Best source of thermal noise

[Raven Luni]

Greetings,

I'm not 100% sure why I'm asking this but my brain thinks I should know the best / easiest way to ensure an even distribution of random bits that doesnt involve a predictable algorithm.

So I'm thinking radio noise, but what to do with it......
How about taking a band that isnt really susceptible to local interference (things like mains hum, computer noise etc. could lead to predictability), pass it to a comparator, XOR it with a square wave and sample the data at 2x the square wave's frequency. Am I right in assuming this would result in a fairly even spread of 1s and 0s?

 

[(*steve*)]

Yes, noise can be a good source. You want the "frequency" of this noise to be much greater than the rate at which you sample it.

Essentially you sample it and see if it is above or below some threshold (passing it through a schmitt trigger can be helpful) and use that as the bit value.

You want to be sure that the signal is sufficiently symmetrical that the chance of getting a 0 is close to the same as getting a 1.

If you want an 8 bit number you ask it 8 times for a bit (with a delay between "asks"). If the probability of 1 and 0 are different, some numbers will be more common than others.

 

[john monks]

I believe the best source of random noise is a reversed bias base emitter junction adjust to about 10 microamps. From there you simply amplify the signal, run it into a logic gate, or anything else you want.

 

[CocaCola]

The thing to remember is once you remove predictability you for all purposes have random... 111111101111 is just as random as 010000000000 or even 010101001011 as long as the outcome is not predictable... Even though our minds suggest that 010101001011 is more random it's simply not in most cases as it's an even distribution of 0 and 1 and that in itself is not random it's generally forced order

This is important when you set a threshold to distinguish 1 and 0, at that point you are establishing order when setting that threshold, a pesky stuck between a rock and a hard place...

 

[(*steve*)]

Whilst you can't prove that 0, 0, 0, 0, 0 is not a random sequence of digits, if your source of noise varies slowly then you may be in the situation where your possible random numbers are:

00000000
00000001
00000011
00000111
00001111
00011111
00111111
01111111
11111111
11111110
11111100
11111000
11110000
11100000
11000000
10000000
If numbers like 11011101 don't show up (or do so with extremely low frequency) then whilst your source may be random, it has a very poor distribution.

A reverse biased BE junction, as mentioned above, or a zener diode. Having a high pass filter before your amplifier and detector will minimise issues arising from low frequencies (such as mains hum) that could introduce bias as I have indicated above.

 

[Harald Kapp]

I'm sure you've read about random number generators in general.

Here's Intel's approach used in the Ivy Bridge CPUs.
general
detail