Authors: Sun, Q.; Wang, S.; Zhang, Y.;Liu, J.; Wiart, J.; Nait-Abdesselam, F.
20th European Conference on Antennas and Propagation (EuCAP), 19 – 24 April, 2026, Dublin, Ireland,
doi: 10.23919/EuCAP68105.2026.11612471
Abstract
This paper presents a deep learning based framework for studying uplink radio-frequency (RF) exposure in urban environments. Large-scale outdoor uplink measurements were conducted using the Nemo device from Keysight Technologies. Key uplink related parameters were extracted, including transmit (Tx) power, frequency band, application type, and reference signal received power (RSRP). The proposed deep learning framework utilizes uniquely environmental features, publicly available datasets, i.e., map imagery and base station information, and actual measurements rather than simulations as model inputs, making it a purely measurement based approach. The proposed framework can predict both RSRP and Tx power, where extra information on the application and band are fed to the part predicting Tx power. Measurement locations were divided into separate sets for training and testing to ensure independent validation. The results demonstrate that the proposed framework achieves a prediction performance of RMSE =7.84 dB for downlink and RMSE=7.5 dB for uplink transmit power estimation.