Re: Looking for schematic to build a -10db to +4dbu converter [Archive]

Mike Rivers

August 6th 04, 01:32 PM

In article > writes:

> On this point I don't agree. If the input level required for 0dB FS is
> fixed, you don't need *ANY* headroom. You simply require a clean analog
> signal up to the 0dB FS point. Any more is simply wasted.

That would be valid if practice agreed with theory, but it doesn't.
Most analog circuits approach "not clean" slowly. In order to be
really clean at the level that produces 0 dBFS, you need to be able to
go above that level by some amount (depending on how much distortion
you're willing to accept at high levels) before reaching clipping.
Clipping isn't the only "distortion" you need to worry about.

Also, subjectively, audio that's clipped for just a couple of samples
is undetectable, so you want to be able to send clean level to the
converter even at its clipping level.

> If it makes you
> feel better, fine. Metering anywhere in a fixed gain system will tell you
> exactly the same thing, once you have established the lowest overload point
> and don't exceed it. I wouldn't want the analog stage clipping first, but
> extra headroom is just a feel good factor with digital.

The metering tells you what's happening right now (or what just
happened now that it's too late to do anything about it). Again,
you're arguing theory and I'm talking practice. If everyone sang level
sine waves, your system would be OK. But meters differ in how they
deal with complex waveforms and with dynamics, so the meters don't
always tell you everything. You want to allow for some ambiguity of
the metering.

> OK, now you have a point, you must first establish exactly where the 0dB FS
> level is.

Exactly - and since there's no industry standard, you have to
determine this for every system if you want to be able to push things
to the limit. And once you know that, you need to choose a source that
can produce that level. And if you're talking about a mic preamp, it
may need more GAIN in order to produce that level. A singer produces a
given sound pressure level. Let's say that with the preamp cranked all
the way open it has 60 dB of gain, and that produces a peak level (a
function of the singer's volume, the distance from the mic, and the
mic's sensitivity - also not standardized) of +16 dBu. Let's assume
that the distortion is acceptable at that level.

Let's say this makes the digital meters on the A/D converter hit
-8 dBFS on peaks. If you're paranoid about losing resolution (maybe
you're working at 8-bits?) what do you have to do in order to make the
meters hit FS on peaks?

Well, you could raise the input sensitivity of the A/D converter, but
we've established that this is fixed in the hardware. So you can't do
that. You could tell the singer to sing louder or move closer to the
mic, or switch to a mic with higher sensitivity, but that might make
changees that aren't acceptable otherwise. You could turn up the
preamp gain, but it's already at maximum. So you need more gain.

> If it's +26dBu, then you will indeed lose resolution by running
> too far below that to avoid analog overload

How much resolution do you think you'll lose, and how do you think
that will affect what you hear? I hate to sound like Phil here, but
you really should do a listening test. There is so much "audible"
headroom with decent 24-bit converters that you can easily work in the
eyeball average range of -20 to -10 dBFS and it will sound just fine.

Another problem that can occur if you're mixing multiple channels all
firing away at full scale output. Converter pairs (such as in a sound
card or a "workstation" converter that has A/D and D/A in a single
box) are generally designed to be "unity gain" so that if it takes
+24 dBU to push the input to full scale, playing back a full scale
recording will produce +24 dBu at the output of the D/A converter. Try
to mix too many of those in an analog mixer and you'll have to reduce
the gain at the mixer. Try to mix too many of those in a digital mixer
(hardware or software) and you may become one of the "mixing in the
box sounds bad" camp.

> However you also lose resolution by running analog stages too
> far below their maximum, EIN being constant.

The problem with that argument is that EIN isn't constant. It's a
function of gain, which is why it's nearly given (on the spec sheet
anyway - the engineering lab is another story) at maximum gain.

Noise floor is more nearly constant, and you can indeed get down into
that noise if things are quiet enough. But given a modern system with
greater than 100 dB of signal-to-noise ratio in the worst of
conditions, it's not really a concern unless you're doing something
really dumb (or are simply trying to prove the point of what happens
when you do).

> Personally I would never design
> such a device. +14dBu FS, with 12 dB pads on the input when necessary, is
> much more useful IMO.

I agree, but you're not the one designing what Marketing has to sell.
The numbers look a lot better when you leave it to "the other guy" to
give you the signal that will get all the performance your device is
capable of.

--
I'm really Mike Rivers )
However, until the spam goes away or Hell freezes over,
lots of IP addresses are blocked from this system. If
you e-mail me and it bounces, use your secret decoder ring
and reach me here: double-m-eleven-double-zero at yahoo