🚀 go-pugleaf

In article <1992Nov18.203413.11509@rchland.ibm.com> lwloen@vnet.ibm.com writes:
>In message <1992Nov18.190513.10997@cis.uab.edu> Jim Dailey writes:

[ahem...actually, I wrote it.  Perhaps I'm taking blame, instead of credit?]
>
> [ long description of transposition system with some
>   "random" pad bytes thrown in before the ordinary
>   transposition step ]
>
>Transpositions are weak because they fall to some pretty universal,
>system-dependent attacks.
>
>The technique of interest here is called "multiple anagramming".
>...
>
>Now, the most common two words in English are "the" and "and".
>Whenever both ended up at the same location, the fragment:
>
>Msg1  t...h...e
>Msg2  a...n...d  (were... represent other ciphertext in between of same length). 
>
>would stand out.  After all, there is no substitution, so the real words are
>in there somewhere! 

Very good.  Thank you.  Let me try to summarize where we are so far:

0) original post: histogram to tell *if* a transposition cipher is in
   use.

1) my post: how about flattening the histogram, to defeat this?

2) your post: NG - we'll try anyway, and use second order statistics to 
   begin to unravel the transposition.  That is - you seem to reject the
   point of view represented by the original scheme, and will look for
   transpositions anyway (perhaps because you know from other sources
   that I'm using transpositions - the original poster was replying to
   a "what if you don't know the method" question)

Have I missed anything, yet?

If not - let's see if we can defeat the second order statistics.  Recall
that my suggestion (old as the hills, I'm sure) was to add pad
characters to defeat the first order statistics attack.  Can I instead
add pad bytes to defeat the second order statistics, instead?  For
example, you show an attack based on common triples - such as {t,h,e}
and {a,n,d}.  Does it help me to include lots of these triples,
scrambled, in the padding?

That is - if flattening the character histogram doesn't help, does it
help to add (scrambled) padding with the *same* distribution of
characters as the likely plaintext?  The theory is that this will give
the multiple anagrammer lots more "possible matches", and cause him to
chase down too many blind alleys.   Perhaps I can even arrange for the
triples (quadruples, whatever) that you are looking for to match many
times in the padding, rather than letting the statistics do this for me?

Ah...a light begins to dawn.  Nevermind...

==================

My other point was - the padding scheme relies on the receiver being
able to recognize the intended message.  You point out that this annoys
people.  To get around this, I proposed that the padding be chosen so
that no fragment of the padding formed a word in the lexicon - so that
an automatic dictionary lookup could extract the words in the message
from the jumble - this eliminates all whitespace, and eliminates the
need for a human to look at the full padded text.  I thought that an
advantage of this scheme might be that message text would be more evenly
distributed in the (untransposed) test.  I asked if this would be
counterbalanced by the "common lexicon" providing too much of a lever
for the opponent.  

I gather from your one-line reply to this point that the restricted
lexicon simply makes multiple anagramming easier.  Is this still true if
the padding has the same statistics as the message - and even
intentional misleading matches?   It appears so - even matches in the
padding tend to reveal bits of the transposition...oh well.

But...distributing the message differently within the padded blocks
makes it less likely that "common 3 letter combinations" will fall in
the same place in 2 different messages.  Does this matter?

===================

Finally (I promise, I'll go away after this) - is multiple anagramming
bothered by schemes which re-write the plain text in shorthand-style
alphabets designed so that the second-order statistics are flattened?
Perhaps this is too much of a tangent from the original point - if so,
sorry. 

Thanks for the reply - I learned from it.


-- 
Kenneth Sloan                   Computer and Information Sciences
sloan@cis.uab.edu               University of Alabama at Birmingham
(205) 934-2213                  115A Campbell Hall, UAB Station 
(205) 934-5473 FAX              Birmingham, AL 35294-1170