1 (edited by ning 2020-12-20 14:26:13)

Topic: oversampling for DSP?

There are quite a few rumors that while oversampling may not improve sound quality of the original audio, it may improve the result of DSP (such as corssfeed, bass/treble boost, or parametric EQ).  this topic is also discussed in this forum: https://forum.rme-audio.de/viewtopic.ph … 04#p157104

I want to understand if such rumor is true --- That is, is there any benefits for me to oversample 48khz format by 4x to 192kHz, in order to get better result from the DSP algorithms the ADI-2 provides.

equivalent question:  does upsampling help to lower the noise of a given IIR (in ADI-2, most  dsp algorithms such as PEQ or Bauer are all implemented with IIR filters [Infinite impulse response])?

2

Re: oversampling for DSP?

The ADI-2 Pro/DAC/TotalMix already use oversampling for the DSP calculations. Pretty much standard for a long time already.

I never heard or measured 'noise' from the IIR in Crossfeed.

Regards
Matthias Carstens
RME

3 (edited by KaiS 2020-12-20 15:41:07)

Re: oversampling for DSP?

I heard an advantage going from 44.1 to 96 kHz, with PEQ treble boosts, as mentioned in the linked posting.
Nowhere else.

The improvement wasn't night and day.

SRC is only available with COAX, OPTICAL and AES, not with USB.
This significantly limits the use for me.

4 (edited by ning 2020-12-20 18:39:11)

Re: oversampling for DSP?

> The ADI-2 Pro/DAC/TotalMix already use oversampling for the DSP calculations. Pretty much standard for a long time already.

Oh, that's amazing! never realized that --- manual only mentions dsp works in 42 bits but never mentions about its sample rate. Since DSP has limitations above 192kHz, is it reasonable to guess everything below 192 will be oversampled to 176.4 or 192 before dsp, and downsample to their original rate after dsp?

> I never heard or measured 'noise' from the IIR in Crossfeed.

by noise I mean quantization/aliasing/whatever error caused by using an IIR filter under a too low sample rate. it's purely digital. Yeah 42bits will eliminate most of the SNR problem caused in the amplitude calculation (such as floating number overflow). But I was curious about the noise in the discrete time domain if source data is not oversampled.


> I heard an advantage going from 44.1 to 96 kHz, with PEQ treble boosts, as mentioned in the linked posting.

this is doubtful. 44.1 to 96 will likely introduce resampling artifacts. probably what you heard is just artifacts. for 44.1 you need to oversample to 88.2.

> SRC is only available with COAX, OPTICAL and AES, not with USB.

This is a different topic. we're talking about when an internal sample rate is given (after the source is SRC-ed), what sample rate is the DSP algorithms working under.

5 (edited by KaiS 2020-12-21 08:48:54)

Re: oversampling for DSP?

ning wrote:

> I heard an advantage going from 44.1 to 96 kHz, with PEQ treble boosts, as mentioned in the linked posting.

this is doubtful. 44.1 to 96 will likely introduce resampling artifacts. probably what you heard is just artifacts. for 44.1 you need to oversample to 88.2.

I didn't do a blind A/B, I stumbled across this by chance.
Then I repeated it to find out what's going on.


What I heard was some kind of silkyness and softness in the treble with some added brilliance - I liked it.


Possible reasons:

• SCR introduces audible artifacts that "smooth" the sound and /or add harmonics.
• The DA-Filter is shifted out of the audible range. I used "Sharp" at that time, that I later found out is not my preferred one.
• The EQ algorithm works different (more correct) on higher SR's. This is true for most EQs, but this case would be clearly measurable, as it affects the frequency response.
With MC's above information this is now unlikely.
I didn't do any measurements.
• A combination, as I heard it only with boosted treble.
• Pure imagination.


SRC chips don't care about integer multiples, good algorithms don't introduce different artifacts with different ratios.
The software SCR's I use in the studio e.g. don't introduce measurable artifacts at all, even at 32bit float files.

Re: oversampling for DSP?

Is there a technical white paper that introduce how TotalMix/ADI-2 Pro/DAC dsp works?
I can’t find relevant information online.
Topics may include general information such as oversampling, internal bit depth, dithering algorithm, DSP pipeline order(the order to apply each DSP algorithm), etc.
Right now there is very little information concerning for example, if crossfeed is performed before EQ or after. They certainly will result very differently.
A PowerPoint slides with diagrams is useful as well.

7

Re: oversampling for DSP?

We don't have plans to release any of this, sorry.

Regards
Matthias Carstens
RME