20Hz-20KHz frequency sweeps

[manisandher]

I did this a while back, but with some new filters in XX, I thought it'd be interesting to repeat it.

I have a 16/44.1 wav file that sweeps from 20Hz to 20KHz in 33 seconds at a constant level. So, I simply play this file in XX and monitor the output of XX using RME's 'Digicheck' software.

A few notes:

1. I don't think it's wise to equate any of these graphs with SQ.
2. I'm not sure what happens below 100Hz with 4x upconversion applied - a bug in the RME software maybe?
3. My ADC/DAC is capable of 4x upconversion internally - the output looks identical to 4AI in XX (I had to use a separate laptop/card to generate a 16/44.1 input for my ADC/DAC and then monitored the 24/176.4 output from the ADC/DAC on my main PC - but I forgot to capture this image).

Peter, are these graphs as you would expect? I was surprised by the 4AP graph...

Cheers,
Mani.

[pedal]

Peter, are these graphs as you would expect? I was surprised by the 4AP graph...

My guess: The "AP" process is triggered by square waves, which aren't present in your test signal. 

[manisandher]

Agreed.

So, I was expecting 4AP to look similar to 4AI (for these sine-wave sweeps at least). Actually, it is pretty similar above 20KHz... but there is a slight HF roll-off beforehand. I guess this due to artifacts above 88.2KHz, which I can't measure.

In which case, I don't understand is why 4AP is more rolled off than 4_no_filter at ~20KHz...

Mani.

[PeterSt]

Well, that costed me a sleepless night ! Ok, a bit then.

I think it is good to look at the "no_filter" first; Here we see what happens when nothing is done. The roll off starts at 3KHz and at 20KHz it is -1.5dB. At 16KHz (say, audible) it also is -1.5dB.
The drop in output volume is caused by not all samples are registered on the peaks of the original wave, so the average output of such a periodic wave gets lower (plus it's no good sine).
The energy beyond 22050Hz are images, and -roughly said- are caused by the wrong shapes of the waveform.

If we now look at the 4AI example, we see that this filter is constructed so that there is no roll off in the audible area at all, at the cost of ringing (which you can't see). Notice that the 0.5dB rise you see at 16KHz is real, because no filter of this kind can be "linear". Notice that these bumps (can be pits as well) depend on the frequency, and 16.00KHz can be totally different from 16.01KHz.

The 4L example (Linear Interpolation) doesn't help a bit, and only the images are lower in volume. The roll off is even higher than with no_filter.

Finally, the 4AP example shows that the roll off has been "postponed" to 10KHz, at 20KHz it is at -3.5dB and at 16KHz (assumed audible) it is at -1.5dB (compare with the -1.5dB at no_filter). Btw, you could try to run this through 2AP, in order to get an idea (extrapolate the data) of what will happen to 8AP which is what it should be at least according to me.
What you also see (as far as the details allow this to see) is that there is no imaging beyond the audioband, and what you again can not see is that there is no ringing.


Having said this all, no, maybe I didn't expect the roll off to start that soon, especially because I never saw it in my own FFTs. However, alhtough the 1.5dB (or 3dB) isn't all that much, how can it be more than no_filter. We must be careful though how the "FFT" of Digicheck works, and whether it's "good enough", so to say;
If we look at the low side, we always see the exact same problem there, while I know for sure this problem doesn't exist with AP. I know this, because it *is* an attention point with normal filtering (think about the slowness of low frequency waves and how many samples it would take to "reconstruct" them), while it per se *can not* be a problem with any means of pure interpolation (like AP and even Linear Interpolation). So, while it must be Digicheck doing something wrong here on one side, the behaviour looks very much to normal filtering on the other. Also notice that measuring devices always offer the option to highpass (like above 20Hz) and lowpass (like under 22KHz), and to me this looks like highpass above 100Hz or something, and THUS exhibiting normal filter behaviour there.

All summarized for now : don't forget 8AP is minimum to let this work properly, and that this is also audible.
Maybe I can try to show the same (via DigiCheck) with 8AP. If time permits.
But there's also the FACT that peaks-points in AP are not touched (as they are not in Linear Interpolation), so they sheer can not be lower than "doing nothing". It must be my conclusion that the FFT of DigiCheck isn't incorporating (or interpreting) the phenomenon "VRMS" very well, because it is that that only can be *higher* compared to doing nothing.

Lastly and additionally, remember that this is about periodic waves (never mind it is a sweep, because per frequency it will be periodic long enough to count as that), and while AI will benefit from that, AP is totally unrelated. And, since music isn't about periodic waves you may wonder how AI would (be able to) perform, while AP just shows reality as if it were music. That "squares" (better : transients) are (almost) fully pertained with AP, while they vanish into sines with AI is also a fact, but hard (impossible ?) to measure.

Peter

[manisandher]

Hi Peter,

Sorry, I didn't mean to give you a sleepless night!

Thanks for your explanations. Yes, it'd be great to see an 8AP sweep to prove that things below 20KHz are exactly as they should be.

4AP (QAP) still sounds great on my DAC - the clarity is amazing, no doubt due to the lack of pre-/post-ringing. However, as I've mentioned over at CA, 4AP seems to draw too much attention to the highs, which kind of detracts from the upper-bass/lower-mids, where I find so much of the emotional content of music resides.

Can't wait to get hold of a NOS1 and hear 8AP for myself. Out of interest, what is the problem with 16AP? In theory, it should work, no?

Cheers,
Mani.

[PeterSt]

A very curious bandwidth problem in the driver. I always have my answers (near) ready, but not to this one. A works as one solution for 384 ( 352.8 ), B works as another solution for 384, this one using all the bandwidth there is already, while the combination of A and B (which should bring 768) does not work because of a bandwidth problem. The only difference happens in the driver, while the master clock goes bananas of it. The most sad thing is that it is "just" over the limit (I have tested that). Something like 700 still works, but the necessary 705.6 not anymore.

But I'm doing "not thought of" things only of course.

[PeterSt]

Maybe I can try to show the same (via DigiCheck) with 8AP. If time permits.

Time permitted allright, but of course this can't be done ...
So, I was fully ready this morning, and moved the mouse to set DigiCheck to 352.8 sample rate ...
Well, you can guess the rest of the story.

Peter

[manisandher]

4AP (QAP) still sounds great on my DAC...

But I have a request: Virtually all I can hear right now are leading edges. Is there a way to change the AP algorithm to keep these leading edges but to add more 'body' and 'warmth'? Of course, I have no idea how you would go about doing this, but it would be cool...

(Incidentally, you may get the impression that I'm using Special mode with low latency. But actually my signature is correct - quite a while ago, I settled on Adaptive with 4096, and haven't touched this since.)

Mani.

[PeterSt]

Hi Mani,

No, I don't think we should do that. This would be real DSP stuff, and we should stay far from that. Instead you better try to find out what's wrong. There always *is* something wrong you know, with such kind of complaints.

Try to think of what you drew out of balance (most probably with your new amp of course) and whether there's a normal way to correct it.
Don't hesitate to think it's just the amps themselves.
Also, it is too "obvious" to think it's the speakers.

But don't fall back to bass and treble knobs ...

Peter


PS: Maybe you can describe better what you perceive, and maybe I can then hear better what is going on. Will be difficult though.

[manisandher]

Oh no, I wasn't suggesting any sort of DSP at all... no one would want that.

I'm still trying to understand why QAP applied to 16/44.1 files sounds so different from native 24/176.4 files, both recorded from vinyl. I've just got my re-wired vinyl rig back up and running and also have a much better phono stage now. I'm going to do some more comparisons between the recorded 16/44.1 and 24/176.4 files vs. the straight vinyl. My thinking here is simply that the best playback means is the one that most closely matches the straight vinyl, though I totally understand that vinyl may not really be the best source for this.

Of course, things may be totally different using a NOS1 with OAP...

Mani.

[pedal]

...Virtually all I can hear right now are leading edges. Is there a way to change the AP algorithm to keep these leading edges but to add more 'body' and 'warmth'?...

10 years ago I had some Norwegian loudspeakers (Forsman Ultima) with a SEAS soft dome tweeter. Changing interconnects to Nordost Quatrofil made a huge improvement. The sound became more "fast" and "alive". Several friends agreed, even those who were/are sceptical about cables in general. I lived happy with the Quatrofil for a year. Later I changed loudspeakers, to Magnepan 3.6. Then I found the treble to be somewhat "hyped". It turned out that the Quatrofil didnt suit the ribbon tweeters of Magnepan. Going back to some "basic no-nonsense" copper interconnects gave a more natural and better integrated treble.

(Just my 2C worth in the writing of the never ending saga of system matching.)

[PeterSt]

Mani, now I think of it ... Q2 and Q3 to 30 (or 28 as others prevail) may give that warmth you talk about. Not sure though.

[manisandher]

Thanks pedal, thanks Peter. I'll try a few different cables and also play around with the Q2/3 sliders... sorry, knobs! But I'm kind of resigning myself to the fact that maybe 4AP doesn't cut it, and that you really need 8AP...

Mani.

[manisandher]

Yes, setting the Q2/3 knobs to 28 (or 30) does reduce the leading edges... but it also 'softens' the whole sound, taking away some of the focus and tightness.

Perhaps more importantly: In the past, I've been kind of indifferent towards PeakExtend. But with my current 'ultra-fast' setup, I'm now finding that it is essential to have this engaged to keep QAP sounding 'natural'.

Mani.

[AUDIODIDAKT]

Maybe its not a bad thing to make some kind of dead-switch (dodemans-knop) in xx or mute,
if you have your dac directly connected to your amps, with digital volume control.
If something happens like static or noise i want to be in control, need to get volume down or mute very fast.

With all those expensive speakers and amps of some people here,

My question is what to do when "it" happens, what button to use, or does have NOS1 some kind of protection.
What if pc craches and there is no mute or something like that,
You will run into troubles sooner or later, what to do?

What are you're thoughts ?, whats the "panic" button.

Does NOS1 has mute for eg.

Roy