Deep Learning-Assisted Phaseless Near-Field Transformation for Accelerating Near-Field Scanning

The electromagnetic field will spread/shrink between a lower/higher plane and a higher/lower plane. It is difficult to simulate such spreading/shrinking by using traditional methods, such as dipole source reconstruction method. In this paper, a deep convolutional neural network (DCNN) is proposed to "learn" the rule of such spreading/shrinking of electromagnetic fields, and then a novel near-field to near-field transform is proposed. To learn such rule, the proposed DCNN combines a height feature matrix with the scanned near-field pattern. The input of the DCNN is the scanned phaseless near-field at single plane and the height label of the target plane. The output is the field pattern at the target plane. Based on the scanned phaseless near-field at a single plane, the fields at other planes can be quickly derived using the trained DCNN. It significantly enhances the scanning efficiency in electromagnetic interference (EMI) near-field measurement. This method is comprehensively validated through numerical and experimental examples. Moreover, the proposed method shows better precision compared with the traditional dipole source reconstruction method. The corresponding code, data, and well-trained DCNN are all available through https://github.com/DongHaoHan/EMC-EUROPE-Near-Field-Transformation.