Deep Learning based Jammer Localization in Mobile Communication Networks

We investigate the localization of a radio jammer disrupting a wideband OFDM-based communication in the uplink of a mobile network. This is done by separating the estimated angles-of-arrival (AoA) into groups associated with the user equipment (UE) and the jammer, respectively. The jammer AoAs are utilized to locate a jammer using a data-driven machine-learning (ML) technique. Low-complexity methods are employed for AoA separation. Two types of jammers are examined, linear-chirp jammers and OFDM-based random jammers. We demonstrate the effectiveness of ML approach for single-cell based jammer localization (under suitable conditions). We show that the ML models can be trained solely on the UE dataset (independent of the jammer) that can be procured in a cooperative setup. Simulation results demonstrate the promise of the ML-based approach in offering superior performance compared to the classical approach.