From: rbmccammon@mmm.com (Roy McCammon)
Newsgroups: sci.electronics.components
Subject: Re: Random Number Generators
Date: 30 Apr 1996 20:56:48 GMT
Organization: 3M

Every time some smart guy thinks he has a source of cripto quality
random numbers, some smarter guy proves him wrong, so I wouldn't
dream of saying the following circuit can produce cryto quality random
numbers, but you may want to try it.

The idea is to have two oscillators that have a random relationship to
each other, and then use one to sample the other.  I would make two different
types at very differnet frequencies and take care that there is no inadvertent 
coupling through poor ground or power supply connections.  At least one
would probably be very drifty.

Start with an about 10MHz crystal oscilator, and a 1000Hz rc oscilator such 
as a 555 or a few cmos gates.  Uses high temperature coefficient capacitors
like Z5U's and even thermisters if you are so inclined for the 1000 Hz 
oscilator.  Run the 10 MHz to a flip flop set up to toggle.  Call the
output of this flip flop T1.  T1 has close to a 50% duty cycle.  Call
the 1000Hz output T2.  

Connect T1 to the serial input of a shift register (8 stages should be fine) and
T2 to the clock input of the shift register.  Take your random bit stream
at the output of the last stage of the shift register.  If you need absolute
equal percentages of ones and zeros do this.  Take your bits in pairs.  Then
let 01 be a one and 10 be a zero.  Throw away 00 and 11.  You can do similar
things on greater numbers of bits if you are worried about higher order
correlations.  Gather the bits up into numbers of the size of your choice.

The purpose of the shift register is to suppress meta-stable outputs.  Use
only the last stage of the shift register.

Opinions expressed herein are my own and may not represent those of my employer.