Deep Learning Propagation Models over Irregular Terrain

Abstract

Accurate path gain models are critical for coverage prediction and radio frequency (RF) planning in wireless communications. In many settings irregular terrain induces blockages and scattering making it difficult to predict the path gain. Current solutions are either computationally expensive or slope-intercept fits that do not capture local deviations due to terrain variation, leading to large prediction errors. We propose to use machine learning to learn path gain based on terrain elevation as features. We implement different neural network architectures with dense and convolutional layers that could include effects difficult to describe with traditional models (e.g. back scatter). We test our framework on an extensive set of measured path gain data and consistently predict with 5 dB Root Mean Squared Error, an 8 dB improvement over traditional slope-intercept solutions.