Long-term spatio-temporal RF-EMF exposure assessment in a sensor network Sam Aerts 1 , Joe Wiart 2 , Luc Martens 1 & Wout Joseph 1 1 Department of Information Technology, Ghent University / imec, Ghent, Belgium, 9052 2 LTCI, Chaire C2m, Télécom ParisTech, Institut Mines-Télécom, Paris, France, 75013 BioEM2018, Piran, Portoroz, Slovenia, Jun 25 - 29, 2018 Summary Our exposure to environmental radiofrequency (RF) electromagnetic fields (EMF) at a given location is inherently dynamic due to the constantly-changing nature of both our environment as well as the telecommunications networks present in it. More than a year of measurement data was collected in a fixed low-cost urban exposimeter network and analysed to build a spatio-temporal surrogate model of the exposure to environmental telecommunications signals. We observed that by taking into account the moment of the measurement in the modelling the accuracy of the resulting surrogate model in the area under study was improved by up to 50% compared to models that neglected the daily temporal variability of the RF signals. Introduction Human exposure to environmental radiofrequency (RF) electromagnetic fields (EMF) is not at all constant in time, due to environmental changes and variations in the number of active users (as well as the nature of their activity) in telecommunications networks. However, previous studies that aimed to characterize this exposure have tended to ignore the temporal dimension. Here, the impact of the temporal variability of telecommunications signals on outdoor RF-EMF exposure characterization was investigated. Measurements of three downlink telecommunications signals – i.e., from base station to user device – were collected during more than a year in a low-cost exposimeter network within an urban setting. From this vast set of data, full spatio- temporal surrogate models of the exposure in the area under study to the considered RF signals were built. Furthermore, global profiles of the daily variation of the signals were composed and used to quantify the improvement of adding the temporal dimension to established RF-EMF surrogate modelling techniques. Materials & Methods Monitoring network In the EU-FP7 LEXNET project, RF-EMF exposimeters were added to the SmartSantander Internet-of-Things (IoT) platform (http://smartsantander.eu/) [Diez2014], covering an area of 0.4 km by 1.4 km. The exposimeters were developed for large-scale deployment, thus as cost-efficiently as possible, and measure the environmental exposure (quantified by the electric field strength E, in V/m) induced by the three most-used telecommunications technologies (at that time, these were GSM at 900 MHz (GSM900) and 1800 MHz (GSM1800), and UMTS at 2100 MHz). Frequency bands specifically used by fourth generation (4G) Long Term Evolution (LTE) were not included, as this technology was not yet in use when the exposimeters were installed in 2014 [Diez2014]. Each of the considered frequency bands were alternately selected using an RF switch in the exposimeter, and the nominal sample collection time (one value for each band) was either 5 or 10 minutes, depending on the specific exposimeter. The data used in this study were collected during 14 months – between 9th of December, 2015, and the 15th of February, 2017. Temporal profiles For the three considered telecom signals, the average trends of the signal strength were identified over a day (i.e., the variation of the signal between 00:00 and 23:59, averaged over all measurement days) or a week (i.e., the daily variation depending on the day of the week). In order to obtain a smoother profile, and because the sampling time was not the same for all exposimeters, hour-aggregated averages were used in this analysis – i.e., for each day, the (24) average electric-field values captured between HH:00 and HH:59 (with HH = 00 to 23) were calculated. Furthermore, to rule out any potential bias due to long-term variations in the signal (due to, for example, changes in the network infrastructure), we used the relative variation of the signal compared to the daily average (η). Finally, by aggregating and averaging all these relative values (over the entire 14-month