On 08/02/2012 15:37, adacrypt wrote:
> Having run the experiment again without changing anything and without
> any tampering I can verify now that the (pseudo) randomness of any
> ciphertext (over and above the cipher randomness imparted by the
> encryption keys internally) that will use this cryptography is 94% of
> true randomness. This is a result that was not easily available to me
> previously because of limitations.
>
> This result is sustainable at any time in the future.
>
> This is a good result of this experiment and I am impressed with the
> performance.
Try this on; the cipher is one-to-many. A particular plaintext P(a)
character will be transformed into one of a set of triples in C(a).
This set of triples C(a) is unique to P(a), so it is possible to
construct a dictionary of C(a) -> P(a) for all a. The dictionary
an no more than an annoyance to create, and currently inefficient
to run as I am using pretty basic scripting tools and linear searches
to prove the concept.
The use of the large "vector" triples is what introduced this weakness.
I discovered this by enciphering 16Mi of one character and looking for
duplicates, and noticing that a MUCH smaller number of triples was
multiply reused.
This cipher is _busted_.
Do you even know you have a bug in your constant tables? Looks like
a silly typo. One of the change-of-origin tables has a mistake in it
that results in the entry being zero. And this has been copied around a
bit. How much care/thought went into those numbers?
M
--
Mark "No Nickname" Murray
Notable nebbish, extreme generalist.