Prediction of RF-EMF Exposure by Outdoor Drive Test Measurements

Abstract

In this paper, we exploit the artificial neural network (ANN) model for a spatial reconstruction of radio-frequency (RF) electromagnetic field (EMF) exposure in an outdoor urban environment. To this end, we have carried out a drive test measurement campaign covering a large part of Paris, along a route of approximately 65 Km. The electric (E) field strength has been recorded over a wide band ranging from 700 to 2700 MHz. From these measurement data, the E-field strength is extracted and computed for each frequency band of each telecommunication operator. First, the correlation between the E-fields at different frequency bands is computed and analyzed. The results show that a strong correlation of E-field levels is observed for bands belonging to the same operator. Then, we build ANN models with input data encompassing information related to distances to N neighboring base stations (BS), receiver location and time variation. We consider two different models. The first one is a fully connected ANN model, where we take into account the N nearest BSs ignoring the corresponding operator. The second one is a hybrid model, where we consider locally connected blocks with the N nearest BSs for each operator, followed by fully connected layers. The results show that the hybrid model achieves better performance than the fully connected one. Among N∈{3,5,7}, we found out that with N=3, the proposed hybrid model allows a good prediction of the exposure level while the maintaining acceptable complexity of the model.