[sdiy] Fw: New waveform generator currency market

[Craig Critchley]

I have this feeling I'm taking this too seriously, but... 

These data sets are examples of random systems that nonetheless have
long-term autocorrelation and self-similarity. They are fractal, in other
words. The classic midpoint subdivision algorithm for generating brownian
paths or surfaces is one example of an algorithm that produces data with
these properties.

Another method is the fractional ARIMA process, which is basically white
noise filtered with an IIR filter that has (in principle) an infinite number
of terms... The transfer function has the form 1/(1-B)^d, where B is the
"one sample delay" operator, and d is some real number parameter. Because of
the fractional exponent, the denominator expressed as a power series has an
infinite number of terms - that's what causes the long-term memory that gets
you the self-similarity. Of course, if you were to implement this on a DSP
(who's got a Chameleon?) you'd approximate it with a finite number of
terms...

There's other ways to generate this kind of noise using FFTs or wavelets or
whatever... I'm not an expert and only fooled around with non-realtime
fARIMA filters.

These sorts of chaotic systems show up all over the place - economics,
politics, biology... I encountered them when learning how to simulate
network traffic, where packet size and arrival intervals are random, but
self-similar.

Maybe you could just sample a large network...

				...Craig

-----Original Message-----
From: owner-synth-diy at dropmix.xs4all.nl
[mailto:owner-synth-diy at dropmix.xs4all.nl] On Behalf Of MED
Sent: Monday, April 12, 2004 10:08 PM
To: Jay; synth-diy at dropmix.xs4all.nl
Subject: Re: [sdiy] Fw: New waveform generator currency market

On Monday 12 April 2004 23:05, you wrote:
> A presidential election poll tracking module would be great as a 
> random noise source.

yes, but the problem with modules like these, is even when you have the
data, it is too difficult to use it!
Unless you make music that spans hours or in some cases, days, the data
doesn't change enough and certainly isn't updated enough, unless you were to
sample it and then store it for later use in a piece.
And if you're doing that, why not just make it all up to begin with? :)

-MED