Cellular Automaton-Based Emulation of the Mersenne Twiste

Abstract

The Mersenne Twister (MT) (MT19937), developed 30 years ago, is the de facto pseudorandom number generator (PRNG) used in many computer programs. This paper proposes a candidate that offers a randomness quality that is better than MT19937 and its sisters SFMT19937 and TinyMT. A special three-neighborhood, two-state cellular automaton (CA), called CA (150′) is the underlying model of this PRNG. The same working style of MT19937 is used, while avoiding the problems of the MT, like a large state space and the zero-access initial state problem. Nonlinearity is added in the base simple linear CA such that the properties of the base CA are not violated. Finally, a PRNG is developed using this CA that beats MT19937 as well as its advanced versions over the standard empirical platforms Dieharder, TestU01 and NIST.