This is a set of reference data for testing an implementation of the
non-iid estimators specified in NIST's SP800-90B document (the January
2018 version).

This archive includes a few example data sets, all approximately
1,000,000 samples, along with the corresponding results from UL's tool
for comparison.

We have included the timing for each test; all of these tests were run
on an Amazon EC2 c5.large instance.

Bit strings are provided in two encodings, "packed" and "nist".
"packed" strings are bits arranged from MSb to LSb. The bytes are ordered
in the order they were produced (so the first bit is the MSb of the
first byte, and the ninth is the MSb of the second byte, etc.) These are
in the "packed" directory, and their file names end with "-packed.bin".

"nist" strings are the format that the 2016 NIST python tool uses; here
there is 1 byte per symbol, in the order they are present in the file. These
are in the "nist" directory, and their file names end with "-nist.bin".

A description of the test cases follows:

random-bits: statistically ideal bit-wise data.
biased-random-bits: Significantly biased bit-wise data.
deterministicRingOsc: The output of a wholly deterministic ring
oscillator, sampled periodically.
ringOsc: The output of a non-deterministic ring oscillator (with
fairly realistic parameters), sampled periodically.


Byte strings (under the "byte-strings" directory): these are all 8 bit symbols, 
one symbol per byte.
	random-bytes.bin: statistically ideal byte-wise data.
	biased-random-bytes.bin: Significantly biased byte-wise data.
	LRS-testcase.bin: an artificial test case that encourages bad performance within the LRS test.
	normal.bin: Normally distributed data.

The corresponding results are in the ".results" files.

Please feel free to contact me with any comments or questions.

Joshua E. Hill <joshua.hill@ul.com>
