Machine Learning for Radio Propagation Modeling: A Comprehensive Survey

Abstract

With recent advancements in the telecommunication industry and the deployment of 5G networks, radio propagation modeling is considered a fundamental task in planning and optimization. Accurate and efficient models of radio propagation enable the estimation of Path Loss (PL) or Received Signal Strength (RSS), which is used in a variety of practical applications including the construction of radio coverage maps and localization. Traditional PL models use fundamental physics laws and regression-based models, which can be guided with measurements. In general, these methods have small computational complexity and have been highly successful in attaining accurate models for settings with trivial environmental complexity (e.g., clear weather or no clutter). However, attaining high accuracy in radio propagation modeling at complex settings (e.g., an urban setting with many buildings and obstacles) has required ray tracing, which computationally complex. Recently, the wireless community has been studying Machine Learning (ML)-based modeling algorithms to find a middle-ground. ML algorithms have become faster to execute and, more importantly, more radio data measurements have become available with the increased deployment of wireless devices. In this survey, we explore the recent advancements in the use of ML for modeling and predicting radio coverage and PL.

Keywords

• Path loss
• machine learning
• radio propagation
• wireless channel modeling
• neural networks