Re: Poker Tests in restricted envionments

From: Jean-Luc Cooke (
Date: 08/24/05

Date: 24 Aug 2005 18:13:52 GMT

Xi-Squared tests would be the classical way of testing an observed
with an expected distribution. Tricky to do, however with only 1
sample. Especially from a non-white source (ie. there is no way of know
if FF will *ever* show up from your device even if it's workng 100%

Random events in nature do not exhibit an even statistical distribution
accross all possible finite values. Keep that in mind.


David Eather <> wrote:
> Hash: SHA1

> I've got a problem and I'll ask here first. I have a random noise
> generator that collects bits of data from a transistor noise source,
> processes it and then send it of to a PC serial port at 4800 or 9600.
> I
> have a maximum of 14 bytes of ram (probably only 9 usable) so I can't
> perform a full poker test.

> What I am thinking of doing is to count the number of 1 particular
> byte in a
> sample of 2**16 i.e count the number of FF bytes in what should be a
> random
> stream of bytes ranging 00 to FF. If the test fails at a reasonable
> p, then
> the byte value gets tested again. A successful test would result in
> testing
> the next byte in sequence - by not advancing after a failed test it
> should
> pick up any "stuck" (i.e. broken) conditions reasonably quickly

> I'm also performing a bit frequency test and steps to also remove
> auto
> correlations.

> I have 2 questions. What is the statistic to use to show if
> particular
> byte samples are over or under represented in a stream " -
> probability is
> not my forte.

> All help appreciated
> The other question is more user friendly related. If you had such a
> device
> how would you prefer it to send out the data? 32-bit words or 8-bit
> words?
> Prefixed with "0x", "&H", a space or nothing at all


> Version: PGP 8.0.3

> iQA/AwUBQwyVGJS9Fk5okqe7EQKgnQCdGcYR8pb6VH+aKXa/lnqjImQ7hYUAoMwb
> jLP5agnyhtbwEpZYv6jsOfX6
> =h9ta