Abstract:
Modern wireless systems for mobile communication use electromagnetic waves to transmit
information over the air, enabling seamless connectivity for a wide range of devices. However,
one of the key challenges in wireless network planning and optimization is accurate prediction of
received signal strength (RSS), especially in complex urban environments. This paper presents a
novel deep learning-based framework for RSS prediction that integrates geospatial data, elevation
profiles, and engineered system parameters. The model leverages satellite imagery and digital
elevation models (DEMs) derived from Sentinel-1 radar data to account for the impact of terrain
and environmental clutter on wireless signal propagation. We presented a custom Convolutional
Neural Network (CNN) architecture that processes multi-channel satellite images, while additional
features such as 3D distance and elevation data are incorporated into a fully connected network to
enhance prediction accuracy. We evaluated our model using real-world measurements collected
from a 5G test site in an urban environment. Our results demonstrate that the model significantly
outperforms conventional propagation models, including free space pathloss (FSPL), 3GPP urban
macro, and the alpha-beta-gamma model (ABG), with improvements in root mean square error
(RMSE) and mean absolute error (MAE). Furthermore, complexity analysis using floating point
operations (FLOPs) showed that the model achieves these improvements without excessive
computational cost. The findings in this study can help better design and implement robust
wireless communication networks with improved signal quality and capacity
