[Gnuk-users] TRNG output

[NIIBE Yutaka]

On 08/28/2015 02:07 AM, Kurt Roeckx wrote:
> What they all seem to lack is a calculation (and proof) of the
> (minimum) entropy.

I agree.  In my opinion, it would not be the job of implementer, but
of scientist.  I found that we have fast ADC these days and we can use
it as a source of noise, I did some research to support this idea, and
I implemented my idea.

For scientists, I suggest having a formal model of a device (and then
doing some simulations and experiments if the model really matches the
implementation), with references.  For NeuG, I even suggest to have a
minimal formal model which only use the LSB from output of ADC.

Currently, NeuG lacks such a (accurate) formal model or proof.

Nevertheless, it implements runtime testing (Repetition Count Test,
Adaptive Proportion Test) against its sequence continuously, following
the draft of NIST SP 800-90B.  It can detect failure.

Please note that runtime testing doesn't fulfill the need of a formal
model at all, because there remain a possibility where a scientist can
guess/control the sequence even if it looks quite random.
--