Application of convolutional neural networks for improving the accuracy of multistatic localization of radio emission sources

Abstract

The study addresses the scientific and practical problem of improving the accuracy of localization of radio emission sources (RES) in the high-frequency (HF) band by combining classical direction-finding and amplitude analysis methods with modern intelligent technologies. The relevance of the problem is determined by the fact that traditional methods for determining the coordinates of signals largely depend on ionospheric disturbances, geomagnetic activity, and intentional electronic countermeasures, which significantly reduce their effectiveness.
The aim of the study is to develop a multistatic architecture of software-defined receivers with embedded convolutional neural network (CNN) modules capable of analyzing spectral–spatial signal characteristics and identifying hidden patterns in input data. The paper describes the principles of constructing algorithms for joint processing of amplitude and angular information, taking into account the electrophysical parameters of the ionosphere and factors of electromagnetic inaccessibility. Simulation results confirmed that the application of a hybrid approach makes it possible to reduce the uncertainty area of localization to 8–30% of its initial size.
Special attention is given to the analysis of localization errors, their physical causes, and minimization methods, which ensure stability and reliability of the system even under conditions of electromagnetic countermeasures. The proposed approach enhances the efficiency of radiomonitoring and electronic intelligence systems.