Ambiguity-reduced phaseless reconstruction of planar antennas using deep learning
Array coefficient reconstruction is a fundamental problem in antenna engineering. Although full-wave measurements enable accurate recovery of source distributions, they are often costly and experimentally complex. As an alternative, phaseless measurements provide a more practical approach; however, they lead to a nonlinear and ill-posed inverse problem due to the loss of phase information, resulting in ambiguity in reconstruction. To address this issue, a preprocessing strategy is applied to the source current distribution prior to radiation. Specifically, an additional phase-dependent term is incorporated into the original current distribution, producing a modified current source that generates more informative phaseless measurements. Based on these data, a dual-branch deep learning framework composed of two U-Net networks is developed to reconstruct the complex source distribution of two-dimensional antenna arrays. The two branches are responsible for predicting the real and imaginary components, respectively. The proposed method enables accurate recovery of both amplitude and phase information using only intensity measurements. Numerical experiments show that the model achieves a coefficient of determination (R 2 ) of approximately 0.98, whereas the conventional approach achieves an R 2 of approximately 0.40. The results demonstrate that the proposed preprocessing strategy significantly reduces ambiguity in phaseless inverse problems and substantially improves reconstruction accuracy. The method effectively resolves the non-uniqueness issue inherent in intensity-only measurements and outperforms conventional reconstruction approaches.