Re: Confirmation of Golic's results for RC4

From: Michael Amling (nospam_at_nospam.com)
Date: 01/08/04 

Date: Thu, 08 Jan 2004 17:48:54 GMT

Bartosz Zoltak wrote:
> I tried to find experimentally the bias described by Golic in "Linear
> Statistical Weaknessof Alleged RC4 Keystream Generator", however I got
> no result (for about 2^40 generated outputs).
>
> Has anybody tried to do this before and could point where I made a
> mistake in my implementation?
>
> I am generating a keystream X with RC4. Then I take, for the following
> n:
> a=X[n] and 1,
> b=X[n+2] and 1,

   I've never seen Golic's paper, but, are you sure that n+2 isn't
supposed to be n+1? Scott Fluhrer would know for surer.

> c=a+b
>
> I increment counter S when c=1. Then I divide S by the length of
> generateg keystream (n) and I was expecting the result to be
> statistically significantly more than 0,5.
>
> However I got a ratio of 0,49999952, which could be statistically
> accepted as the random expected 0,5.

--Mike Amling

Quantcast