Indoor 1D-Localization of Omnidirectional Power-Modulated Jammers: A Machine Learning Approach With Rapid Database Generation

The threat of power-modulated jammers to electronic systems has recently been reported in the literature. These are malicious devices that emit intentional electromagnetic interference whose power changes rapidly over time. Such dynamic power emissions make it hard for traditional localization algorithms to track the jammer position in indoor environments, especially if the shadowing effects of the objects and people nearby the monitoring antennas are strong. In this work, we propose an indoor jammer localization strategy based on machine learning. The machine learning models are built from simulations based on the shooting-and-bouncing rays technique to quickly generate the required databases and provide a parametric study. The simulation model is validated by comparison with the measurement performed in a real room in the presence of a commercial jammer. Decision tree algorithms lead to predictions with an accuracy of tens of centimeters for a constant-power jammer and a power-modulated jammer. This result significantly outperforms conventional trilateration approaches. Furthermore, a new machine learning feature based on power ratios was introduced and provided good predictions even if the jamming power is unknown by the machine learning model. In addition, the main limitations are evaluated according to the uncertainties between measurement and simulations, the learning dataset size and changes in the considered environment. Finally, the proposed framework is validated using measurement as input of the machine learning models.