PSN-L Email List Message
Subject: Re: Sensor noise
From: Larry Conklin lconklin@............
Date: Fri, 01 Jan 2010 10:30:57 -0500
Your goals were a lot more ambitious than mine. I had no thoughts about
trying to use the frequency domain to trigger event alerts. I've pretty
much concluded that any sort of automatic alert scheme requires a MUCH
quieter site than I am able to provide in my residential location.
I have a waterfall display running, but it clearly isn't performing the
FFT correctly. For one thing it looks like the plot is symetrical
around the center point (showing "negative" frequemcies"?). And when I
run it on a data file that contains just a single sinusoidal signal, I
get an output that shows a whole series of lines. So, I've got some
debugging to do, if I ever get around to it.
I found the code (in C) for the transform somewhere and cobbled it into
my data logger/replay program. Kind of a hack job, so I probably
deserve the problems I'm having.
On 12/29/2009 10:17 PM, Barry Lotz wrote:
> I wrote the Quick Basic acquisition program back when storage size was
> an issue :) Initially , before I changed the program to an FFT analysis,
> because of the computer speed I ran what was called a Walsh transform.
> It was similar to an FFT but you used square waves. The math was faster
> and I could write it in assembly language. What I thought was that a
> quake signal is seen as a change in amplitude and the frequency
> composition of the background signal. The "short term-long term" trigger
> and the like routines look at signal amplitude only and not changes in
> frequency. What I did was to run an FFT on a sort of small size window (
> one I could run between the acquisition of two data points) maybe around
> 512 points. At a fairly slow sample rate of about 5 hz ( for teleseismic
> events) it covered a reasonable frequency range and I could do it with
> my "slow" 286 computer. I chose certain frequencies to observe (based on
> FFTs I had run on typical previously recorded events ) and ran the short
> term/ long term style routine on a weighted sum of these frequency
> amplitudes. The thought was that the trigger would pick up the P wave
> arrival. I then started recording beginning with a moving buffer of
> certain # of data points. It would ignore most transients. The "noise" I
> had the most trouble with was wind ( not on the sensor), I could protect
> from that. It was fluctuating wind gusts that generated movement in the
> slab the sensor was on. My trigger threshold varied in amplitude based
> on running data. It could correct for most of this wind also. The
> interesting thing was I could "catch " confirmed events that I could
> barely see with my eye "in the grass". It all became academic when it
> was easier to record continuously and look for events based on what
> showed up on say the USGS web site. Events you can't see very well
> aren't that interesting to look at. ---- You asked :} I don't remember
> right now the numerical specifics, but I can look it up if you are
> --- On *Tue, 12/29/09, Larry Conklin //* wrote:
> From: Larry Conklin
> Subject: Re: Sensor noise
> To: psn-l@..............
> Date: Tuesday, December 29, 2009, 1:16 PM
> Hi Barry,
> It sounds like you've been messing around with something similar to
> a project I started but didn't get around to finishing. I added a
> waterfall FFT plot to my data logging program, but never finished
> debugging it. In part because I wasn't all that confident that the
> plot would be worth all that much.
> I'd be interested in a little more information regarding how you are
> collecting the FFT and some of the design decisions you use (number
> of samples in the window, how frequently you do a conversion, etc.)
> Did you write your own code or are you using some 3'rd party software?
> Larry Conklin
> Liverpool, NY
> On 12/29/2009 12:12 AM, Barry Lotz wrote:
> > All
> > I have successfully used a running fft to sense signal frequency
> > component changes and used this as a trigger mechanism for an event.
> > Could one use the ( I guess you call it ) power spectrum of the
> > just before and during the event to remove the noise? I guess you
> > have to use the same time window so the frequencies would
> compare. Could
> > it be a simple subtraction of the "before" from the "during"?
> This would
> > assume that the background noise didn't change in the period
> during an
> > event. This could be better than trying to shape a multi pole
> filter to
> > eliminate the noise. I have found that often a portion of the event
> > signal is in the same frequency range as the noise.
> > Regards
> > Barry
> > http://www.seismicvault.com
> Public Seismic Network Mailing List (PSN-L)
> To leave this list email PSN-L-REQUEST@..............
> with the body of the
> message (first line only): unsubscribe
> See http://www.seismicnet.com/maillist.html for more information.
Public Seismic Network Mailing List (PSN-L)
[ Top ]
[ Back ]
[ Home Page ]