I am experimenting with plotting 3D spectrograms. Using Scilab I FFT a
somewhat arbitrarily chosen number of windows. I could use Scilab's
plot3d1(), but it "chokes" sooner than Gnuplot due to the "shape" of my
data.
Currently a dataset may be approximately 250x600 points taking 3-4 MB in
an ASCII format file. Gnuplot takes 3-4 minutes to initially display. I
run under WinXP Pro on a 1.6 GHz Pentium with 1 GB RAM.
I'm looking for guidelines on the relative effects of making different
choices.
For example, I originally chose to use
set contour surface
unset surface
because it gave a less cluttered plot and some things were more easily
seen. But it _seemed_ to display faster than doing a more typical #D
surface plot.
I can see wanting to display up to 500x1000 points. What kind of trade
offs will I have to consider. Simply choosing to display every n'th
point in x or y is not attractive due to sampling artifacts.
My Gnuplot script is below.
Thank you for any comments.
set cntrparam linear
set grid
set cntrparam points 30
set cntrparam order 10
set contour surface
set style data lines
set ticslevel 0
unset key
unset surface
set cntrparam levels discrete # 30-60 levels given
set zrange [ 0.0009 :1] # occasionaly replot with set logscale z
splot 'C:spectrogram.dat' using 2:1:3
