Implementation Notes

Signal to Noise Ratio

In computing the signal-to-noise ratio, we compute the FFT of the waveform for the event and pre-event (noise) windows. For the event window, we set spectral values below the lowest resolvable frequency (1/event_duration) to NaN. We do not do this for the pre-event (noise) window, because we compute the Fourier amplitude spectrum of the signal by subtracting the pre-event spectrum from the event spectrum. If we did set the values to NaN, then all signal spectral values below the lowest resolvable frequency in the pre-event (noise) window would be NaN. In most cases where this could be an issue, the signal spectra is well above the noise spectrum in this frequency band and extrapolation of the pre-event spectrum is reasonable, so it is reasonable to use the signal-to-noise ratio in this frequency band.

Note

If the high-pass corner frequency determined from the signal-to-noise ratio is less than the minimum frequency in the pre-event (noise) window, gmprocess will include a warning message in the logging output.

Tip

We recommend using the rider_fchp processing step (included by default) to adjust the high-pass corner frequency. This helps minimize extraneous long-period oscillations in the waveforms.