On a Machine Learning Enhanced Solution of the Inverse Source Problem

The inverse source problem in electromagnetics is present in a wide range of applications, such as antenna characterization and antenna diagnostics. This problem can be tackled by means of the boundary element method, by leveraging the surface equivalence principle. In this context, the system to be solved is plagued by two sources of ill-posedness. The first is related to the presence of evanescent modes, which exponentially decay as the distance between the object under test and the probes increases. Therefore, part of the physical information quickly falls below the noise level, especially in settings in which the field sampling is performed in the far-field region. The second is the null-space due to the existence of non-radiating currents, which are described by the solutions of the homogeneous integral equation associated to the system.