Using White Noise to Choose Between Red Noise and Blue Noise

There is source code that goes with this post, which generated all the images and did all the tests. You can find it at https://github.com/Atrix256/DFTRandomFibonacci

I recently saw a really cool video from @Numberphile which mixed some of my very favorite things: Fibonacci numbers (aka the golden ratio), red noise and blue noise.

The link to the video is below and is very much worth watching.
“Random Fibonacci Numbers”

It had me thinking though… they are using a coin flip (white noise) to determine if they should add the last two numbers (red noise / low pass filter) or subtract them (blue noise / high pass filter).

I was curious what the DFT of that would look like, which would show what sort of frequency content that number sequence had.

BTW I have a post talking about dice, probability distributions and noise color here that is a bit related: https://blog.demofox.org/2019/07/30/dice-distributions-noise-colors/

White Noise

Just to prime things, here is 100 uniform white noise values on the number line and the DFT of those points. The points clump together and leave holes, and the frequency spectrum shows all frequencies.

Here is the average DFT of 100,000 such point sets. The average flattens out and the grey lines showing 1 standard deviation are flat as well.

Regular Fibonacci Numbers

The first 90 Fibonacci numbers look like the below on the number line:

And here’s their DFT, where 0hz (DC) is in the middle:

Nothing super interesting really.

Randomized Fibonacci Numbers

Here is 90 randomized Fibonacci numbers on the numberline, the dft, and the average DFT of 100,000 such number sets.

It’s interesting to see that the individual randomized Fibonacci has strong low frequency content, but other frequencies too, while the average DFT of the number sets shows only low frequency content.

I think what’s going on here is that since the numbers start out small and grow larger over time, that they will always start out clumped together (red noise), but then depending on the coin flip (white noise) which have different frequency content in each set of numbers, as the numbers grow larger. This means that the only thing common among them is low frequency content, but the rest is just white noise and averages out to be flat.

Maybe not that interesting of a result, but it’s an answer at least 😛

Prime Numbers

I got a tweet from Tariq wondering what DFTing the prime numbers would look like.

Here are the first 25 on the numberline, and the DFT:

Here are the first 100:

Here are the first 200:

Here are the first 1000:

I don’t really see much of a pattern, but I guess if someone did, they would have gotten some prize by now? 🙂

“The next coin flip has to be tails!”

If you saw a fair coin flipped and 10 heads came up in a row, you’d probably either think that the coin was a 2 headed coin, or that the next flip HAD to be tails.

In the code that goes with this post (https://github.com/Atrix256/DFTRandomFibonacci check out DoCoinTossTest()), I have a random number generator generate bits until there are 10 ones in a row, and then count how many times the next random bit is a one again.

I do that test 10000 times, and ran that overall test 5 times. Here are the results!

At an infinite number of coin flips, you can expect an even numbers of heads and tails, but until you reach infinity, all bets are off. When a coin has come up heads 10 times in a row, it still has a 50/50 chance of heads in the next coin flip.

That’s the definition of white noise – independent, random events.

Red noise would tend to have similar values – so, heads would clump together, and tails would clump together. This makes more sense with dice, where similar values would be rolled.

Blue noise would tend to have dissimilar values – so heads and tails would specifically NOT clump. And with dice, you’d rarely get the same, or similar values, in 2 dice rolls.

White noise doesn’t care about what came before, and just gives you a new number based on the probability distribution.

Keep this in mind if you ever play roulette!


2 comments

  1. interesting. thinking of roulette- how about an actual red noise or blue noise roulette wheel? i imagine that the betting strategy would be a little different. would the outcomes be more predictable?

    Like

  2. Yeah it would actually be more predictable, but yeah could still make a gambling setup based on it!
    Check out the “noise colors” section here if you want details. It’s come up before funny enough! https://blog.demofox.org/2019/12/01/calculating-information-entropy/

    Another neat test to look at might be using red noise to choose between blue and white, or blue noise to choose between red and white.

    Also now I want to make a flag where red white and blue sections are replaced by sample points following those noise colors hehe.

    Like


Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s