Ted (tchannel) wrote:
Hi Folks,  One issue where I seem to need  help, is how to filter.  I have 
kept notes on the Low and High Pass filter  values used by others. I also know 
that the filter values will change with all  the variables. Let us assume, 
using a vertical similar to the AS1, what would be  good starting points for 
Local, Regional, and Teleseimetic events?    Also could someone explain what the 
poles numbers do? 
 
Hi Ted,
 
  The object of seismic filtering is to reproduce the signal with as  little 
distortion as possible and attenuate the noise. In the case of seismic  
signals, the spectrum containing the signal is dependent on event distance, and  the 
noise is primarily from microseisms and cultural noise, including traffic,  
other human activity, and wind. Microseisms range widely in intensity and  
frequency. Their period ranges primarily from 2 seconds to 7 seconds. Cultural  
noise, in my experience, is generally short period and can be easily rejected  
except for local events, which also have a short period. Wind, by exerting  
pressure on trees and structures, imparts long period motion to the ground, and  
can for the most part be excluded by long period high pass filtering. The same 
 applies to tilt induced by walking anywhere near a horizontal sensor. 
Usually,  the sensor itself is the high pass filter. Amplifier bias drift and 1/f 
noise  are also long period effects. Atmospheric pressure and temperature 
changes cause  the pendulum to drift.
 
  For intermediate distance and long distance events, microseisms are  right 
in the frequency range of interest, and you probably should just suffer  them.
 
  My advice on filtering is to let experience be your guide, but use  the 
best filters you can, and do as little filtering as you can get away with.  In my 
opinion, the lowpass and highpass filters in WinQuake stink, and I have  
developed my own filters. They are implemented in my WQFilter.exe utility  
program, and anyone can download and try them out on WinQuake files. The  approach I 
use is to broadbrand the signal first with my period extending  filter 
(especially necessary with the AS1), and then restrict the high and low  ends of the 
frequency range with Butterworth filters. You should restrict the  number of 
poles used. The number of poles determines the steepness of the  rolloff at the 
corner frequency. However, more poles means more ringing in the  filter 
impulse response, which you do not want. 
 
You can learn a lot about filters by downloading "seismic_dataq.zip" from  
John Lahr's web site. Extract "WQFilter.exe", "WQFilterHlp.txt", and  
"Impulse.psn" from the zip. Use both WinQuake and WQFilter to filter  "Impulse.psn" with 
the various filters of each program. Using WinQuake, examine  the resulting 
filtered response (the impulse response). The FFT of the impulse  response will 
give you the passband spectrum of the filter. The impulse response  itself 
will tell you about the group time delay and ringing response of the  filter. My 
filters have zero time delay and phase distortion. This is achieved  by 
filtering the data forward in time, and then backward in time (impossible to  do 
with an analog filter). 
 
  This is the Digital Age. Analog is dead! Do not listen to anyone  about 
analog filter circuits. Analog filtering has no place anymore except at  the 
front end of the A/D converter, and even that need is minimized by sampling  at a 
high rate and subsequently downsampling with maybe some digital filtering  
thrown in.
 
Bob





Ted (tchannel) wrote:
Hi Folks,  One issue where I seem to need=
=20
help, is how to filter.  I have kept notes on the Low and High Pass fil=
ter=20
values used by others. I also know that the filter values will change with a=
ll=20
the variables. Let us assume, using a vertical similar to the AS1, what woul=
d be=20
good starting points for Local, Regional, and Teleseimetic events? &nbs=
p;=20
Also could someone explain what the poles numbers do? 
 
Hi Ted,
 
  The object of seismic filtering is to reproduce the signal with=20=
as=20
little distortion as possible and attenuate the noise. In the case of seismi=
c=20
signals, the spectrum containing the signal is dependent on event distance,=20=
and=20
the noise is primarily from microseisms and cultural noise, including traffi=
c,=20
other human activity, and wind. Microseisms range widely in intensity and=20
frequency. Their period ranges primarily from 2 seconds to 7 seconds. Cultur=
al=20
noise, in my experience, is generally short period and can be easily rejecte=
d=20
except for local events, which also have a short period. Wind, by exerting=20
pressure on trees and structures, imparts long period motion to the ground,=20=
and=20
can for the most part be excluded by long period high pass filtering. The sa=
me=20
applies to tilt induced by walking anywhere near a horizontal sensor. Usuall=
y,=20
the sensor itself is the high pass filter. Amplifier bias drift and 1/f nois=
e=20
are also long period effects. Atmospheric pressure and temperature changes c=
ause=20
the pendulum to drift.
 
  For intermediate distance and long distance events, microseisms=20=
are=20
right in the frequency range of interest, and you probably should just suffe=
r=20
them.
 
  My advice on filtering is to let experience be your guide, but u=
se=20
the best filters you can, and do as little filtering as you can get away wit=
h.=20
In my opinion, the lowpass and highpass filters in WinQuake stink, and I hav=
e=20
developed my own filters. They are implemented in my WQFilter.exe utility=20
program, and anyone can download and try them out on WinQuake files. The=20
approach I use is to broadbrand the signal first with my period extendi=
ng=20
filter (especially necessary with the AS1), and then restrict the high and l=
ow=20
ends of the frequency range with Butterworth filters. You should restrict th=
e=20
number of poles used. The number of poles determines the steepness of the=20
rolloff at the corner frequency. However, more poles means more ringing in t=
he=20
filter impulse response, which you do not want. 
 
You can learn a lot about filters by downloading "seismic_dataq.zip" fr=
om=20
John Lahr's web site. Extract "WQFilter.exe", "WQFilterHlp.txt", and=20
"Impulse.psn" from the zip. Use both WinQuake and WQFilter to filter=20
"Impulse.psn" with the various filters of each program. Using WinQuake, exam=
ine=20
the resulting filtered response (the impulse response). The FFT of the impul=
se=20
response will give you the passband spectrum of the filter. The impulse resp=
onse=20
itself will tell you about the group time delay and ringing response of the=20
filter. My filters have zero time delay and phase distortion. This is achiev=
ed=20
by filtering the data forward in time, and then backward in time (impossible=
 to=20
do with an analog filter). 
 
  This is the Digital Age. Analog is dead! Do not listen to anyone=
=20
about analog filter circuits. Analog filtering has no place anymore except a=
t=20
the front end of the A/D converter, and even that need is minimized by sampl=
ing=20
at a high rate and subsequently downsampling with maybe some digital filteri=
ng=20
thrown in.
 
Bob