Naïve Analysis of TRNG

Please report issues with the DM41X Beta Firmware in this sub-forum
rprosperi
Posts: 1703
Joined: Mon Apr 24, 2017 7:48 pm
Location: New York

Re: Naïve Analysis of TRNG

Post by rprosperi »

pauli wrote:
Sun Feb 09, 2020 11:45 am
Using a 41 for raffle draws? It's biased and the results are immediately suspect.
Monte Carlo simulations? It's biassed and the results are immediately suspect.
Dice rolling? It's biassed and the results are immediately suspect.
Dividing up pizza slices? It's biassed and the results are immediately suspect.

I'm willing to assist with improving it. Subject to time availability (which is very limited at present but this is something I've experience with).

Pauli
Pauli, I am most grateful for your offer to help! Even if not pursued, a constructive attitude is always appreciated.

As for the examples cited, the part I wonder about is not if bias exists, but rather the "immediately suspect" phrase. I think dividing Pizza can be dismissed as with only 8 pieces I don't think you would notice any bias. :lol:

For the others, is it not the case (I'm genuinely asking, as I don't have deep background here at all) that bias would only be revealed after a very large number of samples have been issued? I'm not suggesting at all that this makes it any more correct, only wondering about the 'immediate' part of the way its characterized.
--bob p

DM42: β00071 & 00282, DM41X: β00071 & 00656, DM10L: 071/100
User avatar
pauli
Posts: 252
Joined: Tue May 02, 2017 10:11 am
Location: Australia

Re: Naïve Analysis of TRNG

Post by pauli »

To be completely honest, the bias cutting pizza is unlikely to matter -- even so I'd position myself favourably if I could :) For 12 pieces, I suspect it would be measurable.

I'm coming from a cryptographic background where any bias is immediately suspect. For example, refer to this article. The bias in the second byte of output is suspect. There are worse problems with this algorithm.

Regardless, it is possible to create unbiassed output from quantifiably biassed input. It is far easier to just produce output :( I doubt the sample size to detect a problem is all that large. Thousands or tens of thousands most likely.


Pauli
User avatar
Walter
Posts: 3070
Joined: Tue May 02, 2017 11:13 am
Location: On a mission close to DRS, Germany

Re: Naïve Analysis of TRNG

Post by Walter »

pauli wrote:
Fri Feb 14, 2020 9:21 am
Regardless, it is possible to create unbiassed output from quantifiably biassed input. It is far easier to just produce output :(
Is your first sentence here correct?? :?
WP43 SN00000, 34S, and 31S for obvious reasons; HP-35, 45, ..., 35S, 15CE, DM16L S/N# 00093, DM42β SN:00041
User avatar
pauli
Posts: 252
Joined: Tue May 02, 2017 10:11 am
Location: Australia

Re: Naïve Analysis of TRNG

Post by pauli »

Walter wrote:
Fri Feb 14, 2020 9:44 am
Is your first sentence here correct?? :?
Yes.

:D

Pauli
User avatar
pauli
Posts: 252
Joined: Tue May 02, 2017 10:11 am
Location: Australia

Re: Naïve Analysis of TRNG

Post by pauli »

You possibly want additional information :twisted:

Assume that a zero bit is produced with probability p and a one bit with probability 1-p. Now instead of considering individual bits, consider only the transitions from zero to one and one to zero and work out their probabilities.

This is assuming that each bit is independent. If sequential samples aren't independent, everything breaks down unfortunately.


Pauli
Post Reply