From: smb@research.att.com
Date: Sun, 19 Jun 94 06:35:46 PDT
To: pfarrell@netcom.com (Pat Farrell)
Subject: No Subject
Message-ID: <9406191335.AA24696@toad.com>
MIME-Version: 1.0
Content-Type: text/plain


There are a few things to watch out for.  First, it's really easy
for subtle (or not so subtle) biases to be present in a noise source.
These can be due to component drift, external noise (i.e., power
supply coupling), etc.  You want a design that isn't sensitive to
such things, if possible.  Second -- and it's partly a corollary to
the first -- the designs I've seen for real RNGs have always included
a scrambler step, to mix up the bits, account for biases, etc.

The first such scrambler was, I think, described by von Neuman himself.
I have the citation in my office; I'll try to post it tomorrow.