[sdiy] Re: GIC String Filter - first experiments

[jhaible at debitel.net]

Hi Ian, and list,

> Congratulations!  That's a big project.  I think I have some idea of what
> it means to "select" the caps.  :-)

I have calculated the C values in Excel, and then searched for combinations of
two capacitors to get the right value by <1%. The idea was to make advantage of 
the
distribution of low tolerance components, and to find some that fit the very 
odd values.
Example: Need 34.6nF, find a nominal 33nF that is actually 33.1nF and parallel 
it with
1.5nF. Or find a 33.4nF and parallel it with 1.2nF. The idea was that there are
several possible combinations for each of these odd values, so one should find
at least a couple of combinations that fit.
Unexpected obstacles: When you need a 995pF cap and you find that the cheap
1nF 20%  that you've bought _all_ have _precisely_   1.01nF. 30 in a row, and
loose (not on a belt). /&%/(&%(%& !!

In the end I had 5 DIN A 4 paper sheets with colorful capacitors taped on it 
and labelled.
It looked like a friend's butterfly collection. (Selected the caps in the week 
before the
PCBs arrived.)


> The demos sound great, although I don't hear the highest freqs any more.

Ahh! Maybe I should add a little of the original (dry) signal before I go thru 
the
filter bank. I did the pre-filtering with 3 BPFs and a Notch filter, not with
a parametriq EQ. So the frequencies outside the Bridge hump may be attenuated
too much. 


> Did you use the usual Burhans frequency ratio?

I think so. I got the spacing and lowest f straight out of Electronotes, then
did the rest in Excel.


>  What is the maximum
> peak-to-valley ratio on the second demo?  It sounds like you are getting
up
> to about the 8-10 range.

I haven't measured it. But I noticed that the high frequencies are ringing more 
than the
lower ones. (I think you wrote about that, too.) The reason is that between the 
bands
there is some phase cancellation, but the highest (and lowest) bands don't have
enough "neigbours" to get the exact same height as the middle ones. Feedback
will make this slight unevenness even worse. I was aware of this before I 
started,
but I thought a little HF rolloff in the feeback path would cure it. But that 
doesn't work,
as that LPF also introduces phase shift. I might try a combination of low pass
and all pass filters - maybe.



> >I think the peak/valley ratio of the main filter bank is ok,
> >but the air/wood/bridge resonance needs some more
> >adjustments. Any hints how to improove this are welcome.
> >(Benade gives detailed information about the frequency -
> >125Hz, 175Hz, 1.5kHz and 2kHz for Cello -, but not
> >about Q and Gain.)
>
> There have been many measurements of violin resonances.  A fairly detailed
> summary can be found in the Fletcher and Rossing book (The Physics of
> Musical Instruments).  Concerning violas and cellos they say "...one must
> search diligently to find 1 or 2% as much published material on their
> acoustical behavior."
>
> For cello, they list the lowest modes as 57, 102, 144, 170, 195, 203, 219,
> 277, 302 (Hz).  The lowest mode is the C1 bending mode and is probably
> quite weak.  102 Hz is the A0 (air) mode and 144 Hz is the T1 mode.  The
C3
> and C4 modes are at 219 and 195.  


Oops, that's quite different from what Benade shows!
( http://www.zainea.com/violinfamily.htm )

If you look at the Benade courve ( http://www.zainea.com/v243.gif - shown for 
violin)
you might also think that the "ripple" filter is only needed in the higher 
frequency range.
I _thought_this was reasonable: A distinct lowest Woond and Air resonance
mode for low frequencies, and higher modes that become a dense cluster of peaks
and valleys in th ehigher range. So my first attempt was to just run the Bridge 
resonance
part (broad 2kHz BPF) thru the resonator bank, and add the 125Hz and 175Hz 
peaks without further processing. But this was far worse than running the whole
thing thru the filterbamk.


> They don't actually show any admittance
> curves for cello, unfortunately.  For violin, the A0, T1, C3 and C4 modes
> are the strongest low-frequency resonances, and look quite a bit stronger
> than the higher frequency modes.  So you might try frequencies of 102, 144
> and ~200+.  Also, violin has an overall -9 db/Oct output above about 700
> Hz, so you might want to try prefiltering the input sawtooth a bit.  Also,
> experiment with a pulse drive with different widths.
>
> Please keep us posted on this very interesting work!


Thanks for all your valuable input - I will try some of these things later.
At the moment, the whole experiment is dismantled, as I'm building a 19"
enclosure for this and must see how I fit it all inside. I plan to make
the viola and violin filters at least, blus maybe the great 3-band string 
equalizer from the Crumar Performer.

Best Regards,

JH.


-------------------------------------------------
debitel.net Webmail