From: Bill Stewart <stewarts@ix.netcom.com>
Date: Tue, 16 Jul 1996 15:21:29 +0800
To: Jack Mott <thecrow@iconn.net>
Subject: Re: Hardware RNG
Message-ID: <199607160009.RAA28737@toad.com>
MIME-Version: 1.0
Content-Type: text/plain


>	Got a geiger counter plugged into the game port
Cool.  As other people have noted, the game port may do timing things
that can degrade the quality of your random numbers a bit, so don't
use too many bits per sample unless you really understand its behaviour.
>	Weak radioactive source next to it (dont worry wont fry you)
>	Use a PRNG string to do create an RC4 S-box
>	Cycle through the S-box in a tight loop, each time checking to 
>see if the geiger counter got a hit, if it did, record that number in 
>the S-box as our first byte, do this 100 times, and we have 100 random 
>numbers.

Other people have also commented that this is bad, unless you can
show that the math behind it works well and usefully.

>any thoughts? It seems to work well, no basic stat analysis reveals any 
>pattern, and physicists have backed me up on radioactive decay being 
>'the great randomizer'.

The real randomness you have is from the radioactive decay intervals,
plus or minus any gain from the game port hardware.  No need to muck that
up with RC4, especially in ways that may add predictability.  
Be very careful with statistics - if something's grossly skewed, the usual
tests will pick it up, but they can't tell if there's a mathematical way
for somebody who knows data point N to predict point N+1.  If you look
at the output of your geiger counter system, you'll probably see a roughly
exponential distribution of time intervals between hits (which is the
result of uniformly distributed events like radioactivity) modified a bit
by your game port behavior (e.g. the times may all be multiples of 1ms or 17ms.)
You can extract decent random bits from this by inverting the distribution;
if you want really high quality randoms, at lower resolution, 
you can do something like pick two samples and return 0 or 1 depending on 
which one is shorter - especially useful if you don't trust the game port.

Your generator can't give you more real bits than that; the RC4 just
obfuscates it, makes it harder to evaluate the real strength, and
evens out the distributions.  You could still crunch the bits through
MD5 or something, but only use as many bits of output as you're sure
the generator is really giving you.

#				Thanks;  Bill
# Bill Stewart +1-415-442-2215 stewarts@ix.netcom.com
# http://www.idiom.com/~wcs
#				Re-delegate Authority!