Chronology Current Month Current Thread Current Date
[Year List] [Month List (current year)] [Date Index] [Thread Index] [Thread Prev] [Thread Next] [Date Prev] [Date Next]

Re: [Phys-L] high-quality random number generator



On 02/23/2014 11:45 AM, Bernard Cleyet wrote:

It’s the detector that creates biases.

Yes, that's an important contribution.

bc thinks no perfect RNG possible.

See previous reply. It said in part:
1) Nothing on this earth is perfect.
2) You can make an RNG that is so close to perfect that is
will be the least of your worries. For example, you can
have one that produces 159.9999 bits of entropy in a 160 bit
word ... with the entropy evenly distributed over all 160
bit positions. For details: http://www.av8n.com/turbid/

Building a decent RNG requires a range of skills, including physics,
electronics, programming, and cryptography.

A cryptographer not versed in physics will never figure it out.
A physicist not versed in cryptography will never figure it out.

The basic block diagram is:

detector --> cryptologic hash --> final output

Yeah, I know the detector contains biases. It also contains
entropy! It might contain only a few bits of entropy in a
32-bit word, but as long as it's not zero, and as long as I
know a reliable lower bound on the entropy content, I can
put it to good use.

Suppose I have 2000 bits of raw data, fresh from the detector,
known to contain at least 170 bits of entropy. I do not care
whether the other 1830 bits are doing. They can be known,
unknown, biased, correlated, or whatever. I run all of that
data through the hash. It produces a 160-bit output word
containing 159.9993 bits of entropy. That's not "perfect"
but it's darn close. Furthermore (!) it is computationally
infeasible for anybody to find the missing 0.0007 bit. That
is to say, it is computationally infeasible to distinguish
this process from a perfectly random distribution.
http://www.av8n.com/turbid/paper/turbid.htm

If this is not good enough for your purposes, please explain
what you are trying to do, and why this is not good enough.