Phase Retrieval Based on Supervised Machine Learning Regression Algorithms

This work introduces a novel phase-retrieval technique applicable to different families of sources. The presented approach uses supervised machine learning algorithms to estimate the phase of electromagnetic fields. The methodology involves training the algorithms with a dataset from spherical near-field data taken at multiple sampling locations. The analyses results evidence the efficiency of standard machine learning regression algorithms in efficiently recovering the phase of electromagnetic waves emanating from different sources. Both statistical and numeric validations have been conducted to validate this approach.