SoylentNews is people
SoylentNews
from the just-pull-numbers-from-a-hat dept.
Researchers in China have developed a way to improve the reliability and security of machines that use quantum phenomena to generate random numbers. This is crucial to the development of other related technologies, such as secure quantum communication and computer simulations used in weather forecasts.
[...] "The output of [...] pseudorandom number generators is in principle predictable," said Xiongfeng Ma, an information scientist from Beijing's Tsinghua University, who was a part of the Chinese group. "They are good enough for most applications like simulations, but not for high security crypto systems."
[...] "Even if you have a very good [quantum] random number generator, there will still be some residual bias, so there needs to be a way to test and clean the data," said Juan Carlos Garcia-Escartin, a telecommunication scientist from University of Valladolid in Spain.
This need for post-measurement processing exposes the system to potential hacking. Ma and his team have developed a way to detect if a system is compromised. The basic concept is pretty simple -- they use the random source to trigger random testing of the data, kind of like pop-quizzes for a class of students.
This involves repeatedly shuffling and dividing the output numbers into four random groups, then testing them and crosschecking their results for anomalies. If the numbers are truly random and unbiased, any manipulation by an outsider would show up in these tests. Once this testing method is implemented, then even an untrusted quantum random number generator can still be used without the fear of compromising the level of randomness generated.
(Score: 2) by FatPhil on Tuesday February 14 2017, @06:53PM
You can "recycle" the 00 and 11 pairs as input for a second von neumann unbiaser. There are diminishing returns, as 2p(1-p) is lower in the subsequent blocks, and also as you need two failed outputs from the previous unbiaser. However, if you have something like a p=0.6 input you'll increase your yield from 0.24 bits per bit to 0.30. I've always suspected that with a tiny amount of additional smarts and no extra debiasers you can squeeze even more entropy out of the stream, but I've never set about to prove it. Perhaps I should...
A single unbiaser will give a maximum entropy rate of 0.25 (at p=0.5+eps).
The method mentioned on the wiki page takes that to 0.28125. My chaining setup peaks at 0.34375, but I'm sure it can be pushed to 0.40625.
Great minds discuss ideas; average minds discuss events; small minds discuss people; the smallest discuss themselves