From:	"Radar" <p_radar AT hotmail DOT com>
Newsgroups:	comp.os.msdos.djgpp
Subject:	Re: real random numbers
Date:	Wed, 25 Mar 1998 00:28:55 +0100
Organization:	Nextel
Lines:	26
Message-ID:	<6f9fku$30q@argo.telecom.cz>
References:	<m0yHU7T-000S2iC AT inti DOT gov DOT ar> <3517F669 DOT 53B2 AT pobox DOT oleane DOT com>
NNTP-Posting-Host:	194.228.30.97
To:	djgpp AT delorie DOT com
DJ-Gateway:	from newsgroup comp.os.msdos.djgpp

>Randomness is a pure mathematical concept. The only way to know whether a
>series is random or not is to test it. Hence my question: has anyone
>tried to test such a "natural noise series" to classical tests for random
>numbers (Knuth's spectral test for instance)? It would be a good way to
>know whether such natural noise sources make good RNG or not (but again,
>I suspect the latter).


I am not sure, but I think that spectral test is only for congruential PRNG.
It is test for correlation of x(i) and x(i+k) and it is theoretical test.
If you want to test physical device for randomness, the only way to do is
use some empirical test, such as Pearsons' chi square test, KS-test, ups and
downs and many others.
If you would like to look at some PRNG and test for randomness, look at
DIEHARD- the battery of tests from G.Marsaglia
http://stat.fsu.edu/~geo/diehard.html. There are included sources for some
simple PRNG. To look at results of test for physical devices look at
http://random.mat.sbg.ac.at .
As for your thinking about best RNG, good results are for combined
generators, for example congruential with random noise.

Paul

- Raw text -