INTELLIGENT METHODS FOR MULTI-POSITION LOCALIZATION OF RADIO SIGNAL SOURCES USING NEURAL NETWORKS AND ADAPTIVE FILTERS

This paper presents a hybrid localization system for shortwave radio signal sources based on ResNet CNN with an attention mechanism and an adaptive Extended Kalman Filter. Inputs include amplitude, interaural phase difference (IPD), time-difference-of-arrival (TDoA), and direction-of-arrival (DoA). Our model achieves a localization error of about 4 km—significantly outperforming traditional amplitude-based (28.5 km), direction-finding (21.3 km), and basic hybrid (12.8 km) methods. The paper includes a comparative analysis, key formulas, tables, and illustrative schematics.