Greg Simmons wrote:
> 2) The second part of my post was in response to someone's previous post =
about upsampling, followed by my own perceived benefits of upsampling (hear=
d through some very high quality equipment) along with what I believe are t=
he reasons for those benefits. I did not mean to imply that upsampling woul=
d be a necessity for a better-sounding SRC...
Okay Greg, understood...
> although if it makes an audible difference in EQ algorithms in channel st=
rip plug-ins, it seems reasonable to assume it would be of benefit to the f=
ilter in an SRC. Perhaps not?
Sure, it is indeed quite difficult to predict the behavior of a specific au=
dio processing application. Actually, I'm not familiar with all the current=
DAW software products - I can just talk about these things based on my own=
practical experience in designing digital signal processing algorithms.
> > Yes, this is true, but the required low-pass filter would be very strai=
ghtforward and can just operate on the original sample stream of 88.1 kHz.
>
> Right... But how straightforward is it, *really*? If you look at the SRC =
comp's site that I linked to earlier, it seems some/most coders have a very=
hard time getting that filter right. And when it is not right, there are s=
ome unpleasant side-effects getting into the audio. Personally, I take a lo=
t of care choosing and using my equipment; I don't want those efforts to be=
stymied by a poor filtering algorithm.
Yes, the stop-band attenuation of this low-pass filter must be sufficiently=
high. If this was not the case, one will get ugly aliasing products that c=
an sound horrible. Fortunately, a high stop-band attenuation can be simply =
achieved by cascading more than one filter.
> Of course the SRC can operate at 88.2ks/s. But *if* upsampling from there=
produced an even cleaner result, that would be worth chasing (in my opinio=
n). The question is, of course, whether that improvement is worth the addit=
ional processing time and whether the user is willing to spend that extra t=
ime.
This is exactly the issue that I was referring to. For theoretical reasons,=
upsampling should have no benefit here. The problem is that it is in any c=
ase required to apply a steep low-pass filter before converting to the fina=
l 44.1 kHz sample rate.
> > Any intermediate up-sampling (to 7.056 MHz) would unnecessarily increas=
e the amount of data that needs to be processed, which would take much long=
er to process.
>
> Why 7.056MHz? I was talking 394ks/s. Or are we using different terminolog=
ies to say the same thing?
|