Obspy and Instrument Response

[sheeny]

Starting to play with Python and Obspy and experimenting with removing the instrument response on my RS&B (R21C0). I’m getting a long period oscillation in the waveform after removing the instrument response. I’d appreciate any clues as to why or how to fix it.

Here’s my code:

Here’s the raw count plot:

and this is the waveform after removing the instrument response:

Being a geophone rather than an accelerometer, I would expect the waveform in velocity with the response removed would be virtually the same as the raw counts, as it appears in DataView, but clearly this isn’t the same shape.

Thanks,
Al.

[PhilipPeake]

I can’t speak to Obspy, but that does look familiar from my experiments writing code to detect events.
The raw data looks like this:

{'EHZ', 1654182923.653, 16636, 16720, 16767, 16711, 16700, 16651, 16651, 16661, 16646, 16606, 16627, 16679, 16662, 16657, 16629, 16637, 16634, 16646, 16672, 16674, 16694, 16698, 16738, 16772, 16774}

{'EHZ', 1654182923.903, 16775, 16766, 16741, 16751, 16734, 16718, 16684, 16640, 16639, 16699, 16754, 16682, 16601, 16543, 16611, 16663, 16632, 16657, 16642, 16636, 16669, 16653, 16639, 16697, 16709}

{'EHZ', 1654182924.153, 16691, 16670, 16680, 16669, 16651, 16707, 16684, 16696, 16668, 16664, 16663, 16658, 16698, 16664, 16719, 16702, 16686, 16620, 16517, 16545, 16575, 16623, 16638, 16596, 16627}

{'EHZ', 1654182924.403, 16682, 16644, 16659, 16662, 16698, 16707, 16698, 16782, 16727, 16716, 16671, 16614, 16651, 16633, 16674, 16692, 16749, 16745, 16669, 16656, 16610, 16579, 16623, 16689, 16686}

{'EHZ', 1654182924.653, 16581, 16624, 16671, 16534, 16565, 16598, 16613, 16658, 16680, 16702, 16685, 16670, 16643, 16658, 16695, 16758, 16734, 16735, 16733, 16675, 16712, 16746, 16779, 16695, 16658}

{'EHZ', 1654182924.903, 16661, 16636, 16640, 16614, 16676, 16641, 16609, 16613, 16598, 16613, 16650, 16684, 16676, 16674, 16630, 16628, 16638, 16613, 16553, 16594, 16688, 16690, 16668, 16643, 16623}

{'EHZ', 1654182925.153, 16608, 16674, 16710, 16677, 16684, 16721, 16726, 16697, 16681, 16721, 16756, 16743, 16758, 16699, 16683, 16695, 16652, 16673, 16655, 16686, 16703, 16650, 16648, 16606, 16597}

{'EHZ', 1654182925.403, 16657, 16688, 16694, 16652, 16579, 16587, 16663, 16755, 16794, 16698, 16611, 16596, 16572, 16679, 16750, 16663, 16620, 16575, 16562, 16599, 16609, 16583, 16624, 16678, 16683}

{'EHZ', 1654182925.653, 16679, 16617, 16636, 16662, 16649, 16683, 16647, 16577, 16596, 16657, 16703, 16727, 16730, 16718, 16684, 16648, 16649, 16710, 16705, 16657, 16664, 16625, 16602, 16622, 16677}

{'EHZ', 1654182925.903, 16721, 16707, 16639, 16639, 16676, 16641, 16629, 16633, 16654, 16596, 16613, 16622, 16565, 16582, 16611, 16677, 16722, 16751, 16692, 16682, 16683, 16688, 16745, 16690, 16674}

You can see that it is centered around a value of ~16660. But not exactly, and over time the “center” value changes.

What I do, and what I think most others do, is to calculate a “running average” over a long period, or, if you wish to think of it that way, a (very) low-pass filter. That gives you a stable average and you plot incoming data as a delta from this (continuously updated) average.

If you don’t use enough samples in calculating the average, you get some residual “ripple” - which looks like what you are seeing.

I average the baseline over 250 data points (10 seconds).

Here is a pair of graphs, raw data on the left, and the average calculated over 100 values on the right:

Same thing, but with the average calculated over 250 data points:

[sheeny]

OK, so you are not using Obspy to remove the instrument response - you have written your own code, correct?

Al.

[PhilipPeake]

Right - I just use the USB data stream which is about as raw as it gets - no corrections at all applied.
The example I pasted is a listing of the data packets - one per second, which you can probably see is:

{STREAM, TIME, val, val, val … x25}

25 samples per second.

[sheeny]

OK, here’s what I’ve found that seems to work:

This is the raw counts:

This is the demeaned counts:

and this is the velocity signal in m/s:

This is using instrument response data from the metadata for my shake, and NOT the Obspy function to remove the response.

Al.

[PhilipPeake]

Looks like the code that applied the instrument response must also have done the demean-ing.
Presumably this is documented somewhere? Or do they just expect you to read the code?

Whatever … looks like you have doing what you want now.

[branden]

Hello @sheeny

We have some example ObsPy code for removing the response here.

branden

[sheeny]

Thanks Branden, that’s basically what I was doing, but having a closer look found the link to here.

The pre filter was the bit I needed.

Thanks!
Here’s the code that works:

and the final result without the wavy low frequencies:

Thanks again.

Al.