Надежность двухуровневого подхода к тестированию в наборе тестов NIST SP 800-22 / А. А. Серов
Year: 2026
Type: article
Abstract: The two-level approach for testing random number generators involving the well-known NIST SP 800-22 test suite, i. e. counting the sequences passing a basic test and checking the p-values distribution with a chi-square test, was considered. Such an approach may increase the reliability of the test. However, it is sensitive to the approximation error introduced by the computing of p-values. In this paper it is shown that for AES-based sequences two-level testing approach is not reliable too. Systematic error in the computing of the p-values is dependent only on the accuracy of ap proxi ma tion of the exact distribution of statistic by its theoretical counterpart and the number of bits in the analysed sequences n. For a reliable second-level test, this error should be smaller, or at least, approximately equal to [формула], where Ϭ = [формула] is the standard deviation from the mean number of particles in a bin for equiproexample in the second-level test of the Frequency test of NIST SP 800-22 test suite with n = 2[20] the number of tested sequences N should not exceed 26 000. To completely eliminate the systematic error appearing in the Frequency test when determining the number of bin from k bins (disjoint subintervals of the interval [ 0,1]) to which the p-value belongs the two-sided estimates of the quantiles of the binomial law are proposed.
Author: Serov, A. A.
Institutions: Russian Academy of Sciences, Steklov Mathematical Institute
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7223477817
