On Wed, Dec 6, 2023 at 11:10 PM Ethan Duni <[email protected]> wrote:

> > On Wed, 6 Dec 2023 at 10:43, robert bristow-johnson 
> > <[email protected]> wrote:
> > >
> > >  ... Let's say that word width isn't a problem (low normalized frequency 
> > > not a problem) ...
> > >
> > > Perhaps I'm wrong but it seems clear to me that any discrete-time model 
> > > of an analog process (linear or not, time-invariant or not) will get 
> > > better with higher sample rate. So however good your filter is at 48kHz 
> > > the model (if it's a good model) will be better at 96kHz.
>
> Seems like this is only true for discretization methods like bilinear 
> transform. For MZTi or similar, it is not clear that larger sample rate gives 
> a better result.
>
> It seems unhelpful to ignore finite word-length effects in this context, 
> given that the method works exactly by driving all roots down to lower 
> normalized frequencies.
>
> Depending on the algorithm and on the implicitly assumed default
word-length, one probably could ignore the quantization effects to an
extent. There are clearly two opposite degradation effects. One grows as
the sample rate gets lower, resulting in various artifacts. I believe
pretty much no matter what's the approach, there will be a degradation in
the frequency response, either in the amplitude or in the phase part or in
both. Conversely, the quantization-related issues grow with the sample
rate. I also believe the same effects are present in pretty much any DSP
algorithm, just instead of degradation of frequency response we might e.g.
have aliasing (which is similar to frrequency response issues), or latency
(if we do oversampling or some equivalent thing). The tradeoff point (even
if we ignore the CPU costs) is to an extent subjective, as different people
may have different judgement about the criticality of the arising artifacts.

So I disagree that it's unhelpful to ignore finite word-length effects.
Depending on what you mean by "ignore". Of course, one should be aware of
them, but they do not necessarily start to show up already in the range of
44-88kHz.

Best regards,
Vadim

--

Vadim Zavalishin

Reaktor Application Architect

Native Instruments <https://www.native-instruments.com/> – now including
iZotope, Plugin Alliance, and Brainworx

Reply via email to