MonitorIA - Radiomonitoragem do Serviço Móvel Pessoal por Crowdsensing

Radiomonitoring involves mapping frequency usage, enabling, for example, the deactivation of clandestine stations. Crowdsensing reduces costs and increases agility compared drive tests, but current proposals demand high resource usage from mobile phones. This paper proposes MonitorIA: a framework that uses smartphones to map channels and identify irregularities through machine learning. The contributions include the evaluation of spectrum mapping models on real data and the use of comparative metrics between maps to identify irregularities. The experiments use data from different applications in models trained with ray tracing, aiming to evaluate the generalization capacity. The results indicate a power prediction error up to 30% lower than that obtained with statistical methods, using less than 4% of the measurements. Furthermore, the proposed Siamese network identifies irregular transmissions at different positions, powers, and frequencies with an accuracy above 92%.