.
this .
.At this late date (You bought or cheated your way to a degree, didn't you) you are still trying to save yourself by presenting the work of your betters as your own. It rarely works out.
.
this .
.At this late date (You bought or cheated your way to a degree, didn't you) you are still trying to save yourself by presenting the work of your betters as your own. It rarely works out.
. . g . r
rator
is-test.pdf)
CC_151-04255.pdf
They use the standard technique of attenuating the light from a laser, to get single photons. They actually make single photon generators.
Leon
(from
conformity
And the photo diodes, do they recover in the nsec range required? And the perfect output data, to remove bias / non linearities they must have used some algorithmic method, but they fail to mention that at all. Did they input enough entropy into the algorithm to justify their claims? You don't know, and I suspect no one does.
It smells of snake oil. They cite meaningless certificates as proof of quality, use a faulty testing methodology - you have to be able to "pass" the diehard tests but this alone is insufficient, plus according to them, their random data does not "fail" tests like a source of true random data would (at a 99% confidence level true random data would fail
1% of the time), and they make claims supported by factual errors (other random sources are deterministic???).Great stuff. I'll get one now for a door stop.
p .
is .
ing
e .
tor
nerator
antis-test.pdf)
e 6es/CC_151-04255.pdf
st
e.
ets
This is how they do the single photon detection:
I don't think there is much wrong with that technique.
Leon
generator
(from
conformity
Yep, that part looks good (seriously). Now if only there weren't those other questions.
generator
(from
conformity
It's just gross overkill, when a Johnson or zener noise source, with a downstream scrambler, is good enough.
John
Why don't you contact them? Their RNGs are very widely used; they should be told if they don't work properly.
Leon
chip .
this .
a ring
one .
ilator
generator
ode.
ers
/quantis-test.pdf)
oise
gle
- 16
files/CC_151-04255.pdf
test
drage.
here
nce
tests
r,
ve
d gil
er
For some jobs overkill is just the right amount. I don't begrudge those who worry about the keys to the kingdom a little extra on random numbers.
You left out other low cost noise sources such as the shot noise in a semiconductor or in some applications, just a radio signal.
There is a very old voice scrambler design that uses two radios tuned to the same radio station. Basically you mix the voice you want to hide with a signal that is known at both ends. Unlike the simple spectrum inversion method, a person listening can't learn to figure out what is being said.
I'm not about to tilt at wind mills. If they didn't care enough to make their product correctly why should I fix it for them? Do you think they even want to have their product modified - with all the costs that would involve? Would they thank me or pay me? Neither.
I should stress that I looked at this product as if it is a cryptographic product, which seems to be one market they are aiming for. In this case 1 output bit should equal 1 bit of entropy and anything else is either unacceptable or it should be fully specified so the user can make a judgement about its acceptability.
If, for example, they feed 256 bits, each with 0.4 of a bit of entropy into a hash function, (this is a common way of un-biasing and distilling entropy), a 128 bit output will have about 102 bits of entropy. Then, there are no practical statistical methods that will uncover this problem and for almost any simulation this could be used, trouble free, as "true random data". In saying this, I am assuming the "never fail statistical test property" is actually an invention of the advertising department and not a property of the device. If it is a property of the device, then it should never be used on a serious simulation.
In a cryptographic setting, 102 bits of entropy is too close to "not enough" to trust your serious secrets to and given the way the product is presented you have no way of telling what the true case of entropy per bit is. It might be fine, but you don't know.
This restriction would not apply to a big TLA (three letter agency) who would pull it apart and figure out which biases the thing has and assuming it is faulty, attack via the defects in the random number generator.
generator
(from
conformity
The above approach was used in a system called "SigSaly". There is a link:
ElectronDepot website is not affiliated with any of the manufacturers or service providers discussed here. All logos and trade names are the property of their respective owners.