I tend to prefer slow roll offs and even the NOS mode - what does this say about my setup and preference? (Conclusion: back to fast linear)

[MRC01]

In summary, the potential advantage of a "slow" filter is that at high sampling rates, it can still start to attenuate just above 20 kHz. This gives a wider transition band, which means a less steep slope, which makes the filter easier to implement while avoiding passband artifacts. One could call this a "well engineered" slow filter, in that it does not attenuate audible frequencies and it fully attenuates by Nyquist.

However, the term "slow" is also used to describe filters that start attenuating well below 20 kHz, or that don't fully attenuate by Nyquist. Filter like this are "poorly engineered", which makes the phrase "slow filter" ambiguous and confusing.

 

[KSTR]

[...]See the crappy output of the 10 and 16 kHz sine? Yes, your ear will filter out the high frequency crap, it will however not fix the amplitude errors. That modulation you will hear in the form of IMD.

That "modulation" actually is the effect of the image tone. Both tones together with almost equal amplitude generate the picture of a single frequency, fs/2, modulated in amplitude.
This effect is not restricted to NOS, though, it appears inevitably with any filter that does not reach full stop-band attenuation at fs/2. With a typical "slow" filter which is only ~ 10dB down at fs/2, the affected frequency range is a way smaller region below fs/2 than with NOS, but it is still there with very high frequencies:

It is distortion in that new frequencies are introduced that have not been there in the original signal, but it is not IMD. The frequency doublets might trigger more IMD analog downstream, though.

With a "fast" filter, the affected frequency range is so small and the level of the mirrors is generally greatly reduced that it is practically irrelevant.

 

[MRC01]

... With a "fast" filter, the affected frequency range is so small and the level of the mirrors is generally greatly reduced that it is practically irrelevant.

By "fast", you mean a filter that fully attenuates by Nyquist, right?
Consider the 96 kHz sample rate...
Filter (A) has flat response to 45 kHz, transition band 45 kHz to 48 kHz, and stop band (-100 dB) at 48 kHz.
Filter (B) has flat response to 20 kHz, transition band 20 kHz to 48 kHz, and stop band (-100 dB) at 48 kHz.
By your definition, both (A) and (B) are "fast"?

 

[KSTR]

By "fast", you mean a filter that fully attenuates by Nyquist, right?

Ehm, no. I'm using the naming convention of manufacturers like AKM, where "fast" means a steep slope into the stop-band... which is above fs/2.
Your filters A and B would fall into the brickwall category (zero output above fs/2) with differently sized transition regions.

 

[MRC01]

OK let me put it differently: regarding the distortion you show above, the width of the transition band is immaterial. So long as the filter fully attenuates by attenuates by Nyquist, you won't get that distortion.
A simple topic is made confusing by the ambiguous terms "fast" and "slow" ...

 

[KSTR]

OK let me put it differently: regarding the distortion you show above, the width of the transition band is immaterial. So long as the filter fully attenuates by Nyquist, you won't get that distortion.

Exactly.

 

[MRC01]

I'd love to devise terminology that is concise, intuitive, and unambiguous for the key aspects of digital filters.
Suggestions:
1. Passband attenuation ("flat" vs. "soft")
2. Width of transition band ("fast"/"narrow" vs. "slow"/"wide")
3. Attenuation at Nyquist ("clean" vs. "dirty")
4. Impulse response ("symmetric"/"acausal" vs. "asymmetric"/"causal")

For example, the typical filters we see in DAC reviews at 44.1 kHz sampling are flat, fast/narrow, dirty, symmetric/acausal. I say "dirty" because most of them are only 6-10 dB down at Nyquist and don't fully attenuate until 24,100 Hz.

The mathematically ideal filter resulting from Shannon-Whittaker reconstruction would be flat, fast/narrow, clean, symmetric/acausal.

 

[dc655321]

I'd love to devise terminology that is concise, intuitive, and unambiguous for the key aspects of digital filters.
Suggestions:
1. Passband attenuation ("flat" vs. "soft")
2. Width of transition band ("fast"/"narrow" vs. "slow"/"wide")
3. Attenuation at Nyquist ("clean" vs. "dirty")
4. Impulse response ("symmetric"/"acausal" vs. "asymmetric"/"causal")

For example, the typical filters we see in DAC reviews at 44.1 kHz sampling are flat, fast/narrow, dirty, symmetric/acausal. I say "dirty" because most of them are only 6-10 dB down at Nyquist and don't fully attenuate until 24,100 Hz.

The mathematically ideal filter resulting from Shannon-Whittaker reconstruction would be flat, fast/narrow, clean, symmetric/acausal.

There is already adequate and wide spread terminology for filter characteristics - engineers use them all the time. Please don't try to invent another vocabulary.

It's audiotools and marketeers that pervert the jargon...

 

[MRC01]

In that case, what exactly do "slow" and "fast" really mean? Or are they among those terms that audiotools & marketeers perverted?

 

[dc655321]

Or are they among those terms that audiotools & marketeers perverted?

Absolutely.
"Fast" and "slow" convey almost nothing useful.

Maybe lawyers were involved?

 

[anphex]

Hmmm I have more questions than I had when starting this thread now.

I play my music only from Windows 10 with lossless streaming services and smack my playback devices to the highest setting they provide (Gustard 384/32) for example. Thinking "well oversampling can't be bad right?". Also I thought that when checking "Applications have exclusive control over device" (loosely translated from german) the playback software would tell the DAC the sample rate and adjust it regardingly. But since the Gustard is display 384khz all the time this can't be right.

So the DAC is when using MSLOW starting it's roll off a few khz before 192khz and then attenuates slowly. Then the hiss that I disliked with LFAST and still noticed a little with MSLOW were the unattuated frequencies from 20-192 khz? And NOS rolls them off slowly at 20khz no matter the sample rate right?

Then, in the end, if probably makes sense. Then I should try 44,1/16 with LFAST and see how it sounds. I would love a software that automatically adjusts the DAC settings to the bit perfect sample rate of the current playing media. Maybe ASIO can do that?

Then there is still the question of pre and post ringing and the transient shape. How audible is it and what is placebo? What would be considered a "right" transient? One that follows the pulse more accurately in the time/amplitude domain or frequency domain? Earlier in this thread it was said that it's frequency but I am not really sure for myself. As I mentioned many times now the NOS mode made it feel like the sound was suddenly free and detached from the speakers with subjective artifacts in my hearing range.

 

[voodooless]

That "modulation" actually is the effect of the image tone. Both tones together with almost equal amplitude generate the picture of a single frequency, fs/2, modulated in amplitude.
This effect is not restricted to NOS, though, it appears inevitably with any filter that does not reach full stop-band attenuation at fs/2. With a typical "slow" filter which is only ~ 10dB down at fs/2, the affected frequency range is a way smaller region below fs/2 than with NOS, but it is still there with very high frequencies:
View attachment 144155
It is distortion in that new frequencies are introduced that have not been there in the original signal, but it is not IMD. The frequency doublets might trigger more IMD analog downstream, though.

With a "fast" filter, the affected frequency range is so small and the level of the mirrors is generally greatly reduced that it is practically irrelevant.

Isn't the amplitude modulation always lower in frequency than the encoded tone?

 

[voodooless]

Let's try to dissect this then:

I play my music only from Windows 10 with lossless streaming services and smack my playback devices to the highest setting they provide (Gustard 384/32) for example. Thinking "well oversampling can't be bad right?".

It can be, but I think by now even Windows manages to do a fairly decent job. Why would nu use NOS then ?

So you're saying the NOS/OS switching you do with upsamples content to 384/32? If so, those EQ tests are quite useless. The artefacts only happen after 192 kHz in that case. I'm pretty sure all differences are then down to the difference in volume that we've seen in your RMAA tests.

Also I thought that when checking "Applications have exclusive control over device" (loosely translated from german) the playback software would tell the DAC the sample rate and adjust it regardingly. But since the Gustard is display 384khz all the time this can't be right.

The checkbox actually means the application CAN have exclusive access. That doesn't mean they actually do it. Usually, you'll need to put them in some kind of special mode to have them do that. And then the application can CHOOSE to set the DAC sample rate. Those are two separate things.

So the DAC is when using MSLOW starting it's roll off a few khz before 192khz and then attenuates slowly. Then the hiss that I disliked with LFAST and still noticed a little with MSLOW were the unattuated frequencies from 20-192 khz? And NOS rolls them off slowly at 20khz no matter the sample rate right?

NOS roll-off is also dependant on the sample rate. Not sure what hiss you hear.. from what can be seen from the DAC review it should be absolutely silent.

Then, in the end, if probably makes sense. Then I should try 44,1/16 with LFAST and see how it sounds. I would love a software that automatically adjusts the DAC settings to the bit perfect sample rate of the current playing media. Maybe ASIO can do that?

What are you using right now? Check your settings for the appropriate options.

Then there is still the question of pre and post ringing and the transient shape. How audible is it and what is placebo? What would be considered a "right" transient? One that follows the pulse more accurately in the time/amplitude domain or frequency domain? Earlier in this thread it was said that it's frequency but I am not really sure for myself. As I mentioned many times now the NOS mode made it feel like the sound was suddenly free and detached from the speakers with subjective artifacts in my hearing range.

Read up on this: https://archimago.blogspot.com/2018/01/audiophile-myth-260-detestable-digital.html

 

[charleski]

Yay, I am doing amateur science. Do I get a badge?

Of course.

 

[MRC01]

... I play my music only from Windows 10 with lossless streaming services and smack my playback devices to the highest setting they provide (Gustard 384/32) for example. Thinking "well oversampling can't be bad right?"...

Depends on the method used. Resampling done properly is transparent. But it's not always done properly. Safest thing is to play each track at its native sampling rate, avoid resampling.

... Then the hiss that I disliked with LFAST and still noticed a little with MSLOW were the unattuated frequencies from 20-192 khz? And NOS rolls them off slowly at 20khz no matter the sample rate right? ...

"NOS" is a general term for filters that are poorly engineered in various ways. It may mean there is no filter at all, or an insufficient one, in which case it leaks all kinds of HF noise that aliases into the passband creating audible distortion. You'd have to look at the specs to know what exactly what this particular NOS filter is doing. The other common filter options: minimum vs. linear phase, fast vs. slow attenuation, usually sound identical with music, though some people may be able to hear subtle differences with specially crafted sounds like square waves or jangling keys.

... I would love a software that automatically adjusts the DAC settings to the bit perfect sample rate of the current playing media. ...

Yes, that is the best way to listen. Set up the PC to deliver the audio output of each track "bit perfect" unmodified. When it does this, the DAC will automatically detect the sample rate & bit depth.

 

[anphex]

At the end of the day I can just guess that since choosing a more loose roll off gives a sharper, more "accurate" transient with less pre and post stuff is the reason I like it better. The only other high impact criteria might be the roll off starting at about 17khz. Everything is just so precise, clean, natural ... I wish I could invite someone of you over to Berlin to show you.

You laid out so much as to why NOS isn't a good choice technically and audibly, yet I like it. Maybe this is my tube amp placebo haha.

 

[voodooless]

Or it’s just the volume difference..

 

[anphex]

Or it’s just the volume difference..

I am confident that this would only matter when doing direct AB because of the contrast. But I have it enabled for hours and days and still notice the improvement from before.

 

[MRC01]

... You laid out so much as to why NOS isn't a good choice technically and audibly, yet I like it. Maybe this is my tube amp placebo haha.

It wouldn't be all that surprising. On a related note, lots of people prefer vinyl.

 

[voodooless]

I am confident that this would only matter when doing direct AB because of the contrast. But I have it enabled for hours and days and still notice the improvement from before.

That doesn’t really make it a very reliable comparison. Especially if you have the upsampling to over 300 kHz enabled, there should be no audible difference. If not I can imagine that there is a difference besides the volume. May very well be that you prefer it.

To really know a double blind test should be done..