Subject: Re: [linux-audio-dev] Linux support for IEEE1394 mLAN?
From: Paul Winkler (slink_AT_ulster.net)
Date: Fri Sep 22 2000 - 07:33:05 EEST
Tom Pincince wrote:
> Regarding the float a/d, I originally thought of it only
> in terms of potentially superior sound quality by making
> the input passive, providing impedance matching only and
> eliminating the need for mic preamps,
That's a fascinating idea. But of course many people in pro
studios would keep using their Neve and Manley and JoeMeek
pres because of their distinctive "character"...
> and moving the
> amplification to the digital side of the s/h where the
> signal is dc and not reactive. I later realized that if I
> amplified the signal by a factor of 2 and did this
> multiple times until the signal became > = a reference
> voltage, that by counting the number of times the signal
> was amplified I would have the exponent of a normalized
> signal. This would give all signals, large and small, 24
> bits of precision.
Sorry if I'm dense; I understand the
counting-up-to-the-exponent idea, but how do you then
measure the mantissa?
I can think of one way, but it seems very problematic. This
is all just off the top of my head, mind you; I'd love to be
enlightened. The way that occurs to me is to just take the
difference between the amplified signal and the reference
voltage, and feed it to a standard analog-to-digital
converter. For each sample you'd need to switch this ADC's
input to the output of the correct gain stage. You'll be
getting all the analog noise and distortion introduced by
however many amplification stages it took to get the signal
above the reference level. If as you suggest we have a chain
of amplifiers, each with gain of 2, then the lowest-level
input signals will get correspondingly more noise and
distortion because they'll go through lots of gain stages.
I can think of a way to measure the exponent without
amplifying the input (by feeding it to a bunch of gates in
parallel; for gate X, calibrate it to turn on when the input
reaches 2 ** X ... ) BUT this is in fact totally useless,
because then you're still left with a non-normalized
mantissa to measure.
So we're back to keeping the reference voltage constant and
amplifying the input signal; which means that for very low
signal levels, we're still dealing with possible noise and
distortion from an extremely high-gain amplifier.
But maybe that's not so bad. Put the amplifiers in parallel,
not series; calibrate their gains to 1, 2, 4, 8 ... Now we
can tailor each amp to the job at hand. At the low end of
the input signal range, we need the highest-gain amplifier,
so we use the highest-quality, lowest-noise one available;
but don't worry so much about its clipping characteristics
because anytime it clips, we'll be ignoring its output in
favor of the output of a lower-gain amplifier. At the top
end of the signal range, noise floor would be relatively
less important, but distorting more gracefully may be more
significant. The difficulty would be, again, calibrating the
gain of each amplifier precisely enough.
I think this is extremely important. If I'm doing my math
right, to get true 24-bit accuracy you need to calibrate
each amplifier to within +/- .0000000298 of its nominal
value. I calculated that by 1 / (2 ^ 24) / 2
If that formula is correct (and it might well be wrong - I
just made it up!), then just to get 16-bit accuracy would
require amplifiers calibrated to within 1 / (2 ^ 16) / 2 =
.000007629
Also, the amps would need to have very flat frequency
response - you wouldn't want to hear the tone quality change
as the signal level changed!!
> I remembered discussions here that
> referred to the desirability of float for dsp. If the a/d
> and d/a can be done in float then the entire digital
> domain can be float, if all one would wish to do digitally
> can be accomplished in float. Is this desirable? Please
> reply because your answers will weigh greatly in my
> decision to pursue this or not.
I'm intrigued by the idea, even if I'm not knowledgeable
enough to know whether it really makes any practical sense
or not. :)
--
................. paul winkler ..................
slinkP arts: music, sound, illustration, design, etc.
web page: http://www.slinkp.com
A member of ARMS: http://www.reacharms.com
This archive was generated by hypermail 2b28 : Fri Sep 22 2000 - 08:25:05 EEST