(Fwd) RE: Noise Sphere in Java

[Deranged Mutant]

On 18 Apr 96 at 11:13, Hal wrote:

[..]
> There was a post on cypherpunks a few days ago claiming that the output
> of "pgp +makerandom=2000 rand.pgp", which uses the PGP internal RNG,
> showed visible structure in the output of a noise sphere program.  I
> tried modifying Chuck's Java version and ran it with data from pgp
> +makerandom, and it looked fine to me.

Hm. I didn't see that post... (I wonder if he plotted an ASCII
file... or I suspect a troll?)

> However, that isn't very meaningful, because I find the that program
> output looks fine even with Chuck's version, which is using the Java
> internal RNG, a LCM from Knuth.  Specifically the Java code is:
[..]

A lot of bad PRNGs look good with noise spheres.... but what I found 
while experimenting with them was data based on timer drift samplings 
that passed a variety of randomness tests but showed up with very 
clear patterns on a noise sphere... so I made the original post to 
the list and sci.crypt about it.

Noise spheres will show certain correlations in the data.  It doesn't 
mean that the RNG is crypto-usable if it looks good. But if it looks 
bad then you know to be suspicious.... a visual image makes a clear 
point better than analyzing a lot of numbers on a page.

By no means should it be relied upon as a sole test.

[..]
> see it.  Oh, sometimes I can almost convince myself I'm seeing structure,
> but it is never repeatable from run to run.  And I see as much with the
> output of pgp.  Particularly when the graph is sparse you can see some
> clumping, but I think it is just random noise.

> If the noise sphere can't even reject an LCM RNG, it doesn't sound
> that useful to me for crypto purposes.

If you see definite spirals and loops then it's more than random 
noise.

It doesn't pass or reject anything. It plots data. If the data shows 
*definite* patterns, you can reject the RNG.  If not, use more 
sophisticated tests.

[..]
> at right angles to this.  Because polar coordinates are not uniform in
> space, the points are clustered along the north-south axis inside the
> globe.  You can see this in the views, where the upper right view should
[..]
That's probably what the poster to c'punks wrote about...

*sigh*



 
Rob. 

---
Send a blank message with the subject "send pgp-key"
to <WlkngOwl at unix.asb.com> for a copy of my PGP key.