Measuring RME ADI-2 Pro with QuantAsylum QA401

[Miska]

Anyone want to draw the input circuit

View attachment 23338

If you want clean audio band, you need to have >144 dB attenuation above Nyquist...

Looks like AKM ADC chip. Have you checked the data sheet how it's digital filter looks like at the sampling rate you are running at? The analog filter looks maybe 3rd order. And if you want fairly flat phase response in 100 kHz bandwidth it means fc around 400 kHz or so. Or if just for 20 kHz then fc around 100 kHz.

Then again, many ADC and DAC chips also exhibit noise modulation anyway... So could be that too.

 

[March Audio]

Well ordinary audio interfaces seem to manage quite well without significant noise floor modulation so it's quite acheivable.

Matts information shows only a few dB change in noise floor which matches the AKM data, so something is amiss. I'm sure the digital filters are adequate. Whilst I'm interested in hearing what the ultimate problem is, I don't have time to persue it.

 

[QAMatt]

If you want clean audio band, you need to have >144 dB attenuation above Nyquist...

Looks like AKM ADC chip. Have you checked the data sheet how it's digital filter looks like at the sampling rate you are running at? The analog filter looks maybe 3rd order. And if you want fairly flat phase response in 100 kHz bandwidth it means fc around 400 kHz or so. Or if just for 20 kHz then fc around 100 kHz.

Then again, many ADC and DAC chips also exhibit noise modulation anyway... So could be that too.

Hi Miska, it's a second order filter on the QA401--pretty much the same filter used by AKM in Figure 41 of the ADC spec and also on the AKM eval board.

The tradeoff here is CMRR. The more components you have in the path between in the input and ADC, the more you hurt CMRR, and CMRR is important to THD, especially in certain class D applications. So, yes, you could build a higher order anti-alias filter but you'll degrade other things that matter. There's no getting around that unless you have a lot of hand tuning on each board. But then you move to a much higher price tier, routine calibrations, etc.

And 105 dB of rejection is actually very workable. You can take a 300W class D amp, short the input, and run that amp's output straight into the QA401 inputs. Remember: even though no sound is playing, the amp's output is still slamming +/-50V before pumping through the output LC--and you'll measure very close to TI's published noise numbers for that amp (85uV A weighted, 20 to 20 KHz) in spite of TI using an AES17 filters.

A really quiet headphone amp might have noise at -110 dBV (unweighted, 20 KHz). To impact that measurement via aliased noise means your out of band noise would have to be nearly 1Vrms at a few hundred KHz. Or maybe 100 mVrms if you are looking across a hundred KHz bandwidth. If someone is building a product with that much noise at the higher frequencies (but dead quiet at audio frequencies), something is really, really wrong with the product--you could see the noise easily on a cheap oscilloscope.

PS. I think all (?) >20 bit audio ADC and DAC exhibit a rise in N+D as the output level increases near the upper limits. If they didn't, there would be no "valley" in the THD+N output curves. I think the particular issue here is whether or not the rise was excessive in the OPs QA401. In some DAC products, you can set your max output to be 10 dB below the 0 dBFS and avoid the valley (early RME ADI had 0 dBFS set at ~17 dBu and +4 dBu max I vaguely recall), but the core converter itself has the valley.

 

[Miska]

OK, I'm a bit curious because those plots with QA401/AK5397 seem to have some similarities with ADI-2 that has AK5574 which seems to exhibit somewhat similar behavior what I've seen, but has 85 dB stop-band attenuation.

 

[edechamps]

I tried to reproduce the "rising noise floor" phenomenon today with my QA401 in loopback. Here's the difference between ~-28 dBFS and -~-1 dBFS:

If I understand correctly, what @March Audio is worried about is the difference in noise floor that I annotated with a red arrow.

Just for fun, I took the opportunity to design a custom THD+N vs. output level measurement procedure and I ran it against the QA401 in loopback:

The above is THD+N. If we only look at absolute N+D, then the "rising noise floor" is extremely obvious:

We're still looking at around -98 dB THD+N when running near the top of the scale, though, which is good enough for many use cases.

 

[amirm]

We really need to test the QA401 with a high performance analog generator rather than its own DAC. My AP has both options. Here is the difference:

As you see with the red curve, you get much better performance with the analo generator as levels get stronger. I hope to test the Q401 at some point which I get some breathing room.

 

[March Audio]

I tried to reproduce the "rising noise floor" phenomenon today with my QA401 in loopback. Here's the difference between ~-28 dBFS and -~-1 dBFS:

View attachment 23778

If I understand correctly, what @March Audio is worried about is the difference in noise floor that I annotated with a red arrow.

Just for fun, I took the opportunity to design a custom THD+N vs. output level measurement procedure and I ran it against the QA401 in loopback:

View attachment 23781

The above is THD+N. If we only look at absolute N+D, then the "rising noise floor" is extremely obvious:

View attachment 23782

We're still looking at around -98 dB THD+N when running near the top of the scale, though, which is good enough for many use cases.

Thanks thats good information. My experience was that as you approached 0dB the noise floor (with a 32k FFT for example) would sit around -125dB with pretty much any DAC that I attached to it. By -60dB you will be down to below -150dB

Typical example here.

https://www.audiosciencereview.com/forum/index.php?threads/audio-measurement-gear.113/post-62413

 

[Deleted member 2944]

We really need to test the QA401 with a high performance analog generator rather than its own DAC. My AP has both options. Here is the difference:

View attachment 23784

As you see with the red curve, you get much better performance with the analo generator as levels get stronger. I hope to test the Q401 at some point which I get some breathing room.

I would be glad to loan you mine. I live close.

Dave.

 

[amirm]

I would be glad to loan you mine. I live close.

Dave.

Thanks Dave. Company has also offered one but yours may be more convenient.

 

[Blumlein 88]

I tried to reproduce the "rising noise floor" phenomenon today with my QA401 in loopback. Here's the difference between ~-28 dBFS and -~-1 dBFS:

View attachment 23778

If I understand correctly, what @March Audio is worried about is the difference in noise floor that I annotated with a red arrow.

Just for fun, I took the opportunity to design a custom THD+N vs. output level measurement procedure and I ran it against the QA401 in loopback:

View attachment 23781

The above is THD+N. If we only look at absolute N+D, then the "rising noise floor" is extremely obvious:

View attachment 23782

We're still looking at around -98 dB THD+N when running near the top of the scale, though, which is good enough for many use cases.

Yes that noise floor change. For comparison here is a March DAC1 feeding the ADC in the Zen Tour from Antelope audio.
Blue is maximum level at 1 khz. Red is -18 db. Green is -60 db. You see distortion rise at max. You see some higher orders rise at -18 db which I wouldn't expect. But the ADC noise floor is not effected by the level changes in the signal.

 

[QAMatt]

Yes, agree, edechamps, the N+D is the correct way to look at this. I've attached the N+D plots for both the AK5397 and AK4490. On the QA401, both ADC and DAC are 0 dBFS @ = +6 dBV, so if you connect in loopback, and then drive to 0 dBFS, then both are at the same point of operation, which would be at the very far right on the AK plots. In this region, according to AKM, there could be as much as a 29 dB rise in N+D. Some from harmonics, and some from noise. You'd see about 23 dB of that from the ADC, and 6 dB of that from the DAC.

But notice as you drop to -5 dBFS, the combined N+D degradation drops to just a few dB.

Blumlein 88, this behavior is in all sigma delta converters to some degree...the way you "hide" this in an audio DAC is you scale all incoming digital data by, say, -5 dB in the driver before you feed that to the hardware. In other words, if you send it a true 0 dBFS signal, the DAC driver would give it -5 dB of gain, and then it would output a -5 dBFS signal when you told it to output a 0 dBFS signal. And then you call that new lower signal 0 dBFS. And just like that, the DAC (or ADC) exhibits no growth to N+D at 0 dBFS. But in doing so, you throw away dynamic range/noise floor/etc....it's all a tradeoff. The QA401 does some of this (and thus avoid the full 29 dB penalty shown in the graph), but preserving the -115 dBV noise floor is important.

AmirM, I'm always happy to drop a unit in the mail on short notice when you get a time slot opened. Just let me know

 

[proj964]

AmirM, I'm always happy to drop a unit in the mail on short notice when you get a time slot opened. Just let me know

amirm Hope you can take Matt up on his great offer!

 

[amirm]

AmirM, I'm always happy to drop a unit in the mail on short notice when you get a time slot opened. Just let me know

Thanks. If you can spare one for 2 weeks or so, I can take it anytime. Hard to make a schedule for one as stuff keeps arriving and piling up.

 

[QAMatt]

Thanks. If you can spare one for 2 weeks or so, I can take it anytime. Hard to make a schedule for one as stuff keeps arriving and piling up.

Awesome, and no rush. It should go out this week.

 

[proj964]

Awesome, and no rush. It should go out this week.

Its been over two months since QAMatt's offer. amirm did this ever arrive? Analyzed? ...

 

[jackenhack]

You only need to terminate the unused input when doing unbalanced measurements. So remove the 50 ohm terminator on the input.

 

[jackenhack]

And when doing balanced, you don’t need to terminate at all.

 

[QAMatt]

Its been over two months since QAMatt's offer. amirm did this ever arrive? Analyzed? ...

Looks like it went out on April 19. Amir has explained he has a lot of stuff in the queue, which I completely understand.

 

[restorer-john]

Looking forward to this.

 

[amirm]

Its been over two months since QAMatt's offer. amirm did this ever arrive? Analyzed? ...

It arrived. I ran a few tests which I did not understand and ran it by Matt. Then the backlog of gear attacked me and I have not been able to get back to it. I will get to it soon.