Abstract
Electromagnetic radiation source imaging and localization are crucial to the diagnostics of electromagnetic interference (EMI) problems in complex electronic devices and systems, but are usually a time-consuming and tedious process. This article presents a novel and automatic method based on active and unsupervised machine learning for the rapid localization of EMI sources using the sparse emission source microscopy (ESM) technique. The query-by-committee (QBC) method is adopted to select sparse scan samples with high uncertainties on a scanning plane. A real-time source image is reconstructed on the device-under-test (DUT) plane using the ESM method. Also, a K-means clustering (KMC) method is used to automatically identify the number and locations of EMI sources, and automatically determine when the scanning process can be terminated. The results of simulation and measurement examples show that the proposed method is much more time-efficient compared with the conventional full-scanning ESM method, and can automatically identify the number and locations of the dominant EMI sources without human intervention. Also, the method demonstrates a noticeable advantage over two commonly used sparse scanning methods, including greedy sampling and random sampling, with more clear reconstructed EMI source images with the same number of sparse scanning samples.
Funding Information
  • National Natural Science Foundation of China (62571480)
  • National Natural Science Foundation of China (62071424)
  • National Natural Science Foundation of China (62027805)
  • National Natural Science Foundation of China (LQ23F010020)
  • National Natural Science Foundation of China (LD21F010002)

This publication has 34 references indexed in Scilit: