Coverage Area Estimation Using a Multi-Branch 1D Convolutional Neural Network

Abstract

Coverage estimation in cellular networks is crucial for network planning and optimization. Although traditional ray tracing models radio wave propagation, it faces limitations in large-scale applications due to high computational costs. Deep learning-based methods also estimate signal propagation accurately but are restricted by data requirements. To address this issue, this study generates large-scale synthetic datasets using Uniform Theory of Diffraction (UTD)- based ray tracing simulations. The proposed method analyzes electromagnetic propagation paths from 3D digital terrain maps to produce 2D coverage maps and uses a \"Multi-Branch Coverage Estimation Network\" to predict signal propagation quickly and accurately. The model can analyze an area with a 1 km diameter in 80 seconds, offering a significant speed advantage over conventional methods. The RMSE remains below 0.07 dB for all points and below 0.03 dB for 95% of them. In this way, high accuracy in wireless network planning can be achieved without the need for actual measurements.