Notes from Allan Herrimen on his three page spreadsheet file, found here in
graph.zip.

He will be updating this material soon.

 - Robin Whittle 14 September 1999


The simulation consisted of an average of 1024 trials.  Each trial did a
524288 point FFT of 524288 pink noise samples.  A different random number
seed was used for each trial.

I used a raised cosine window in the FFT.

Page 1 is called "Bottom end".  It represents the bottom octaves of my
simulation.  I just collected the first 16384 points from the average of
the FFTs (after skipping the DC bin).  Frequency '1' is then Fs/2^19 =
0.09Hz for 48kHz Fs.

Page 2 is called "Top end".  It represents the full bandwidth of the
simulation.

I reduced the size of the output down to 16384 points by summing the power
of groups of 32 bins.

This makes it look like the sample rate is 16384.  You can see the image of
the noise start to rise above a frequency of 8192.

Page 3 is called "Lines".  It defines the guide lines on the other two pages.

BTW, by average, I mean I averaged the power, not the amplitude or
magnitude of the FFT bins.

The spreadsheet contains links to the data files.  It will probably
complain about not being able to find them, but this shouldn't matter, as
you probably don't want to massage the results further.
(I can provide the data files if you want them.)

The Y axis in the graphs doesn't really mean anything.  The important thing
is the slope.