Active-Learning-Based Sparse Near-Field Scanning With Time-Domain Current Measurement for Conductive Coupling Path Visualization

This article presents a novel and efficient near-field scanning method for coupling path visualization of conductive electromagnetic interference (EMI) using sparse scanning samples based on active machine learning. A high-voltage discharge system with a false triggering problem is studied as the device under test (DUT). Near-field scanning is adopted to measure the time-domain waveform of the conductive current and investigate the coupling path of the conductive EMI that causes the false triggering issue. To facilitate the time-consuming near-field scanning process, an adaptive sampling approach based on active learning is proposed to seek the most informative locations and minimize the number of scanning samples. Through the developed active learning method, the scanning process is more efficient and intelligent than random sampling and a uniform-like sampling strategy called greedy sampling (GSx). Moreover, the dominant EMI coupling path in a key area is clearly visualized. By cutting off the identified coupling path, the false triggering phenomenon is effectively eliminated. The proposed active learning method is much more efficient than the conventional full scanning and can be applied to expedite various near-field scanning scenarios.