Re: FFT test with few kbits

From: Cristiano (cristiano.pi_at_NSquipo.it)
Date: 02/28/04 

Date: Sat, 28 Feb 2004 10:51:38 GMT

"Paul Pires" <diodude@got.net> ha scritto nel messaggio
news:jiS%b.31428$2j2.22436@newssvr25.news.prodigy.com...
>
> Cristiano <cristiano.pi@NSquipo.it> wrote in message
news:IeR%b.48271$IT2.1375675@news4.tin.it...
> > "Paul Pires" <diodude@got.net> ha scritto nel messaggio
> > news:ydO%b.17974$wj3.9189@newssvr29.news.prodigy.com...
> > >
> > > Cristiano <cristiano.pi@NSquipo.it> wrote in message
> > news:K5E%b.42021$IT2.1218078@news4.tin.it...
> > > > Paul Pires wrote:
> > > Regardless of the name,
> > > it will be a usefull test if it can discriminate better than other
tests used or
> > > detect a bias other tests are not sensitive to.
> >
> > I dislike to compare the tests.
>
> I didn't say you should compare the tests. I suggested you characterize
> your test. One way to do so would be to run your test and many other
> known tests on the same data from many good and bad generators .
> This information would be helpfull. It would show if it detects similarly
> (or even identically) to other tests.
>
> Doing this, I found that the 4 tests supplied in the ENT test package were
> actually only three. One test result when inverted, baseline shifted and
> scaled was always EXACTLY the same as another test result for the
> same data tested. Someone put in a lot of work into developing a test and
> coding it when it was directly equivalent to another test already
installed :-)

If EXACTLY means that you get the same p-values, then I agree.
For example, in FIPS-140 when you take the poker test with m=1 (m is the
block length) you get _exactly_ the frequency test; this means that the
former works _exactly_ like the latter and it gives _exactly_ the same
p-value, so the frequency test is totally useless when you have the poker
test and you use it with m=1.

Ok, I can try what you said, but it will take long time...

Cristiano