Speaker
Description
The goal of this contribution is to develop a deep learning framework that combines real-world measurement data with publicly available but incomplete base station information to predict EMF exposure in complex urban environments. Previous research has explored integrating real-world measurements with partially available public data to train ANN models for urban EMF exposure prediction. For example, early work demonstrated the feasibility of training ANNs on simulated data to predict urban exposure levels. Later study expanded on this by incorporating drive test data into ANN training. Additionally, methods such as feature selection based on propagation models and Gram-Schmidt orthogonalization have been introduced to optimize ANN input features. Building on these advancements, this study proposes a deep learning method that uses the geospatial and base station antenna data as the input, to predict the average electric field (E-field) level within a given area. Notably, this approach relies only on real world measurements and publicly available datasets, without the need for extensive simulation-based input.
