Matrix-based tests. Although the previous statistical test looks at the sequence of
numbers linearly, a large portion of the more advanced testing techniques view it in terms
of vectors and matrices. These tests will take the values of the generator, and sequentially
add them into a hypercube, which is a cube with k dimensions (L’Ecuyer, 2007). From
here the tests vary, although many of them conclude with a chi-squared calculation on
their output (Chan, 2009). In a nearest pair test, Euclidean distances are computed and the
nearest pairs are used to look for time-lagged patterns. It has been proposed that a large
number of pseudo random generators based on circuits will fail the nearest pairs test, and
it is one of the more stringent tests available. Multidimensional tests are often specifically
tailored to look at random number generators, unlike the linear tests which are
RANDOM NUMBER GENERATION 25
multipurpose. The calculations in matrix-based tests are complex, and will not be
presented in this thesis.
Do'stlaringiz bilan baham: |