[BC] dueling algorithms and audio quality

[lists at loudandclean.com]

    As I see it, the issues with 16 bits are (including but
not limited to):

[1] when you are down only 6 dB below "all-ones", you've lost
_half_ your quantization steps.  Instead of 65,536 "steps"
between the positive and negative peaks of the audio waveform,
you only have 32,768.  Go down another 6 dB and you only have
16,384.  So, you lose resolution really fast when you leave
any headroom at all.  But wait, we _must_ leave headroom!

[2] recent papers about the performance of D/A converters in
CD players seem to indicate that constant energy in the top
6 dB may cause error correction circuitry to activate and
never recover, introducing significant distortion.  OUCH!

[3] The "loudness wars" among the record companies fighting
to be the loudest CD a radio station music director hears
on "new music Tuesdays" constantly exacerbates issue #2.
Why can't they only do that to the promo CDs and not to the
CDs sold to consumers?

[4] Because the source-to-source level matching in a typical
radio station is poor, lightly processed audio from a radio
station on-air studio should have at least 10 dB of headroom
going through a 16-bit Studio-to-Transmitter-Link.

[5] In the last few years, I've done some live multitrack
recordings of fairly complex musical ensembles, and then
mixed them down to stereo and 5.1 mixes for DVDs and the
like.  It is an eye-opening experience.  Keeping things
in the 24-bit/96 kHz world until the absolute final moment,
so that I could leave 15-or-so dB of headroom until the
final mix and overall processing makes a dramatic difference
in the sonic details.  Furthermore, the music is actually
psycho-acoustically louder-sounding than if I drop to 16
bits earlier in the process.  2-gigabyte audio files are a
bear to store and manipulate, but it turns out it's worth it.

    So, I agree with Mark's comments below.  As long as the
STL is a linear 16 bit device, I like to do as much of the
processing as I can in front of the STL, workin' those bits
as much as I can, while still leaving about 6 dB of headroom
for D/A error correction safety margin.

    My goal at this point is to do the "HD" processing at
the studio (AGC, multi-band platforming, multi-band
compression and non-pre-emphasized limiting of the summed
bands), send that to the transmitter and just put it right
into the HD box.  Then I take that same audio, lightly
process it a bit more, do the pre-emphasized limiting,
HF limiting, clipping, stereo generation, light composite
overshoot clipping and HD/analog alignment delay, and shove
that into the composite input jack of the exciter.

    And, I REALLY WANT a 24-bit digital audio router in the
station.  I just don't have the money.

Grady

> From: Mark Croom <markc at newmail.kinshipradio.org>
> Subject: Re: [BC] dueling algorithms and audio quality
>
> Duh, I left the multiband box in front of the STL, and sent
> fully processed audio out to the new site. I put the single
> rack-unit brand O box out there to function as overshoot
> compensation and stereo generator.  Worked great and I think
> the station sounds pretty decent partially because we're
> filling as much of the available 16-bits in the STL as I can
> with my available equipment.
>
> Mark
> MN