochastic geometry-based analytical expressions for coverage are only valid for a restricted set of oversimplified network scenarios. Deriving such expressions for more general and more realistic network scenarios has so far been proven intractable. In this work, we adopt a data-driven approach to derive a model that can predict the coverage probability in any random wireless network. We first show that the coverage probability can be accurately approximated by a parametrized sigmoid-like function. Then, by building large simulation-based datasets, the relationship between the wireless network parameters and the parameters of the sigmoid-like function is modeled using a neural network.