Best spec ADC Chip currently.... ??

[IVX]

MC_RME, we gonna make our Windows app and I believe all related problems are trivial My app will 100.0% match vs AP. Maybe this weekend I'll start to play Matlab to realize the potential issue.

 

[msmucr]

@IVX I didn't think about ARTA as a general alternative, but just to check input from the ASIO driver at the same setup.

From my point of view, REW has one big advantage - it's multiplatform. I've successfully used is also at macOS. And as I like ARTA, its development stalled, I've tried to contact Ivo couple times over years about small refinements, but no avail.
Also there's lot of great functionality in REW, I was super surprised, when I tried it last year or so (it was always somewhat under my radar, because I thought it's focused primarily on acoustic measurements). So it's IMO worth of making it work with the hardware (and ASIO driver). I'm not sure, if it makes so much sense to devel own accompanying application, unless there will be some super unique functionality over multiplatform REW.. but of course you know your plans that better than me

Anyway, I've made few tests at the right computer
ARTA and REW comparison with generated symmetric distortion.. THD+N readout matches well between programs (0.000063 to -124.01 dB), if I calculated it correctly.
Also other REW plots just to show resolution of its measurements with different inputs and windowing functions.
I've used virtual ASIO driver with 32 bit float path and for comparison with interface also RME card fed by the generator with 24 bit TPDF dithered sine. It's as good as it gets..

Michal

 

[IVX]

msmucr, huh, -200db noise floor makes that test clearly THD-N but THD+N

 

[msmucr]

msmucr, huh, -200db noise floor makes that test clearly THD-N but THD+N

Yes, that was intentional for those plots. I just wanted to demonstrate, it resolves and measures without skirting, jumping, level discrepancies etc. and that if you're fine with ARTA calculations in comparison to your AP, you could be fine with REW too

I've mixed some noise with generator for comparison of results between ARTA and REW in those two plots. And also defeated my default 20k LPF in REW to have minimize any differences, as ARTA uses whole passband up to fs/2 for level calculations (unless you enable weighting of course).

 

[KSTR]

I can't comment on "many other software", but that is not true for REW. REW's calculation of the power in the fundamental takes into account the spreading span of the window and includes the areas of the skirt that are monotonically decreasing. I would nonetheless recommend a low sidelobe window for measurements that have very high dynamic range.

OK, I stand corrected (and just checked) though I think I had issues with this in a very early version...

 

[KSTR]

In general, do we need THD+N at all? I never use it. I prefer to look at true THD (sometimes even with noise reduced by sync'd averaging) and noise floor seperately. One has to do that anyway otherwise you won't find an elevated noise floor with a signal present (like from Jitter).

 

[IVX]

if you're fine with ARTA calculations in comparison to your AP, you could be fine with REW too

Ok but 24db of difference is slightly too HUGE to me
Just updated the REW, the same 20+db mismatch to AP. I see no any settings there which I didn't try to change. The FFT itself looks normal but calculation..

 

[IVX]

https://yadi.sk/d/wjvTTqENQDSAiA this is a 24/48 record from WASAPIAP andArta shows about the same THD+N -118.3db(ES9822 stereo mode, I can make a mono record as well with -120db but I think it is not critical) and 0.00012%. AP's BW 22-20000Hz, RMS detector type, the level is -.286dbfs also matched Arta.

 

[JohnPM]

Pretty noisy at low frequencies. The THD+N figure will depend on the choice of where to draw the line on what constitutes the fundamental and what is signal-related noise provoked by the fundamental.

 

[JohnPM]

As suspected, the main factor in THD+N is how much of the surrounding of the fundamental is removed. AES-17 specifies its "standard notch filter" rather loosely, requiring a Q range of between 1.2 and 3 which is a bandwidth of between 1.17 and 0.48 octaves. Using 1 octave gives a THD+N result that ties up with the AP.

 

[JohnPM]

My case is AP reads steady THD+N -118db, REW -90db with +/-50% fluctuations, and amplitude is -3dbfs instead 0dbfs.

The -90 dB result is probably due to using Java drivers on Windows. The Windows Java runtime only provides 16-bit audio data, use ASIO drivers on Windows to get full sample resolution.

 

[msmucr]

Ok but 24db of difference is slightly too HUGE to me
Just updated the REW, the same 20+db mismatch to AP. I see no any settings there which I didn't try to change. The FFT itself looks normal but calculation..

Hi,
I see and thanks for the captured audio clip.

I briefly looked at that and discrepancy looks much more clear to me.
The signal has quite high level of infrasonic noise bellow 20 Hz and observed higher "skirting" of fundamental tone isn't there because of FFT windowing function in analyzer (see the previous examples with pure digital sine for a comparison), but most likely because of generator itself (oscillator modulation by low frequencies) or happened during capture with the ADC (higher LF jitter).

I've came to similar conclusion like John above.
The notch filter used for THD+N measurement in ARTA and AP analyzer (according to your readouts), is apparently wider than the one used in REW (plus Wavespectra and some of my other tools, which also gave me readouts with similar higher noise level around -90dB).
So the wider filter is more efficient with attenuation of low level skirting, which surrounds the fundamental.
Additionally that infra noise also plays role there. ARTA has 20 Hz HPF for THD+N in its default setup (screenshot).

Finally I've simulated it in a DAW with EQ and FFT analyzer with RMS readout.

12dB notch Q=1 for fundamental, readout was cca -90 dBFS
12dB notch Q=1 and 12dB/oct HPF at 20dB, gained 7 dB lower readout
24dB (stacked) notch Q=1 combined with previous HPF, result was over -118 dBFS, so it aligns with your previous results.

So thankfully no magic happened there And it was also enlightening for me, because I previously didn't have practical examples of discrepancies between notch filter designs among different analyzers (be it software or hardware), which can apparently play significant role with analog signals.

Michal

 

[KSTR]

The -90 dB result is probably due to using Java drivers on Windows. The Windows Java runtime only provides 16-bit audio data, use ASIO drivers on Windows to get full sample resolution.

Or just throw a recorded file on the RTA window ;-)

 

[ksaugrin]

If we could have a APx555 like ADC to do some DIY measure i will buy for sure.

 

[Spocko]

HpW matches the AP THD+N values near perfectly. There are some simple reasons for that once you know them. Unfortunately I can not disclose these.

As a consumer who does not want to overthink it, I always look to @MC_RME to see if they have a product that meets my needs first. I needed a solid PCI card engineered to the same level as their desktop DAC/ADC and the RME AIO HDSP Pro has not disappointed. Highly recommended if you need well implemented ADC.

 

[IVX]

JohnPM, it is not about Jawa, I used ASIO4ALL v2 and 20-20k REW BW limit was turned On. I did filter up AP's output with Sallen-Key LPF to get low as possible harmonics level to accurately compensate ADC's distortion, at 1Hz that LPF has quite high impedance. Regarding the copy of an analog notch, I thought last night too. I'm sure that AP had to keep that trick even in a digital domain to avoid a mismatch with pure-analog legacy analyzers. Even if that implementation isn't 100.0% correct, AP has to repeat that notch in all new models. And msmucr's experiment confirms that as well.
PS: I did repeat your settings but after 500 AVGs got -114db, AP reads today(slightly lower level about -1dbfs) -118.8db.

 

[JohnPM]

You don't have the code I used, so you won't get the same answer. I'll send you a link to a test build if you'd like to try it.

 

[IVX]

JohnPM, yes, now it works I confirm that. Actually, 20, 100, and 400Hz(-119db, I'm sure if I'll add HPF before my LPF, 20-20k BW will give me -119db as well, hence mono mode 121db, will test it now) BW limits are also matched. Very good!
However, we anyway starting our own FFT app for Windows because we gonna offer additional software tools for ADC and DAC THD compensation, stereo mode, and perhaps superfast measurement mode by convolution. I made a relay-free front-end, and I can try to get some speed profit on it. Let's see..
EDITED: in the mono mode, I see a mismatch again. AP THD+N -120.5-120.9db, REW -119.1-119.5db. And SNR also about 1db mismatch, AP shows about -127db(A) REW -126db(A), didn't test unweighted SNR yet but I don't believe if A-curve may have a noticeable error, right?
Oh, no, sorry, my mistake, I did mix up something and will remeasure again!

 

[somebodyelse]

Will you have enough documentation so that others on non-windows platforms can implement similar things?

 

[IVX]

I retested that, seems REW shows a bit better THD+N(something like -.3db.) at -120db level but SNR A-weighted wise versa .3db(A) worse than AP. Anyway, it is a good result. Actually, today is a bad day for the Cosmos proto, almost 1db worse SNR and THD+N in the mono-mode, probably, it needs to be warmed up 1 hour to get 127db(A) and slightly less than -120db THD+N.