PSN-L Email List Message

Subject: Re: event detection
From: barry lotz gbl@.......
Date: Sun, 06 Aug 2000 08:00:26 -0700

Sean & Larry
    I have , with reasonable success been using a modified version of  the
routine by Tom Goforth & Eugene Herrin "Automatic Seismic Signal Detection
Algorithm Based on the Walsh Transform" in BSSA vol. 71, no. 4, August 1981.
I originally ran it with the Walsh Transform then changed to the FFT. It
basically runs an overlapping frequency analysis then allows one to modify
certain frequencies and then sums to spectrum. It compares this sum with a
threshold value which is a function of the median sum. It also keeps a
leapfrog set of three buffers of raw data in case the is a trigger. I like
this routine because it considers frequency content not just amplitude. It is
also can adjust for background noise variations. If one knows a particular
frequency which is troublesome one can weight the FFT output before summing
the spectrum. I have modified this somewhat by keeping a running average FFT
and comparing the new FFT with the running average. Once a trigger has
occured I  keep recording data til the FFT falls below a certain level and
therefore can account for short and long events.The routine works well on a
386. However I only run it at 4 hz and use it with the smt8 style sensor for
teleseismic events. The only problem I have had to date is with wind. It
seems to be of the right frequency and amplitude for the computer to think it
is an event though the eye can tell the difference. Wind is not as symetrical
as an event. Now I need to figure a way to represent this. Counting crosses
alone doesn't seem to be totally effective though I do use it to prevent
triggers due to major offsets during recording.


Public Seismic Network Mailing List (PSN-L)

[ Top ] [ Back ] [ Home Page ]

Larry Cochrane <cochrane@..............>