PSN-L Email List Message

Subject: Re: Noise Reduction
From: ChrisAtUpw@.......
Date: Fri, 30 Jun 2000 21:08:15 EDT

Dick Webb,

    The Infrasound CTBT Stations use three sensors in a more or less 
equilateral triangle with a fourth in the centre. They look for cross 
correlation between the four signals assuming that the sound source comes 
from one direction, scanning 360 Deg and at a range of propagation 
velocities, by storing and then addressing signals taken at different times. 
Whether you would want to a) fund the hardware which includes a serious 
amount of computer power and b) could get hold of / write suitable software, 
is your decision, but the system does apparently work.

    The root N reduction in noise only applies to random noise. If the 
'noise' isn't random, or is common to two or more sensors although with 
different phases, it may or may not add, just like signal. I have a 12 bit 
A/D converter, but it is noisy, so I can only rely on it to 10 bit accuracy. 
If the input signal stays the same, averaging 4 separate conversions will 
give me ~11 bit accuracy. To get ~12 bit accuracy I have to average 16 
conversions (minimum). To be absolutely certain of 12 bit accuracy, I have to 
average 64 conversions..... I wrote a short programme storing and averaging 
the data stream to convince myself that this would work and it does. If your 
16 bit A/D converter is noisy, keep it working hard!

    Have you tried altering the bandpass of your electronic filters, or maybe 
having two sets of filters and comparing the outputs? How about recording a 
higher frequency 'environmental noise' channel so that you can try to 
identify false signals? Could this give you the best improvement / cost + 


    Chris Chapman

Public Seismic Network Mailing List (PSN-L)

[ Top ] [ Back ] [ Home Page ]

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