[Randy Yates[_2_]]

PStamler writes:

On Thursday, October 8, 2015 at 4:50:35 PM UTC-5, Randy Yates wrote:
(Scott Dorsey) writes:

PStamler wrote:
On Monday, October 5, 2015 at 8:20:00 PM UTC-5, Trevor wrote:
On 6/10/2015 5:29 AM, PStamler wrote:
Record in 24 bits (or 32 bits floating); after you've done everything y=
ou want to do to the
signal, convert to 16 bits for burning a CD.
=20
Had to smile that you think you will need more than 96dB DNR to record=20
old vinyl :-)
(Most DAWS work internally at 32bit floating, or better now anyway)
Still, won't hurt at least.

That's a long and complicated discussion, which I don't have time for while=
the rice is on the stove. Suffice it to say that every declicking algorith=
m I've ever used works more effectively, with fewer artifacts, on 24-bit fi=
les (or the 32-bit version of them) than on 16. Worth using just on that ba=
sis.

That's weird. I have noticed dramatic improvements in declicking from using
higher sampling rates and wideband preamps; even though there isn't much signal
at 30 KHz there is enough to make for a better-defined edge detection.

But I have not noticed improvements from the longer sample size.

Now... I would most definitely believe that the longer sample size would be
a huge win if you're doing the RIAA de-emphasis in software, since the needed
dynamic range in that case is pretty huge. Even with the de-emphasis, if
your preamp is very wideband you may see that half of your dynamic range is
being eaten up by rumble that you're just going to filter anyway, so there is
a need to have considerably more dynamic range than the record itself may
have usable.

With software, you can always resample internally to a higher resolution,
perform the algorithm, then (if desired) requantize back to 16 bits at
the end.

But there's no reason not to digitize it at 24 bits in the first
place, with storage so cheap now.

No argument there. I'm just saying that if you did, that shouldn't cause
a problem for deemphasis software. I mean, I would hope the (plugin?)
implementer(s) would know the dynamic range requirements of the
algorithm and resample/requantize if necessary.
--
Randy Yates
Digital Signal Labs
http://www.digitalsignallabs.com