PDA

View Full Version : Questions about new 24/96 support



freelunch
2006-09-20, 07:56
I just installed the 9/19 nightly which features 24/96 downsampling to 24/48..

THANK YOU!!

I found it interesting that the downsampling is being done via the
firmware and apparently at the SB. I had always thought we'd see
this support appear via transcoding.

Does anyone have any info on how the downsampling was implemented? I
did some searches but could not come up with any info. I don't think
the firmware source is public so I can't check it out that way. I'd
like to avoid asking the devs because I know they're busy....


Thanks,

FL

freelunch
2006-09-22, 22:10
Bump.......

freelunch
2006-09-25, 07:05
Support suggested I keep this bumped.

freelunch
2006-09-27, 10:59
Bump...........

cliveb
2006-09-27, 11:12
Does anyone have any info on how the downsampling was implemented?
Since you have now bumped this twice, and nobody better qualified seems willing to say anything....

I don't *know* how the SB does the downsampling, but it would surprise me if it wasn't done the obvious way. That is, to downsample 24/96 to 24/48, all you have to do is throw away every other sample. You're not changing the word length, so no dithering is needed, and since the sample rate is reduced by a factor of two, no fancy DSP is required.

dean
2006-09-27, 11:14
The 96k to 48k downsampling is just done by dropping samples, i.e. no filtering or interpolation. There's not enough CPU power to do more.

T
2006-09-27, 13:32
> I don't *know* how the SB does the downsampling, but it would surprise
> me if it wasn't done the obvious way. That is, to downsample 24/96 to
> 24/48, all you have to do is throw away every other sample. You're not
> changing the word length, so no dithering is needed, and since the
> sample rate is reduced by a factor of two, no fancy DSP is required.

That's the simple way, but not the correct way.

The correct way is first to filter it for the new Nyquist before thowing
away the samples.

Tom

Mark Lanctot
2006-09-27, 16:56
The correct way is first to filter it for the new Nyquist before thowing
away the samples.

Sure, but as Dean just said:


There's not enough CPU power to do more.

Ben Sandee
2006-09-27, 20:02
On 9/27/06, Mark Lanctot <
Mark.Lanctot.2etidz1159401601 (AT) no-mx (DOT) forums.slimdevices.com> wrote:
>
>
> T;140337 Wrote:
> > The correct way is first to filter it for the new Nyquist before thowing
> >
> > away the samples.
>
> Sure, but as Dean just said:
>
> dean;140322 Wrote:
> > There's not enough CPU power to do more.


I think he'd rather hear silence.

Ben

cliveb
2006-09-28, 02:12
> I don't *know* how the SB does the downsampling, but it would surprise
> me if it wasn't done the obvious way. That is, to downsample 24/96 to
> 24/48, all you have to do is throw away every other sample. You're not
> changing the word length, so no dithering is needed, and since the
> sample rate is reduced by a factor of two, no fancy DSP is required.

That's the simple way, but not the correct way.

The correct way is first to filter it for the new Nyquist before thowing
away the samples.
Yes, good point. The pre-filtering is required if the signal contains frequencies above half the target rate (ie. 24kHz in this case). But in practice, typical musical signals contain little or no such frequencies, so simple sample dropping is usually safe. I think the SB approach is a very pragmatic solution, and is the way I expected it to work.

freelunch
2006-09-28, 10:33
On 9/28/06, cliveb <cliveb.2eu8301159434902 (AT) no-mx (DOT) forums.slimdevices.com> wrote:
>
> Yes, good point. The pre-filtering is required if the signal contains
> frequencies above half the target rate (ie. 24kHz in this case). But in
> practice, typical musical signals contain little or no such frequencies,
> so simple sample dropping is usually safe. I think the SB approach is a
> very pragmatic solution, and is the way I expected it to work.

At last some discussion! ;-)

My 24/96 sources are all live recordings and, not sterilized by
processing, contain significant detail above 24khz that isn't noise.

So we've been told that the hardware can't do 24/96. Can the SPDIF
output 24/96?


FL