Spectrum separation for electromagnetic noise using eigenvalue decomposition of correlation matrix

It is required to reduce undesired emissions from electric equipment. Spectrum measurement is the first step to reduce the undesired emissions. We propose a method to separate spectra based on the statistical independence of sources. Waveforms received are sampled by a digitizing oscilloscope. After applying FFT to the obtained waveforms, a correlation matrix is calculated. Finally, by applying eigenvalue decomposition to the correlation matrix, separated spectra are obtained. The independence of sources is used as a priori knowledge to separate the spectra. The validity of the proposed method was demonstrated experimentally.