Towards Efficient Reflectarray Digital Twins - An EM-Driven Machine Learning Perspective

The concept of digital twins (DTs) for reflectarray (RA) unit cells (UCs) is discussed and implemented by exploiting electromagnetic-driven machine learning (ML) techniques. Toward this end, several open challenges are addressed, such as the reliability and the effectiveness of using surrogates for modeling different, in terms of descriptors and complexity, UCs, the accuracy in predicting the scattering matrix entries of both single- and dual-polarization elements, the implementation of effective and efficient strategies for the setup of the training set, and the definition of generalized and robust guidelines for the use of popular ML techniques to the problem at hand. Representative results of an extensive numerical validation are presented to assess the performance and the potentialities of DTs when dealing with different RA modeling problems and training sets.