• Welcome to the General Discussion forum for UAD users!

    Please note that this forum is user-run, although we're thrilled to have so much contribution from Drew, Will, and other UA folks!

    Feel free to discuss both UAD and non-UAD related subjects!

    1) Please do not post technical issues here. Please use our UAD Support Forums instead.

    2) Please do not post complaints here. Use the Unrest Forum instead. They have no place in the the General Discussion forum.

    Threads posted in the wrong forum will be moved, so if you don't see your thread here anymore, please look in the correct forum.

    Lastly, please be respectful.

Upsampling - geek q

Macc

Established Member
After a somewhat in-depth sample rate conversation with someone, I have to ask;

Assuming one is working at 44.1kHz, what benefit does using, say, the 1073 have over the SE?

Obviously oversampling is used to minimalise anti-aliasing in the output/downsampled version, but if frequencies above 22050Hz aren't present in the initial version, what difference does it make?

The only thing i can think of is; Is the upsampled information averaged, by using all the extra samples to create an average for a given initial sample? If that is the idea, isn't a far greater factor than 4x or 8x oversampling required to be of any benefit?

I know a lot of people don't give a *bleep* but I'd really like to understand this, from a pure physics point of view. Don't be afraid to get technical :twisted:

:?

Thanks!
 

kleinholgi

Shareholder
1. As far as I know the full versions do a lot more than just upsampling.
The SEs often leave out nonlinearities in amplitude or frequency response, that cost DSP time. Also sometimes special analog hardware effects that result from e.g. transformer coupling in the audio path is simulated in the big but not the SE versions.

2. Even if we leave out the special vintage emulation component, EQs can profit from bandwidths beyond the typical Nyquist frequency fs/2.

If you e.g. want to increase gain @17kHz with a Q factro not too sharp, a nice behaviouur would include effect modeling of frequencies above 22Khz. Because artifacts can sometimes be mirrowed at the Fs/2 axis down into the audible spectrum, it can be possible that the differences are really heard and not only theoretical.

The same is true for the \"poles\" of the lowpass filter in any DA converter. The higher the sampling frequency, the narrower the filter can be designed (because the cutoff can be set higher as well, leading to less phase shifting, less side effects, better quality overall).
 

Cabbage

Active Member
I am one of those guys who do give a *bleep*. ;)

A digital EQ always has to have unity gain at the Nyquist frequency (i.e. 22.05 kHz when using a 44.1 kHz sample rate). Let's say you have a high shelf with a 10dB boost. In order for the EQ to reach 0dB at 22.05kHz, the gain will have to tail off from 10dB to 0dB over a audiable frequency range just below 22.05 kHz. This means that you will not have a true shelf response. If you have a higher sample rate, this effect is clearly out of the audiable range, and what you hear is a true shelf.

This would be one reason that upsampling is good, but I am sure UA have quite a few more (probably related to modeling of non-linearities).

Are you sure there is no \"Ask the Doctors\" article about this?

Petter
 

Macc

Established Member
Thanks for your replies gents :)

Yes, the Neve SE wasn't a good example really thinking about it :oops: I'm fully aware of the nonlinearities, I just neglected to think about it when choosing my example :lol: :roll: :idiot:

It was really a question just about upsampling. I think I have read all the 'Ask The Doctors' articles that interest me (which would include this) but I'll double check.

So then :D ;

kleinholgi said:
If you e.g. want to increase gain @17kHz with a Q factro not too sharp, a nice behaviouur would include effect modeling of frequencies above 22Khz. Because artifacts can sometimes be mirrowed at the Fs/2 axis down into the audible spectrum, it can be possible that the differences are really heard and not only theoretical.
Hmmm - 'sometimes be mirrored'? Aren't such potential aliases always removed by the anti-alias filter (at least waaaaaay below being audible after mirroring)?

I don't buy that to be honest - either the AA filter works or it doesn't, surely? :? :)

The same is true for the "poles" of the lowpass filter in any DA converter. The higher the sampling frequency, the narrower the filter can be designed (because the cutoff can be set higher as well, leading to less phase shifting, less side effects, better quality overall).
That's the reason I mentioned for oversampling though I think? That oversampling allows gentler anti-alias filter slopes (and so easier design/construction and better results)? :? I think we're saying the same thing there, sorry if I have misunderstood.


I feel funny replying to a cabbage :lol: , but;

Cabbage said:
A digital EQ always has to have unity gain at the Nyquist frequency (i.e. 22.05 kHz when using a 44.1 kHz sample rate). Let's say you have a high shelf with a 10dB boost. In order for the EQ to reach 0dB at 22.05kHz, the gain will have to tail off from 10dB to 0dB over a audiable frequency range just below 22.05 kHz. This means that you will not have a true shelf response. If you have a higher sample rate, this effect is clearly out of the audiable range, and what you hear is a true shelf.
Hmmm (again :D ) - but if your project is at 44.1kHz, you have the following;

44.1 signal> plugin upsamples to 192 signal > plugin does calculations and downsamples > 44.1 signal

So here, with your first sentence being true, there will still have to be the 'from 10dB to 0dB over a audiable frequency range just below 22.05 kHz' after the plugin has done its work, the same as before upsampling, right?

Of course, if you are simply working at a higher SR, or not downsampling for some reason, then yes, you can keep the true shelf response up to 22050. But if you are at 44.1, then surely as you say, you won't get the true shelf response?


So if that's correct (PLEASE correct me if I am wrong!! :lol: ), then why upsample?

Again. sorry for the inital confusion - the nonlinearity point was a silly oversight on my part.


Mmmmmmmmm..... Nonlinearities
 
UAD Bundle Month
Top