Hi,
I need to correct some waveforms from microtremor.
I am using the remove_response code outputting accelerations When playing with the water level parameter I observed a significant senstivity on this parameter. By setting it to the default 60 dB or setting it to none, the difference is huge and the results with the setting to none are definetely not correct and out of scale. Is there some guidance on how best to tailor this paramter to obtain reasonable results?
Thank you
You want to identify what part of your spectrum is actual data and what is just instrumentation noise. You want to set water level and/or pre_filt to suppress that part that is not actual data.
In the bottom plot on the bottom left you can see in red the inverted spectrum that is multiplied on your data in frequency domain. Note the logarithmic scale, that part at the left end of your spectrum at 10e-4 Hz is definitely not real signals and it gets amplified by 10e5. Also that high frequent part around 100 Hz seems to get amplified by 10e6.
thanks a lot for looking into this issues and for the clarification. So, if I set the water level to 85 dB it should be the right value to choose in order not to have unrealistic overamplification on both the low and high frequencies . We are mainly interested at the high frequencies, below 0.001 Hz it is presumably just noise? Would you agree on the selection of the 85 dB capping in this case ?
Personally, these days I usually do pre_filt instead, which basically sets a pass window in the frequency domain aka. applies a taper to the spectrum in frequency domain. If you plan to filter the output later anyway you can just set these frequencies a bit outside of your final bandpass window
Just from the plots you shared the last one seems to be OK ish, though. But yeah, can’t tell with 100% certainty just from looking at these plots you shared. You always need to look at the data in detail, and I’d just vary the stabilization parameters a bit and see how it affects your output. You’ll get a feeling for when unwanted parts of the spectrum blow up in a bad way and when you should be more lax to not suppress your actual data spectrum
