New algorithm of RNG to replace / dev / random

As you know, the generation of random numbers is a key element for applications of cryptography, economics, information security, simulation (weather forecasts) and many others. All existing RNG algorithms use an external entropy source, for example, a processor clock counter, sound card noise, user mouse movement, etc.

German researchers Bernard Fechner (University of Hagen) and Andre Osterloh (BTC AG) have announced a breakthrough in random number generation methods. They developed an algorithm that provides a discrete uniform distribution up to 20 times better than existing methods. The paper includes test results comparing the quality of random numbers generated by different methods.

The RNG developed by the Germans reads a sequence of bits from memory (a metastable state ) and swaps 0 and 1 according to the table. The table is an additional level of randomness that allows us to increase the quality of random numbers by an order of magnitude. The specificity of the method (reading from memory) ensures the maximum possible speed of the algorithm. It is not only better, but also an order of magnitude faster than the LFSR algorithm plus a processor clock counter - this bundle is used in the standard program / dev / random , which comes with Unix.

It is assumed that the ideal sequence of the RNG is close in properties to real random numbers, such as, for example, readings from cosmic radiation detectors or ionizing radiation events, that is, there is not even a theoretical way to predict the next number in the sequence.
')
The results of their work, Bernard Fechner and Andre Osterloh, published in the journal International Journal of Critical Computer-Based Systems. [A meta-level true random number generator. Int. J. Critical Computer-Based Systems, 2010, 1, 267-279, PDF ]

via ScienceDaily