Proposal of A Novel EMI Diagnostic System using Independent Component Analysis

This paper presents a noise diagnosing system that can separate multiple interfering signals due to crosstalk or common-mode noise problems. In estimating original time-series signals from measured data sets, our system uses the independent component analysis (ICA) scheme, which uses a statistical technique to separate blind source signals from measured data sets. In addition, mutual-correlation coefficients between separated and measured signals are used to obtain a set of measured and estimated signals. Evaluations of transmission lines confirmed that our system is effective for diagnosing multiple electromagnetic interference (EMI) problems, resulting in the mutual-correlation coefficients between the original and estimated signals being more than 0.9 for 1-20 MHz sinusoidal signals.