Re: advice sought on key/data histogram analysis of rijndael/128 and serpent
From: lkcl (lkcl_at_lkcl.net)
Date: 23 Oct 2005 09:16:08 -0700
> That the
> statistical code written by me reflects that as the number of blocks
> increases any such correlation disappears
then, to me, that would indicate that the use of additional
blocks is not a suitable test for looking for problems :)
i'm comfortable with increasing the _number_ of tests
rather than increasing the block size as a means to
increase detection of unexpected correlations.
i'm presently working on some variations on the theme,
to endeavour to show correlations between bits set in
the input data and bits set in the output data.
i took inspiration from your idea of measuring all counts
rather than just p-values < 0.01 - i now have TWENTY
sets of histograms, representing the range of p-values
0 <= pv <= 0.01, 0.01 < pv < 0.04 etc.
the news isn't good, i'm afraid: there are definite
and bluntly obvious correlations evident in the
histograms i'm getting. i don't want to be crying
"wolf" so i'll only post results and code when i know
a bit more: being able to show correlations between
input data and output data - independent of the key -
is _really_ really serious "oops" that i'd better be
damn sure about, first.