12 May
2010
12 May
'10
6:55 a.m.
Mark Knecht wrote:
I think also that an FFT assumes some sort of
longer-term steady state
Its actually worse than that. The FFT assumes that the input is steady
state periodic with a period length equal to the FFT length.
repetitive nature which isn't in the stock data.
The good news is that your stock data isn't much worse of a fit to
the assumptions of the FFT than real audio data :-).
I could take a
representative sample - the last 1000 bars, and then treat that as if
it was a repeating cycle and extract frequencies.
Then you loose time locality. In music, its not uncommon for the
chord to change once every bar. By doing the FFT over 100 bars
you are effectively 'averaging' the chord over those 100 bars.
Erik
--
----------------------------------------------------------------------
Erik de Castro Lopo
http://www.mega-nerd.com/