Quantum Random Numbers 167
tqft writes "What the world needs is more truly random sources of numbers. Researchers from Australian National University have found a brilliant way to make one: 'We do this by splitting a beam of light into two beams and then measuring the power in each beam. Because light is quantised, the light intensity in each beam fluctuates about the mean. Those fluctuations, due ultimately to the quantum vacuum, can be converted into a source of random numbers. Every number is randomly generated in real time and cannot be predicted beforehand.' So if you need some really random numbers, just use their generator service."
Re:Quantum (Score:3, Informative)
In this case, it doesn't matter. Random number generation is in itself interesting and a very important part of computer science... or better said, the problem that with a finite state machine, like a computer, we cannot generate truly random numbers. Computers can generate pseudorandom numbers, but they are only random within the constraints set, are repeatable and will have a periodicity. Getting "true" random numbers only is possible from physical processes.
There have been several articles about random number generation on slashdot. On the top of my head, random numbers generated from a lava lamp, or from a CCD with a (disassembled?) smoke detector (which contains Americum, a radioactive element). Let's just say that random numbers are interesting unto themselves. That's they're generated by quantum fluctuations is just an added bonus.
There is no need for this new method at all (Score:5, Informative)
Zener-noise at 5V6 or NPN transistor EB noise is already about half quantum effect noise. Just use that, plenty of recipes on the web. Cost is at a few USD/EUR for the raw generator and you can get it as an USB stick.
Examples:
http://www.cryogenius.com/hardware/isarng/ [cryogenius.com]
http://www.tonbandstimmen.de/evpmaker/random-bit-generator/index_e.htm [tonbandstimmen.de]
http://www.entropykey.co.uk/ [entropykey.co.uk]
Seems to me the quantum folks are getting a bit desperate to prove they actually are doing something worthwhile.
Re:A cheap non-quantum option (Score:5, Informative)
How "very random" do you need, that a Mersenne Twister is not good enough, but an untested signal is? And if you only looked at the 1 bit word mean, you basically didn't test it at all. If you don't care about specifics, you can just use the Testu01 [umontreal.ca] suite of tests. If you care about specifics and want to do your own tests...you should do the same bias test for various word lengths. So besides making sure the 0 and 1 appear about 50% of the time, you should make sure that 00,01,10,11 appear about 25% of the time, 000,001,etc are all uniform for 3 bit words, and so on until at least several hundred bit words (yes, it will take you a long time to sample enough words!) Besides those simple "fair dice" tests, you can also pick a dimension, (2 or more) and sample a lot of points (using whatever bit words you like) and then find the minimum distance between two points. With repeated tests this minumum distance should follow an exponential distribution, with the parameters describing that distribution depending on the range of possible values and the number of dimensions. Similarly, if you treat your n bit words as floats from [0,1) (by dividing by 2^n), then sampling a large number and adding them together should get you a normal variable, where the mu and sigma depend on n and how many you are adding together. There's also a 'pigeon hole' test where you have 2^n bins and sample n bit words and put them in the bin, and then take the maximum occupancy. Again, statistically this value should have a known distribution and you can test against that. A related one is called a "parking lot" test, where you sample in 2 dimensions and treat points as circles of some radius, and you only place a circle if it doesn't overlap with other circles already placed. After 10,000 samples you should have less than 10,000 placed, and the number you actually placed should be normally distributed, with the mu and sigma depending on the size of the lot, the number of attempts, and the radius of the circles used. There are plenty of such tests. They all revolve around using your generator to simulate a simple statistical system where the distribution of the result is known by you, so you can see how consistent your repeated simulations are with the expected distribution. A 1-bit mean test is just simulating coin tosses, while a 2-bit mean test is simulating 4 sided die tosses, etc.
But a Mersenne Twister does quite well on Testu01, so you have to ask yourself why a software generator can't possibly be sufficient.
As for the mechanical "what's wrong" side of things, if you don't know the physical mechanism, you can't know what conditions are required for your tests to remain valid. Where is the line noise coming from? Just noise from the physical components, or is there a radio wave factor? Will a new tower broadcasting alter the behavior of the noise in a meaningful way? What about wifi devices? I know of sound cards that pick up the signal sent when a cellphone rings so you can tell moments before it actually rings. (You can get a TARDIS for your keychain that lights up when a nearby cellphone is being rung, which works by the same mechanism but on purpose).
Re:dead link (Score:3, Informative)