[Cryptography] new PRNG family
ab at bri-c.de
Tue Dec 16 16:46:27 EST 2014
I finished the tests of all three variants of breeze generator for output doublets.
I did the usual testing for random characteristics (Each variant passed NIST Suite on 20 sets of 100 sequences of 10^6 byte length).
Furthermore i checked for output doublets (indication of repeating period) in 100 | 200 | 400 GB output using a bloom filter (2**36 bit / 7 lockers / hash: siphash) on 2 consecutive output-words of length 256 / 512 / 1024 from the breeze128/256/512 respectively. I found one (possible) solitary doublet within the 87th GB in the 98th sequence of breeze128 output and a second possible doublet in the breeze512 output within 91GB of the 61rst sequence).
I found no pattern of degradation.
I take this being the result of pseudo orbit hopping instead of exploring the pseudo orbits. Such degradation had been shown in studies of chaos in logistic maps in finite (digital) environment. But degradation is related to rounding in exhaustive exploration within pseudo orbits. But breeze works different. By using the outcome of one calculation from one pseudo orbit for calculating a function of another map with different pseudo orbits, degradation can be prevented effectively.
More information about the cryptography