Machine Learning Based Data Analysis for Electromagnetic Reverberation Chambers*

In addition to standardized methods for hardness, emissivity, or shield damping tests, electromagnetic reverberation chambers (ERCs) offer the opportunity of microwave spectroscopic scans to quickly find critical frequencies which can subsequently be further examined by another method. In this work, different data analysis techniques for power spectroscopy in ERCs are presented. As a baseline, an established way of fitting a physically based model via non-linear regression to power absorption data is recalled. Next, neural network based classification and the use of neural network based auto encoders are presented as new approaches to analyze ERC-data. For these approaches, a data augmentation technique is introduced to generate datasets from the frequency-dependent power absorption of devices under test, which resembles a bootstrapping method.