The Random Page

2023-05-27 | [radiclock]
next: Atomic Clock

lol so random

As part of my atomic clock project I also produce random numbers (why not). Radioactive decay is a well known method to generate true randomness because no one can predict which or when an atom will decay. This technique was used for years in the famous HotBits project which is sadly shut down as of 2023. I have a radioactive sample of my own, but, there's a ways to go from a Geiger counter to a number. How does this work?

  1. Radiation from my sample of pitchblende gets detected by one of my radiation sensors. These are the "clicks" you hear from a geiger counter.
  2. There is a timer running all the time. When a pulse from the sensor comes in, take the timestamp of it. If you want to slow down the rate, you can take the time every nth pulse instead. Whatever the setup, do this until you have 4 timestamps.
  3. Do a quick check to make sure the timestamps are in order (in case the timer overflowed)
  4. Take the 4 timestamps, and generate 2 intervals:
  7. If interval0 > interval1, that's a 0. If interval1 > interval0, that's a 1. If the 2 are equal, throw out the sample and try again.

I actually have 2 sensors; one is a classic Geiger–Müller tube, and the other is PIN-diode based. I collect random numbers from each. I didn't know quite what to do with the numbers I collected, so I figured I would publish them live.

The 2 detectors are set up to generate random bits at around the same rate - about 32 bytes every 55 seconds. The full 1024 byte frames shown and linked below will update about once every half hour. Therefore, please do not spam us with requests. Also, be sure to check the ISO 8601 timestamp at the start of each file, and make sure it's somewhat close to the current time. If it isn't, the project is either broken, or you are getting a cached copy (check your browser).

Also, a most important note: the data listed here is published globally, not individually, like HotBits or random.org. In other words, it is not unique to you. I highly recommend you do not use this for security purposes as it it possible another party may have a copy of the data.

From the GM tube (hex format):

From the PIN diode (hex format):

Ok, so you want decimal digits instead? I've come up with this technique:

  1. Look at each byte. It's a number from 0-255. Throw out any samples greater than 199, and we now have a range from 0-199
  2. Convert those samples to a 3 digit string, then throw away the first digit. Now, we have random numbers from 0-99
  3. Concat those strings into a really long string. Now, we have a sequence of random digits. But, the string can vary in length.
  4. To deliver something with consistent length, I can just trim the string. But by how much? Since the probability of a skipped byte is 56/256, we would expect about 800 of 1024 bytes to be usable samples. There's a 50% chance there will be more than 800 good bytes. But, there's about a 99.99% chance there will more than 750 bytes. I still double check to make sure, but that's what I publish. Of course, that gives 1500 digits.

To determine where to make the cut, look up the binomial distribution probability. It's already built into spreadsheets as BINOM.DIST, which is what I used.

From the GM tube (digits):

=2){$lines[1] = wordwrap($lines[1], 50, "\n", true);}echo implode("\n", $lines);?>

From the PIN diode (digits):

=2){$lines[1] = wordwrap($lines[1], 50, "\n", true);}echo implode("\n", $lines);?>

Or, perhaps you want just the raw bytes themselves. Just make sure to trim away the first 33 bytes which are the timestamp header:

Here is the python program I use to generate the reports above.

comments







comments rss link

Alnwlsn 2024