[sdiy] VERY cool/bizarre: Buchla noise

[Rainer Buchty]

>OK philosphy students... this is a good one frankly. Whenever I try to
>explain PRBS stuff to people (usually customers, usually in regards to
>internet traffic data patterns or framesize distributions) I end up using
>phrases like "it's not really random but it's pretty darn close" and
>"subjectively random" and "it might as well be truly random" and "but it
>will repeat after a while" and "it's generated by a digital machine so it's
>not truly random" etc. etc. Looks like the definition of "random" is what's
>in question here.

It depends on what the use of your "random" signal is. For *strong* 
cryptography PRNGs are of no use cause once you discover the seed that 
very PRNG sequence is useless.

If your cryptographic system has only *one* defined end-to-end
connection, e.g. those little token computers for company VPNs or TANs
for online banking, then big (and complex, i.e. a more than
one-dimensional seed, and a more than one-dimensional computation)  
enough PRNGs can be useful cause you can easily -- and at very little
cost -- re-initialize the PRNG and spit out an entirely new sequence of 
random numbers.

If it's just for audio reasons, make it a 24-bit (or the minimum size
which isn't perceived as a looped sequence by human ears) feedback shift
register and avoid lock-up as mentioned by Harry by design.

>If we can all agree that that is random, then the cutoff point between
>psuedo-random and random is somewhere in between.

Mathematically (and especially cryptographically), we can't, no matter
how long the period of a PRNG is.

That's why private/public key generators ask you to type in random key 
patterns and wiggle the mouse to get some *real* source of uncertainty, 
and why crypto cards feature good old transistor noise circuits.

Rainer