Single-Channel Blind Source Separation for Periodic Electromagnetic Switching Noise

In this article, a single-channel blind source separation method is proposed to decompose time-domain measurement results of electromagnetic noise into the underlying periodic switching noise source signals without requiring any information about the sources. This is realized by clustering the time-domain waveforms based on their period and similarity. At first, the start and end times of the waveforms need to be defined. Then, periods are determined and a waveform cluster is assigned to each period using a new probabilistic approach. Only in the final step, the waveform shape is considered, when the clusters are filtered by removing the waveform outliers that are too different from the remaining waveforms of each cluster. To this purpose, a distance definition based on the cross correlation among the waveforms is proposed. Using numerical and measurement examples, it is shown that it is possible to determine and rank the contribution to the spectrum of each source separately. This helps individuating the main noise sources more quickly and to analyze the contribution of single sources in a complex noisy environment.

Automated summary

This article proposes a single-channel blind source separation method for decomposing time-domain measurement results of electromagnetic noise into underlying periodic switching noise source signals without any prior knowledge of the sources. The method clusters the time-domain waveforms based on their period and similarity, assigns waveform clusters to each period using a new probabilistic approach, and filters the clusters by removing waveform outliers that are significantly different from the remaining waveforms. The proposed method allows for the determination and ranking of each source's contribution to the spectrum, aiding in the identification of main noise sources and analysis of individual source contributions in complex noisy environments.