Mono and AES-17 Notch

@Matt, a couple questions.

  1. Is there any way to mono the inputs to reduce noise as in the Cosmos ADC?
  2. Can we get the equivalent of the AES17-2015 Notch that REW uses?
1 Like

Hi @Moto

There’s not a way to currently use mono mode to reduce the noise. But you can use the cross-correlation mode, which is even more powerful.

Below is THDN of notch on QA404:

Same setup and settings, except this time run the same notch signal into L and R channels. The cross correlation will do the vector average on the left and right channels. As you accumulate samples the same signals present on the left and right channel will be made stronger, and the different signals on the left and right channels (self noise) will be average to zero. What is left is the noise of the DUT–the analyzer noise gets averaged away.

Below you can see over 134 acquisitions in cross-correlation mode the noise (N-D) in the upper plot went from -101.2 to -106.7, which is about 5 dB. And the THDN went from -112.62, which is about 3.6 dB.

The AES17 technique shouldn’t really change the measurement. The AES17 spec says to use the total power in one octave around the fundamental as the reference level. Otherwise, the power in the fundamental is normally used. If your signal level is 30 dB or more above the noise, the distinction shouldn’t matter. Also, for higher frequency tones, the one octave notch is a lot to take out and tends to be give a result that is too optimistic.

Thx. I’ll try the cross correlation.
Re the aes17 notch, in REW thd+n for the E70 into the Cosmos apu notch and then the Cosmos adc worsens by 5db when you turn off the aes17 notch feature. This doesn’t happen when you don’t use a notch filter like the apu. The reason is that a notch filter dramatically widens the skirt around the fundamental and the aes17-2015 compensates for that. From various forum threads the adjustment in REW seems to be comparable to the AP555 results.

I’m not sure I follow: Why would a notch widen the skirt? Can you share a picture?

  1. E70 directly into Cosmos ADC
  2. E70 into Cosmos APU notch but no aes17 adjustment in REW
  3. E70 into Cosmos APU notch but with aes17 adjustment in REW

OK, I see what you are talking about. The AES17 doc prescribes a one octave measurement around the fundamental to determine the fundamental energy. If you use a notch, that means the fundamental is very suppressed, and so the skirt noise starts to make the fundamental higher than it is (because you are including skirt noise in the level/power calc). In other words, AES17 confuses the measurement.

So, to combat this, REW allows you the ability to NOT measure per AES17, which in that case looks at the main lobe only instead of the octave around the 1 kHz.

This is the default for the QA40x. That is, the QA40x ONLY looks at the main peak, and ignores the skirt energy in the one octave swath. In other words, the QA40x functions as if the AES17 is disabled on REW.

Yes so you are saying that qa40x is equivalent to REW without the aes17 notch feature checked, right?
That is why I was wondering if we could get the aes17 notch feature added to qa40x in the future.

That is why I was wondering if we could get the aes17 notch feature added to qa40x in the future.

If AES17 were the default, then the result for the nominal measurement (strong fundamental) would be the same as it is today. And for a weak fundamental it’d be erroneous (which is why you need the ability to turn it off).

In other words, what does the AES17 buy you?

In a few forums over the last year there has been discussion about whether using the aes17 notch made results more similar to APx measurements. I guess the conclusion was yes. So I guess its useful if for nothing else than being better able to compare to published results from APx systems.
Not sure why John decided to include it in REW software.
Also in some ways if you don’t use the notch when the skirt widens, you are including a lot of energy from the widened skirt in noise, right? So that makes thd+n look worse than it perhaps should be.
Of course as you point out using the notch adds excess energy to the fundamental when the skirt is wide.

Here is a discussion re the REW aes17 notch and Audio Precision measurements.

Here is a discussion re the REW aes17 notch and Audio Precision measurements.

Hi @Moto, I think these are different measurements. The REW measurement fundamental at -6.42 dbFS at the link, and THD+N at -123.4. So, this suggests the N+D is -124.4 - -6.42 = -130.8. That figure is very close to the displayed N+D of -130.6 dBFS A.

However, note the A, which suggests A-weighting (but maybe I’m wrong). For most measurements, A-weighting will give you about another 2.5 dB reduction in N+D. But THD+N is normally an unweighted measurement.

I’m pretty sure that AP isn’t using A-weighting for THD+N measurements. AES17-2020 specifically says “and the unweighted level of the residual shall be measured” in section 6.3.1.

Can you confirm your understanding?

From the author of REW,

Post #2
N+D is A weighted thd+n.

So messed up with aes17 notch compared to the normal couple points difference we expect to see.
I was more focusing on IVX’s comments in the audioscience post #16 and later about the similarity of the AP results and REW with aes17 notch when looking at 1khz.

OK, thanks. I got the E70 Velvet. What do you see as the noise floor on this? With E70 hardware set to 0 dB and REW set to -1 dBFS, I set peak to -1 dBr:

Then set the smallest tone possible (-144.4 dBFS) and measure noise total noise (20k) at -126.76dBr. The tone at -153 dBr isn’t contributing anything here, but it does indicate the DAC isn’t sleeping.

And then short inputs to external gain box/notch (from class D thread) and noise drops to -130.65 dBr, demonstrating about 4 dB of margin wrt noise.

At -100 dBFS, N-D is still -128.6 dB

And using notch N-D remains at -127.65, with THD+N at -121.6 dB

Hi @Matt. I noticed that you are using version 1.170 of the QA40X software. I noticed that in the “Measurements” area there is also the possibility to choose the “N+D” measurement. In my version I’m using (also QA40x version 1.170), I find the possibility to take the “N-D” measurement, but I don’t find the “N+D” measurement (for me it would be very useful to be able to view it directly without having to calculate it). Where am I going wrong? thank you

@matt here is a plot from my other post.

This is using the cosmos apu Q=50 notch.
So similar results to you and a narrow foot.
As you can see from my other plots when using REW with cosmos apu and cosmos adc, the foot is wide. Not sure what the difference is between REW calcs and yours.
Here however you show a wide foot with the qa480 ( from your previous posts)
Interestingly the distortion components in mine are much lower than the ones you show with the e70 even though thd+n is essentially the same.

The release of 1.171 (coming soon) has the ability to export the captured waveform as a WAV (and it also has PC mirroring). You can take a measurement, export the WAV, and then load it into REW or other app if you wish to compare.

Now, the plot below isn’t an exact comparison because the export is time domain, but the averaging done on the QA40x is frequency domain. And so, the freq domain display is very averaged, but the exported WAV is not. A better comparison would be to export a 128K acq without any averaging, and then compare that. But, as a first comparison, note the skirts are very similar here. The QA40x THDN is 1 dB better, but freq domain averaging tends to help there. The N+D are close, however REW should be doing A weighting on that and I’d expect the REW measurement to be lower there.

Another way to look at is that if N+D is -115.8 dBFS with A-weighting, and if the noise is flat (which it is), you’d expect the unweighted N+D to be about 2 or 2.5 dB higher, and thus the THD+N would be around -113.0 or so.

Great. Look forward to this!

Release 1.171 with mirroring is up.

Hi @Matt. I tried version 1.171 of QA40X and found it very interesting. For me it is very useful to have the possibility to display the “N+D” measurement without having to calculate it every time, if necessary. The ‘Export as Wav data’ option is also very useful. It allows me to save a measurement session and then, by reloading it into QA40x itself, to be able to study it at my leisure using the QA40x tools. For me, this is a big step forward! It is very useful to have the very powerful Blackman-Harris window available rather than the little-used Bartlett window. In short, in my opinion, this version of QA40x is a great improvement. Thank you