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)

[voodooless]

Note that the image you linked is NOT frequency response, FR is NOT down 25dB at 20kHz. The image shows FFT spectrum of white noise, to show how the filter works.

You can see the effect of NOS in the RME ADI-2 here https://www.audiosciencereview.com/...ac-and-headphone-amp-review.13379/post-850429
Treble does roll off early, -1dB down at 10kHz, -4dB at 21kHz. Not 25 dB down. The Slow filter actually rolls off harder, about 8dB at 21kHz.

Should white noise not show the frequency response? Why would there be a difference?

 

[shoto]

Note that the image you linked is NOT frequency response, FR is NOT down 25dB at 20kHz. The image shows FFT spectrum of white noise, to show how the filter works.

You can see the effect of NOS in the RME ADI-2 here https://www.audiosciencereview.com/...ac-and-headphone-amp-review.13379/post-850429
Treble does roll off early, -1dB down at 10kHz, -4dB at 21kHz. Not 25 dB down. The Slow filter actually rolls off harder, about 8dB at 21kHz.

And how is frequency response plot produced?

FFT spectrum of white noise? do you think this dac has all harmonics at 0dB below 20k?

 

[Veri]

Should white noise not show the frequency response? Why would there be a difference?

The white noise FFT and the frequency response as shown in the RME manual is a totally different measurement. Amir often measures the white noise response too. This does not show the effect on FR but rather the filter/imaging characteristics. Maybe someone else can explain better..

 

[voodooless]

The white noise FFT and the frequency response as shown in the RME manual is a totally different measurement. Amir often measures the white noise response too. This does not show the effect on FR but rather the filter/imaging characteristics. Maybe someone else can explain better..

I hope so because as far as I know, these things should be equivalent. White noise should follow the frequency response of the filter. If not, then all kinds of strange stuff are going on. Possible content dependant filtering?

Edit: but wait: you're talking about the ADI here? The measurements were from the X16.. Not the same device.

 

[Veri]

Edit: but wait: you're talking about the ADI here? The measurements were from the X16.. Not the same device.

My point was that the ADI has proper FR measurements. You can easily see that NOS for example, the FR response does not at all follow the white noise FFT. They are different measurements...

For example, Project S2 Digital

FFT Spectrum

FR

The 'optimal transient' / pseudo-NOS is the black in the first one. In the FR it is the first orange one, earliest roll-off. White noise FFT != FR. Different measurement alltogether.

 

[anphex]

So what do you guys say? Should I still try and do that RMAA thingy or would it be useless?

 

[voodooless]

So what do you guys say? Should I still try and do that RMAA thingy or would it be useless?

Definitely

 

[voodooless]

White noise FFT != FR. Different measurement alltogether.

You can show pictures all you want, that doesn't bring us any closer to the why of the matter..

 

[shoto]

My point was that the ADI has proper FR measurements. You can easily see that NOS for example, the FR response does not at all follow the white noise FFT. They are different measurements...

For example, Project S2 Digital

FFT Spectrum

FR

The 'optimal transient' / pseudo-NOS is the black in the first one. In the FR it is the first orange one, earliest roll-off. White noise FFT != FR. Different measurement alltogether.

you are not saying how they are different, only they are.
trying to get past difference scales used I see optimal transient about 3db down at 20k in both graphs, it cant be compared anywhere else. where its supposed to be different? I dont understand.

frequency response has cutoff at 20khz, all this means is you get less information than white noise test.
how is frequency response made differently? I think the same method, using white noise signal...

 

[AnalogSteph]

So what do you guys say? Should I still try and do that RMAA thingy or would it be useless?

Oops. I think the white noise + REW approach is actually better if you want to show the filter's out-of-band response. I have never verified the accuracy of RMAA's MLS frequency response test in this area. Dragging out some old results, it doesn't look all that promising. Sorry, my bad.

In-band response is much more accurate and less noisy though. So if you want to know in-band filter ripple (to within +/-0.01 dB or so), go with RMAA, for out of band go with white noise.

 

[Veri]

you are not saying how they are different, only they are.

?? FFT spectrum of white noise is not identical to plain frequency response measurement. That's what I am saying. The proof is there. One tests for filter attenuating/imaging, the other measures the response on frequency. They are not at all comparable. Don't know what else I can say about this, haha.

 

[voodooless]

?? FFT spectrum of white noise is not identical to plain frequency response measurement. That's what I am saying. One tests for filter attenuating/imaging, the other measures the response on frequency. They are not at all comparable. Don't know what else I can say about this, haha.

But why would they be different? It doesn't make any (apparent) sense... Let me rephrase the question then: how do you think the frequency response is measured? Because according to Wikipedia using white noise to deduce the frequency response is a perfectly valid method:

The frequency response of a system can be measured by applying a test signal, for example:

• applying an impulse to the system and measuring its response (see impulse response)
• sweeping a constant-amplitude pure tone through the bandwidth of interest and measuring the output level and phase shift relative to the input
• applying a signal with a wide frequency spectrum (for example multifrequency signals [2] (nonorthogonal frequency-discrete multiplexing of signals (N-OFDM[3][4] or as the same SEFDM[5]) and OFDM), digitally-generated maximum length sequence noise, or analog filtered white noise equivalent, like pink noise), and calculating the impulse response by deconvolution of this input signal and the output signal of the system.

The proof is there.

Not really. I cannot verify where these images came from. I trust the ones from Amir and Wolf, but otherwise... All bets are off. And what we also don't know is if meanwhile firmware was changed (Wolf often gets to test pre-production models). So possibly the filters have changed. All in all, there are way too many variables.

 

[Veri]

Not really. I cannot verify where these images came from. I trust the ones from Amir and Wolf, but otherwise... All bets are off. And what we also don't know is if meanwhile firmware was changed (Wolf often gets to test pre-production models). So possibly the filters have changed. All in all, there are way too many variables.

Amir has measured both filter white noise FFT and overall frequency response before they are not the same thing. Anyway maybe @amirm @solderdude someone can explain what I'm failing to
Edit: thanks, solderdude to the rescue~~

You can't look at a white noise sweep and expect to read its frequency response from it. Different thing, different use.

 

[dc655321]

You can't look at a white noise sweep and expect to read its frequency response from it. Different thing, different use.

Sure you can.

White noise probes all frequencies simultaneously over a given bandwidth, whereas a sine sweep obviously probes a discrete frequency at any given time.

 

[solderdude]

Note that roll-off is really only seen in plots where there is averaging.

A filter with the same treble roll-off as seen in the plots will sound different (rolled-off) compared to a filterless NOS 44.1/48kHz audio file.

I'll show why this is so:

Here is a 16kHz continuous sine passed through a gentle low-pass as found in most DACs as post filter:

On average the level is not reaching peak levels but rather is kind of AM modulated. It varies in amplitude.
It can be seen that certain peaks can still reach almost 0dBFS and thus part is not rolled off while at other points in time the amplitude can be 6dB lower. In music we don't see continuous tones (in general, they do exist in some recordings) but signals are dynamic. It is quite possible that a sample is (near) 0dB and thus may not sound rolled-off but at the proper level. This all depends on the moment the actual sample was taken. On average this means treble will be rolled-off but not as bad as a filter would be.

More on this discussion here

 

[voodooless]

So, does that mean that because the reconstruction is not perfect (enough), it kind of misses the mark a lot, and because of that the roll-off in an averaging is more pronounced than it would be with real music?

 

[anphex]

Ok, I've no idea what I'm doing. The DAC and ADC are setup and working but I don't know what I should measure, what parameters etc. Also, RMAA loves to crash on me. I seem to have gotten a measurement but how do I get the graph? Exporting wav and analyze spectrum?

 

[voodooless]

From what I remember RMAA just makes a report with all the graphs? Do you have them on screen, then just post screenshots?

 

[anphex]

From what I remember RMAA just makes a report with all the graphs? Do you have them on screen, then just post screenshots?

The only graph I get is during the level adjustment, then it just shows a progress bar and slaps numbers onto the screen and a "saving" slot.

Edit: Okay, so when I run a test, export the wav and import that wav as spectrum analysis I get this illegible thing. Though that weird succesion of steps still feels wrong to me.

 

[voodooless]

The test results screen has a button to generate a html report.