“This title is misleading. It’s just a random number generator written in assembly. I thought this was going to be something cool creating a number generator from the bluetooth frequencies, or data “sniffed” over bluetooth or something.”

Exactly what I thought as well.

MS3FGX, I’m interested to know more about these “true RNG” projects, and what your definition of that is. I would think that true RNG is impossible as long as the data is generated algorithmically. To extend that further, even very random naturally occurring data can’t be truly random. It’s also generated algorithmically according to the laws of physics.

@MS3FGX. The biases in the distribution of your random digits is “self-imposed”.

You are printing out an integer (0-65535) and then analyzing the individual digits.

If you were printing 00000 – 99999 each time then you would see an even distribution. But since the numbers you print only go up to 65535 and you don’t print leading zeros you would expect to see a much smaller number of 0, 7, 8 and 9 digits. The expected probability of getting a number in the 60000-65536 range is going to be about half that of the range 50000-59999, so the expected probability of seeing a 6 will be midway between the 1-5 and 0,7-9 probabilities.
(This is precisely what you see in your results)

If you took the 16 bit value and printed it as 4 hex digits and then did a statistical analysis on 0-F you’d probably get pretty good results.

The biases in your Interpreted Binary mode are also caused by this skewed distribution.

If you still wanted to use decimal, then you should truncate the numbers from 0000 – 9999 (and print leading zeros!). If the number you get back from the device is >60000 throw it away, otherwise reduce it modulo 10000. Do not be tempted to just do modulo 10000 and use that number — you’ll just introduce bias back into your results… see the following as to why,

0 – 9999 -> 0000-9999 equally distributed
10000 – 19999 -> 0000-9999 equally distributed
20000 – 29999 -> 0000-9999 equally distributed
30000 – 39999 -> 0000-9999 equally distributed
40000 – 49999 -> 0000-9999 equally distributed
50000 – 59999 -> 0000-9999 equally distributed
60000 – 65535 -> 0000-5535 un-evenly distributed

Also check out the analysis done on LavaRND

http://www.lavarnd.com/what/nist-test.html

(disclaimer – I am a co-Inventor of LavaRND)

Another truly useless project which just proves how bored people can get. Want random numbers? use GetTickCount() …

@ds :-)

As compared to the probabilities of hex digits…

$ perl -e ‘for ($i=0;$i<65536;$i++) { $a=sprintf "%04x", $i; for $j (split (//,$a)) { $x{$j}++ }} for $i (sort keys %x) { print "$i $x{$i}\n" }'

And the evenly distributed way of using digits from a decimal expansion….

$ perl -e 'for ($i=0;$i 59999); $a=sprintf “%04d”, ($i%10000); for $j (split (//,$a)) { $x{$j}++ }} for $i (sort keys %x) { print “$i $x{$i}\n” }’

wordpress == fail. It cannot handle code posted into comments!

The processing of < and > is borked!

#!/usr/bin/perl
for ($i=0;$i<65536;$i++) {
next if ($i > 59999);
$a=sprintf “%04d”, ($i%10000);
for $j (split (//,$a)) { $x{$j}++ }
}
for $i (sort keys %x) { print “$i $x{$i}\n” }’