A Proposal of a New Statistical Testing Framework for Random Number Generators

Abstract

Abstract. This paper proposes a new statistical testing framework for random number generators. First, we show that the evaluation of “generators” using NIST SP.800-22, which is widely used today, is incomplete. Next, we show that the statistical evaluation of the generator can be interpreted as a “random sampling” evaluation and give some examples of random-sampling-tests. After that, under the assumption that the test will be repeated many times, this paper proposes a new framework of statistical testing for generators where the testing procedure and final judgement rules are clearly defined. Finally, we show some examples using the proposed framework.