A Data-Driven Approach for Designing Wireless Signal Propagation Model

In the study of wireless signal propagation, how to choose a suitable location for a base station of signal transmission is a noteworthy problem. The empirical models were often used to tackle this problem. However, the features used in the empirical models lacked the global information of real-world scenarios. To address such issues, we considered that the process of establishing a wireless signal propagation model was essentially a data-driven function fitting. Then, we utilized machine learning to model the existing features and estimate the Reference Signal Receiving Power (RSRP) between candidate base stations and the users, and thus to determine a better location for the base station. Futhermore, we had also designed some new features for modeling based on the surface information. In particular, we considered the impact of different types of surface information in the two-point paths between the candidate base stations and the users, and designed an effective approach to extract these features. Our experiments on different scales of data sets showed that our machine learning models had a significant improvement over the empirical ones.