From: norm@netcom.com (Norman Hardy)
Date: Mon, 25 Sep 95 14:25:25 PDT
To: wb8foz@nrk.com
Subject: Re: "random" number seeds vs. Netscape
Message-ID: <ac8ccae902021004990a@DialupEudora>
MIME-Version: 1.0
Content-Type: text/plain


At 12:29 PM 9/24/95, Ed Carp [khijol SysAdmin] wrote:
....
>I learned the hard way - keep the transmitters away from a reverse-biased
>doide acting as a noise generator.  Only until I examined the output did
>I realize it wasn't random.  I fixed it, though, by looking at the output
>and testing its randomness.
....
Very interesting. I wouldn't be too sure that a transmitted signal at a
single frequency is the only signal that an opponent could use to bias your
random numbers. How do you "test for randomness". I think that signal to
noise arguments, phrased in terms of entropy, can protect you against
unknown and unwanted signal. (Ironically you want a very low signal to
noise ratio!) Perhaps you merely take n/(S/N) bits from the HRNG when you
need n bits and run them thru MD5. Here S is the signal strength of the
maximum plausible unwanted signal, and N is the noise of the diode.

I encourage both diode theorists and information theorists to quibble with
the above formula!