To be or not to be... randomly speaking

[ericfarmer]

Originally Posted by iCountNTrack

Quatum randomness based on quantum events are the most random occurrences known to date, which means nobody has been able to discern any patterns so far.

I tried something like this many years ago as something of a science project. I yanked the chunk of Americium out of a First Alert brand smoke detector, and placed it next to a radiation monitor (http://www.aw-el.com/) that was linked to the serial port on my PC. Each detection caused a modem status register (MSR) interrupt, at which point I used the time stamp counter (rdtsc instruction) to record the time of the detection.

The idea is that the Americium decays, emitting alpha particles as a Poisson process, meaning that the random times between decay events are independent and identically (*) distributed. If two consecutive intervals between successive decay events are measured, then the first interval is equally likely to be longer or shorter than the second... yielding one random bit (e.g., 1 if the first interval is longer, 0 if it is shorter). Slow but effective.

(*) But the pairs of intervals are not quite identically distributed. As the Americium decays, the Poisson rate (\lambda) very slowly decreases, which causes a slight bias toward 0 bits in the above approach. A simple band-aid for virtually eliminating this first-order bias is to simply reverse the sense of the comparison for every other bit.

Finally, this is nothing new. I think the HotBits site (a web source of random numbers, which strikes me as a spectacularly bad idea) used this approach, and a quick search yields a page describing almost exactly the same setup here: http://www.etoan.com/random-number-g...ion/index.html.

[aslan]

Originally Posted by ericfarmer

I tried something like this many years ago as something of a science project. I yanked the chunk of Americium out of a First Alert brand smoke detector, and placed it next to a radiation monitor (http://www.aw-el.com/) that was linked to the serial port on my PC. Each detection caused a modem status register (MSR) interrupt, at which point I used the time stamp counter (rdtsc instruction) to record the time of the detection.

The idea is that the Americium decays, emitting alpha particles as a Poisson process, meaning that the random times between decay events are independent and identically (*) distributed. If two consecutive intervals between successive decay events are measured, then the first interval is equally likely to be longer or shorter than the second... yielding one random bit (e.g., 1 if the first interval is longer, 0 if it is shorter). Slow but effective.

(*) But the pairs of intervals are not quite identically distributed. As the Americium decays, the Poisson rate (\lambda) very slowly decreases, which causes a slight bias toward 0 bits in the above approach. A simple band-aid for virtually eliminating this first-order bias is to simply reverse the sense of the comparison for every other bit.

Finally, this is nothing new. I think the HotBits site (a web source of random numbers, which strikes me as a spectacularly bad idea) used this approach, and a quick search yields a page describing almost exactly the same setup here: http://www.etoan.com/random-number-g...ion/index.html.

The Dilbert comic on your link was so funny, I couldn't resist putting it here for all who don't follow the link to see.