Since a crap random number generator has been recently discussed...
A few years ago I put on the internet two type of hardware random number
generators. They both used a simple noise circuit which solved many
adjustment problems and allowed for cheap or even home construction.
While the design solved some problems it introduced autocorrelations
which were dealt with by using a new/untried technique suited for very
low performance micro-controllers i.e. use a bit from the ADC large
enough not to be significantly affected by LSB non-linearities and throw
the rest of the bits away (compared to sending all the samples through a
hash function).
With my own software based on the tests described in HAC I had good
results and also good results with diehard.
I ran the ent tests over a 512Mb data file collected early in the piece
and was heartened by the result
>
> C:\>ent e:\data.rnd
> Entropy = 8.000000 bits per byte.
Woo Hoo!
>
> Optimum compression would reduce the size
> of this 536870912 byte file by 0 percent.
!
>
> Chi square distribution for 536870912 samples is 222.47, and randomly
> would exceed this value 90.00 percent of the times.
This is OK. I wish I had collected a few more large samples.
>
> Arithmetic mean value of data bytes is 127.5009 (127.5 = random).
> Monte Carlo value for Pi is 3.141765710 (error 0.01 percent).
> Serial correlation coefficient is -0.000020 (totally uncorrelated = 0.
!! Not bad for a new and untested method.
