Re: Leopard15/L15

Greg Rose wrote:

Now you claim somewhere that this is better than
RC4 because in RC4 the "i" variable is known to
the attacker, whereas here your "x" and "y" both
start out unpredictable. Fine. But in RC4 the "j"
variable bounces around in ways that are very hard
to predict. Here, your "x" and "y" both vary
completely predictably, and their configuration
repeats with a known period of (256*255). I really
can't believe this is a good thing.

I wrote a simple autocor test for the lags 255, 256 and 256*255 with
1GB of output it passed. My summation function was simply

sum += (data[x] == data[x-j]) ? 1 : 0;

So if there IS a correlation at one of those lags, it isn't as simple
as bytes being equal. Maybe build a digraph table for the tuple
(data[x], data[x-j]) and find out which is the most probable?

Tom

.

Relevant Pages