[LAU] turning a consumer soundcard into "prosumer" w/ quasi-balanced outs

Niels Mayer nielsmayer at gmail.com
Thu Jun 10 01:05:35 UTC 2010


On Wed, Jun 9, 2010 at 3:05 PM,  <fons at kokkinizita.net> wrote:
>> 127 causes clipping.
>
> Turning down the source by a dB or two should cure that.

Thus my empirically derived value of 126, which is nice to know is a
tiny bit below unity gain.

Also, FYI, the gain issue may be ultimately what's causing the
difference between soundcards. The Dynex dx-sc51 was actually easily
overloaded by the standard amplitude output of many of the XG-midi
files available on the net, which means  in order to hear the db50xg
without distortion, I had to turn down the master volume on the synth
-- there is no other gain reduction available on the input. However,
the Terratec DMX6fire24/96 doesn't have this problem, plus it does
allow for gain reduction. Therefore, with the terratec, bigger voltage
swings are happening. The S/N ratio is improved because the signal is
bigger.

In other words If I really want to fairly A/B test the two cards, I
need to set the volume output by the db50xg to the exact same low
level as needed for the dynex. and them boost the gain on the input of
the soundcard to compensate back to 0db full scale swing.  After all,
a full-swing signal would be testing the slew-rate limitation of the
synth's own output opamps and caps, in series with the slew-rate
limits on the card's input opamps.

And also, I truly think burn-in and not my ears adjusting is the
reason why this card and synth now "thump" like it's supposed to,
after playing full-volume through the soundcard for 20+ hours nonstop.
And you can't just eq away a lack of thump (i tried): just makes the
bass louder and more "muddy". (although you could go all "aural
exciter" ( http://en.wikipedia.org/wiki/Exciter_(effect)#Aural_Exciter
) and try delaying the high-frequencies to match the phase-delayed
bass)

>> How old is too old for a decoupling electrolytic??
>
> Impossible to tell, there's too much spread in quality.
> Some are defective from the start :-)

I think this would be an excellent area to investigate with your
tools. Almost a better use of the scientific method than whole
soundcard testing because you can focus on a single variable that is
known to be problematic.

And unfortunately, electrolytic capacitors are the one component that
are about as far from their theoretical ideal of any component.  In
fact, i think it's totally braindamaged to be using electrolytic
capacitor for audio coupling... but then again, a non-electrolytic
coupling capacitor would be huge, and would also probably have a ton
of self-inductance given the amount of conductor needed to make a
large value capacitor using a standard dielectric. IMHO, the next
advance in circuits would be totally DC-coupled designs, with built-in
microcontrollers to auto-trim away any DC offset....  they should just
integrate the whole thing into a new "super op amp" containing a
digital control "watchdog" that automatically adjusts its DC offset,
output impedance, output damping factor, etc...

> Well, if a circuit is linear (i.e. it doesn't produce distortion),
> and time-invariant (it doesn't change its response over time,
> except maybe *very* slowly), then either the impulse response
> or the *complex* (including phase) frequency response will tell
> you all there is to know. If you dispute this you have to talk
> to the mathematians :-)

I don't dispute this, however, can meaningful impulse responses be
made of an entire  soundcard from analog
input->A/D->DSP/Mixing->D/A->analog output? It's giving me a headache
thinking about even a simplified case -- an analog sample-and-hold
circuit. Isn't "sampling" anti-causal?

> A 'magnitude only' frequency response, which is what you usually
> get and also what JAAA provides, throws away half of the available
> information. OTOH, in 'simple' circuits the phase response will be
> related to the magnitude response.

Yes, such circuits are easy to understand and model:
http://en.wikipedia.org/wiki/Low-pass_filter#Passive_electronic_realization

IMHO, if done correctly, the significant phase and LF rolloff will
happen entirely outside of the audible spectrum. So it's really a
matter of calculating the input resistance, and coming up with a
capacitor value large enough  that the rolloff is below 5hz or so ....

Which is probably how http://www.jensen-transformers.com/an/an003.pdf came up
with the 220μF value for the coupling capacitors, versus 47μF typically used.

-- Niels.
http://nielsmayer.com


More information about the Linux-audio-user mailing list