Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!uwm.edu!uakari.primate.wisc.edu!brutus.cs.uiuc.edu!usc!ucsd!helios.ee.lbl.gov!ux1.lbl.gov!mikec
From: mikec@ux1.lbl.gov (Mike Chin)
Newsgroups: comp.dsp
Subject: FFT S/N replies
Message-ID: <4648@helios.ee.lbl.gov>
Date: 19 Jan 90 01:48:03 GMT
Sender: usenet@helios.ee.lbl.gov
Distribution: usa
Organization: Lawrence Berkeley Laboratory, Berkeley
Lines: 28
X-Local-Date: 18 Jan 90 17:48:03 PST

The problem with my trying to find S/N from an FFT was 
that I was doing something like:
	sum(bins due to noise)
	----------------------------
	sum(bins due to fundamental)

I should have been doing:
	sqrt{sum((bins due to noise)^2)}
	----------------------------
	sqrt{sum((bins due to fundamental)^2)}
Thanks to  oakhill!radio!charlie@cs.utexas.edu (Charlie Thompson)
for this.

Also, Steve Omner (steveno@tekigm2.men.tek.com) reminded me to
look at the effects of my windowing. My blackmann-harris used
a0-a3 coefficients that has sidelobes at the level I'm trying
to resolve. The ADC manafacturer is sending me their new window
set; we'll see if it makes any difference.

Bob Jewett <jewett%hpl-opus@hplabs.hp.com> advocates using a direct
sine fitting, and sent me a nice program that crunched my simulated
data very well.

Thanks also to the others who gave hints.

Mike Chin
Lawrence Berkeley Labs
MJChin@lbl.gov