Newsgroups: sci.crypt
Path: cactus.org!milano!cs.utexas.edu!qt.cs.utexas.edu!yale.edu!think.com!
+     mips!pacbell.com!network.ucsd.edu!qualcom.qualcomm.com!chicago.qualcomm.
+     com!karn
From: karn@chicago.qualcomm.com (Phil Karn)

Subject: Re: generating one-time pads
Message-ID: <1992Jun27.093604.9831@qualcomm.com>
Sender: news@qualcomm.com
Nntp-Posting-Host: chicago.qualcomm.com
Organization: Qualcomm, Inc
References: <2668@accucx.cc.ruu.nl> <2796@accucx.cc.ruu.nl> <1992Jun27.064216.
+           21116@cactus.org>
Date: Sat, 27 Jun 1992 09:36:04 GMT
Lines: 13


I propose a simple experiment. Widen the counter that is incremented
in the tight loop to 16 or 32 bits, and save the full counter value on
each interrupt. Then run an autocorrelation on the results.

This should give you a good idea of just what the oscillator phase
noise process, plus other perturbing processes, like memory refresh,
looks like. And from that you should be able to determine how often to
sample, and how many bits to save from each sample, without impairing
the randomness of the result.

Phil