Library
Module
Module type
Parameter
Class
Class type
This module contains predefined batteries of statistical tests for sources of random bits or sequences of uniform random numbers in the interval [0,1). To test a RNG for general use, one could first apply the small and fast battery SmallCrush
. If it passes, one could then apply the more stringent battery Crush
, and finally the yet more time-consuming battery BigCrush
.The batteries Alphabit
and Rabbit
can be applied on a binary file considered as a source of random bits. They can also be applied on a programmed generator. Alphabit
has been defined primarily to test hardware random bits generators. The battery PseudoDIEHARD
applies most of the tests in the well-known DIEHARD suite of Marsaglia. The battery FIPS-140-2
implements the small suite of tests of the FIPS-140-2 standard from NIST.
The batteries described in this module will write the results of each test (on standard output) with a standard level of details (assuming that the boolean switches of module Swrite
have their default values), followed by a summary report of the suspect p-values obtained from the specific tests included in the batteries. It is also possible to get only the summary report in the output, with no detailed output from the tests, by setting the boolean switch Swrite.set_basic
to false
.
Some of the tests compute more than one statistic using the same stream of random numbers and these statistics are thus not independent. That is why the number of statistics in the summary reports is larger than the number of tests in the description of the batteries.
The number of p-values in the array get_p_val
. For small sample size, some of the tests in the battery may not be done. Furthermore, some of the tests computes more than one statistic and its p-value, so get_n_tests
will usually be larger than the number of tests in the battery.
This array keeps the p-values resulting from the battery of tests that is currently applied (or the last one that has been called). It is used by any battery in this module. The p-value of the j
-th test in the battery is kept in (get_p_val()).(j-1
)], for 1 ≤ j ≤ get_n_tests()
.
This array keeps the names of each test from the battery that is currently applied (or the last one that has been called). It is used by any battery in this module. The name of the j
-th test in the battery is kept in (get_test_names()).(j-1)
, for 1 ≤ j ≤ get_n_tests()
.
val small_crush : Unif01.gen -> unit
Both functions applies SmallCrush
, a small and fast battery of tests, to a RNG. The function small_crush_file
applies SmallCrush
to a RNG given as a text file of floating-point numbers in [0,1); the file requires slightly less than 51320000 random numbers. The file will be rewound to the beginning before each test. Thus small_crush
applies the tests on one unbroken stream of successive numbers, while small_crush_file
applies each test on the same sequence of numbers. Some of these tests assume that the generator returns at least 30bits of resolution; if this is not the case, then the generator is most likely to fail these particular tests.
The following tests are applied:
Smarsa.birthday_spacings
with N = 1, n = 5×10⁶, r = 0, d = 2³⁰, t = 2, p = 1.Sknuth.collision
with N = 1, n = 5×10⁶, r = 0, d = 2¹⁶, t = 2.Sknuth.gap
with N = 1, n = 2×10⁵, r = 22, Alpha = 0, Beta = 1/256.Sknuth.simp_poker
with N = 1, n = 4×10⁵, r = 24, d = 64, k = 64.Sknuth.coupon_collector
with N = 1, n = 5×10⁵, r = 26, d = 16.Sknuth.max_oft
with N = 1, n = 2×10⁶, r = 0, d = 10⁵, t = 6.Svaria.weight_distrib
with N = 1, n = 2×10⁵, r = 27, k = 256, Alpha = 0, Beta = 1/8.Smarsa.matrix_rank
with N = 1, n = 20000, r = 20, s = 10, L = k = 60.Sstring.hamming_indep
with N = 1, n= 5×10⁵, r = 20, s = 10, L = 300, d = 0.Swalk.random_walk_1
with N = 1, n = 10⁶, r = 0, s = 30, L₀ = 150, L₁ = 150.val repeat_small_crush : Unif01.gen -> int array -> unit
repeat_small_crush gen rep
applies specific tests of SmallCrush
on generator gen
. Test numbered i in the enumeration above will be applied rep[i]
times successively on gen
. Those tests with rep[i]
= 0 will not be applied. This is useful when a test in SmallCrush
had a suspect p-value, and one wants to reapply the test a few more times to find out whether the generator failed the testor whether the suspect p-value was a statistical fluke. Restriction: Array rep
must have one more element than the number of tests in SmallCrush
.
val crush : Unif01.gen -> unit
val repeat_crush : Unif01.gen -> int array -> unit
Similar to repeat_small_crush
above but applied on Crush
.
val big_crush : Unif01.gen -> unit
val repeat_big_crush : Unif01.gen -> int array -> unit
Similar to repeat_small_crush
above but applied on BigCrush
.
val rabbit : Unif01.gen -> float -> unit
Applies the Rabbit
battery of tests to the generator gen
using at most nb
bits for each test. See the description of the tests in rabbit_file
.
Applies the Rabbit
battery of tests to the first nb
bits (or less, if nb
is too large) of the binary file filename
. For each test, the file is reset and the test is applied to the bit stream starting at the beginning of the file. The bits themselves are processed in nearly all the tests as blocks of 32 bits (unsigned integers); the two exceptions are Svaria.appearance_spacings
, which uses blocks of 30 bits (and discards the last 2 bits out of each block of 32), and Sstring.periods_in_strings
which uses blocks of 31 bits (and discards 1 bit out of 32). The parameters of each test are chosen automatically as a function of nb
, in order to make the test reasonably sensitive. On a PC with an Athlon processor of clock speed 1733 MHz running under Linux, Rabbit
will takeabout 5 seconds to test a stream of 220bits, 90 seconds to test a stream of 225bits, and 28 minutes to test a stream of 230bits. Restriction: nb
≥ 500.
Smultin.multinomial_bits_over
Snpair.close_pairs_bit_match
in t = 2 dimensionsSnpair.close_pairs_bit_match
in t = 4 dimensions.Svaria.appearance_spacings
Scomp.linear_comp
Scomp.lempel_ziv
Sspectral.fourier_1
Sspectral.fourier_3
Sstring.longest_head_run
Sstring.periods_in_strings
Sstring.hamming_weight
with blocks of L = 32 bits.Sstring.hamming_corr
with blocks of L = 32 bits.Sstring.hamming_corr
with blocks of L = 64 bits.Sstring.hamming_corr
with blocks of L = 128 bits.Sstring.hamming_indep
with blocks of L = 16 bits.Sstring.hamming_indep
with blocks of L = 32 bits.Sstring.hamming_indep
with blocks of L = 64 bits.Sstring.auto_cor
with a lag d = 1.Sstring.auto_cor
with a lag d = 2.Sstring.run
Smarsa.matrix_rank
with 32×32 matrices.Smarsa.matrix_rank
with 320×320 matrices.Smarsa.matrix_rank
with 1024×1024 matrices.Swalk.random_walk_1
with walks of length L= 128.Swalk.random_walk_1
with walks of length L= 1024.Swalk.random_walk_1
with walks of length L= 10016.val repeat_rabbit : Unif01.gen -> float -> int array -> unit
Similar to repeat_small_crush
above but applied on Rabbit
.
val alphabit : Unif01.gen -> float -> int -> int -> unit
alphabit gen nb r s
applies the Alphabit
battery of tests to the generator gen
using at most nb
bits for each test. The bits themselves are processed as blocks of 32 bits (unsigned integers). For each block of 32 bits, the r most significant bits are dropped, and the test is applied on the s following bits. If one wants to test all bits of the stream, one should set r = 0 and s = 32. If one wants to test only 1 bit out of 32, one should set s = 1. See the description of the tests in alphabit_file
.
alphabit_file filename nb
applies the Alphabit
battery of tests to the first nb
bits (or less, if nb
is too large) of the binary file filename
. Unlike the alphabit
function above, for each test, the file is rewound and the test is applied to the bit stream starting at the beginning of the file. On a PC with an Athlon processor of clock speed 1733 MHz running under Linux, Alphabit
takes about 4.2 seconds to test a file of 2²⁵ bits, and 2.3 minutes to test a file of 2³⁰ bits. Alphabit
and AlphabitFile
have been designed primarily to test hardware random bits generators. The four multinomial_bits_over
tests should detect correlations between successive bits by applying a serial_over
test to overlapping blocks of 2, 4, 8 and 16 bits. The hamming_*
tests should detect correlations between the successive bits of overlapping blocks of 16 and 32 bits, and the random_walk
tests consider blocks of 64 and 320 bits.
Smultin.multinomial_bits_over
with L = 2.Smultin.multinomial_bits_over
with L = 4.Smultin.multinomial_bits_over
with L = 8.Smultin.multinomial_bits_over
with L = 16.Sstring.hamming_indep
with blocks of L = 16 bits.Sstring.hamming_indep
with blocks of L = 32 bits.Sstring.hamming_corr
with blocks of L = 32 bits.Swalk.random_walk_1
with walks of length L = 64.Swalk.random_walk_1
with walks of length L = 320.val repeat_alphabit : Unif01.gen -> float -> int -> int -> int array -> unit
Similar to repeat_small_crush
above but applied on Alphabit
.
val block_alphabit : Unif01.gen -> float -> int -> int -> unit
Apply the Alphabit battery of tests repeatedly to the generator gen
or to the binary file filename
after reordering the bits as described in the filter Unif01.create_bit_block_gen
. Alphabit
will be applied for the different values of w ∈ {1,2,4,8,16,32}. If s < 32, only values of w≤s will be used. Each test uses at most nb
bits. See the description of the tests in alphabit_file
.
val repeat_block_alphabit :
Unif01.gen ->
float ->
int ->
int ->
int array ->
int ->
unit
Similar to repeat_small_crush
above but applied on BlockAlphabit
. The parameter w is the one described in block_alphabit
. Restrictions: w ∈ {1,2,4,8,16,32} and w≤s.
val pseudo_diehard : Unif01.gen -> unit
Applies the battery PseudoDIEHARD
, which implements most of the tests in the popular battery DIEHARD or, in some cases, close approximations to them. We do not recommend this battery as it is not very stringent (we do not know of any generator that passes the batteries Crush
and BigCrush
, and fails PseudoDIEHARD
, while we have seen the converse for several defective generators). It is included here only for convenience to the user. The DIEHARD tests and the corresponding tests in PseudoDIEHARD
are:
Smarsa.birthday_spacings
with n = 512, d = 224, t = 1 and r = 0,1,2,3,4,5,6,7,8,9 successively. The test with each value of r is repeated 500 times and a chi-square test is then applied.Smarsa.matrix_rank
.Smulti.multinomial_bits_over
with Delta = −1, n = 221, L = 20.Smarsa.collision_over
with n = 221, d = 1024, t = 2 and all values of r from 0 to 22.Smarsa.collision_over
with n = 221, d = 32, t = 4 and all values of r from 0 to 27.Smarsa.collision_over
with n = 221, d = 4, t = 10 and all values of r from 0 to 30.Sstring.hamming_indep
.Snpair.close_pairs
with N = 100, n = 8000, t = 2, p = 2, m = 1.Snpair.close_pairs
with N = 20, n = 4000, t = 3, p = 2, m = 1.Smarsa.savir_2
.Sknuth.run
.The NIST (National Institute of Standards and Technology) of the U.S. federal government has proposed a statistical test suite for use in the evaluation of the randomness of bitstreams produced by cryptographic random number generators. The test parameters are not predetermined. The NIST tests and the equivalent tests in TestU01 are:
Sstring.hamming_weight_2
with L = n.Sstring.hamming_weight_2
.Sstring.run
.Sstring.longest_head_run
.Smarsa.matrix_rank
.Sspectral.fourier_1
.Smarsa.cat_bits
.Smultin.multinomial_bits_over
.Svaria.appearance_spacings
.Scomp.lempel_ziv
.Scomp.linear_comp
.Smultin.multinomail_bits_over
with Delta = 1.Smultin.multinomial_bits_over
with Delta = 0, and to Sentrop.entropy_disc_over
or Sentrop.entropy_disc_over_2
.Swalk.random_walk_1
.Swalk.random_walk_1
.Swalk.random_walk_1
.val fips_140_2 : Unif01.gen -> unit
These functions apply the four tests described in the NIST document FIPS PUB 140-2, Security Requirements for Cryptographic Modules, page 35, with exactly the same parameters (see the WEB page at http://csrc.nist.gov/rng/rng6_3.html). They report the values of the test statistics and their p-values (except for the runs test) and indicate which values fall outside theintervals specified by FIPS-140-2. The first function applies the tests on a generator gen
, and the second applies them on the file of bits filename
. First, 20000 bits are generated and putin an array, then the tests are applied upon these. The tests applied are:
Smultin.multinomal_bits
with s = 32, L = 1, n = 20000.Smultin.multinomial_bits
with s = 32, L = 4, n = 5000.Sstring.run
.Sstring.longest_head_run
.