5:15 PM - 7:15 PM
[PEM13-P22] Development of software for removing sferics and broadcast radio wave noises on magnetospheric ELF/VLF wave analysis
Keywords:VLF, magnetosphere, noise removal, spectrogram
The analysis of magnetospheric ELF/VLF waves provides valuable information for understanding the dynamic variation of Earth's magnetosphere and ionosphere. However, these signals can become difficult to distinguish in spectrograms due to interference from sferics noise generated by lightning and broadcast radio wave noise in the VLF band.
In this study, we developed software to remove noise from sferics and broadcast radio waves while preserving the spectral form of magnetospheric ELF/VLF waves as much as possible. The data used in this study were ELF/VLF waveforms captured at a 40 kHz sampling rate at Athabasca in Canada and Oulujarvi and Angeli in Finland, as part of the PWING project.
Sferics noise appears on the high-frequency end of the frequency-time dynamic spectra, above 4-5 kHz, spanning over a wide frequency range. The sferics removal process consists of two steps, i.e. noise detection and interpolation. First, we used wavelet transform to extract only the high-frequency components above 5 kHz from the waveform data. Then we identified wave components exceeding a certain amplitude threshold as sferics noise. Next, we performed a Fourier transform on the waveform before and after the detected noise region to obtain their spectra, averaged them, and applied an inverse Fourier transform to generate a waveform used to interpolate the noise-affected region, effectively removing the noise. Through this process, we found that increasing the number of detected data points effectively eliminated sferics noise; however, it also affected the waveform of magnetospheric ELF/VLF waves.
In the spectra, broadcast radio wave noise appears as intermittent horizontal lines at fixed frequencies. To detect this noise, we used the average and/or median value of the power spectral density over 590 seconds. The interval corresponds to the size of a single data file. We then compared three adjacent frequency points before and after each data point. If a data point exhibited power at least ten times higher than its neighboring frequencies, it was identified as broadcast radio wave noise and replaced with the average of the surrounding values. A comparison of both methods revealed that using the average value resulted in better noise removal performance.
In this study, we developed software to remove noise from sferics and broadcast radio waves while preserving the spectral form of magnetospheric ELF/VLF waves as much as possible. The data used in this study were ELF/VLF waveforms captured at a 40 kHz sampling rate at Athabasca in Canada and Oulujarvi and Angeli in Finland, as part of the PWING project.
Sferics noise appears on the high-frequency end of the frequency-time dynamic spectra, above 4-5 kHz, spanning over a wide frequency range. The sferics removal process consists of two steps, i.e. noise detection and interpolation. First, we used wavelet transform to extract only the high-frequency components above 5 kHz from the waveform data. Then we identified wave components exceeding a certain amplitude threshold as sferics noise. Next, we performed a Fourier transform on the waveform before and after the detected noise region to obtain their spectra, averaged them, and applied an inverse Fourier transform to generate a waveform used to interpolate the noise-affected region, effectively removing the noise. Through this process, we found that increasing the number of detected data points effectively eliminated sferics noise; however, it also affected the waveform of magnetospheric ELF/VLF waves.
In the spectra, broadcast radio wave noise appears as intermittent horizontal lines at fixed frequencies. To detect this noise, we used the average and/or median value of the power spectral density over 590 seconds. The interval corresponds to the size of a single data file. We then compared three adjacent frequency points before and after each data point. If a data point exhibited power at least ten times higher than its neighboring frequencies, it was identified as broadcast radio wave noise and replaced with the average of the surrounding values. A comparison of both methods revealed that using the average value resulted in better noise removal performance.