On Sun, Dec 23, 2012 at 01:07:19PM -0500, Joe Hartley wrote:
The usual combination of technical and economic factors.
Every AD or DA needs an anti-aliasing filter. Say you want a
flat response up to 20 kHz, and at least -120 dB at half the
sample rate. At 44.1 kHz that means roughly -60 dB/kHz, at
48 kHz -30 dB/kHz for the transition band of the filter.
Now the complexity of an HP filter (the number of coefficients)
is proportional to those numbers. That means that for the same
specs at 44.1 kHz you need twice the number of gates, chip area,
power etc. as at 48 kHz. That is a simplification, if you take
everything into account the ratio is even bigger. This, plus the
fact that 44.1 kHz is considered to be a 'consumer' rate anyway
means there is a strong incentive for AD/DA chip designers to cut
corners at 44.1 kHz. And that is exactly what happens in almost
all cases, with very few exceptions. The alternative, which makes
sense, is to forget about HW support for 44.1 kHz and let the
driver resample from 48 to 44.1 if the user wants to use that
This is also valid for higher sample rates.
At 96 kHz, assuming you want a flat response up to say 32 kHz,
you only need 7.5 dB/kHz. Since the _delay_ of the filter is also
proportional to the transition band slope, such a filter can have
much less latency than at 48 kHz, and that's the reason why e.g.
digital snakes and mixers designed for PA use operate at 96 kHz.
Low latency is essential for PA, in particular if in-ear monitors
A world of exhaustive, reliable metadata would be an utopia.
It's also a pipe-dream, founded on self-delusion, nerd hubris
and hysterically inflated market opportunities. (Cory Doctorow)
Linux-audio-user mailing list