@matt I just got a Topping E70 and started to do some measurements on it. Looking at Audioscience Review and others, the 1Khz numbers are around -123db thd+n and thd close to -140db.
I tested the E70 directly into a Cosmos ADC at -135db thd and -122.9db thd+n at 5vrms out.
I wanted to see what things would look like on the QA402 and went back to the specs doc on your site for the QA. There is page there titled Plot:THD+N Spectrum from External Source where you are using the oscillator from the QA480 which does not appear to have specs any better than the E70 as far as just the basic oscillator( of course the QA480 has all kinds of additional functionality) so it seemed as if I should be able to get close to the numbers you showed in that plot using the E70 as an external source. The distortion at 14dbv out from the E70 and 18dbv input setting on the QA402 just blows up the QA402 ADC distortion. That distortion is not in the E70 based upon my and others testing.
What am I doing wrong?
I just realized you may have been using the notch filter for this plot. Is that what was happening?
Itâs hard to know from a single plot what might be going on. You nudge the level down 1-2 dB does the spectrum change appreciably? Also, what do the harmonics look at like when you switch to +24 dBV full scale input?
The plot you reference isnât using a notch.
Also, double check your loopback performance at that operating point (18 dBV full scale with 14 dBV input balanced). I just ran it for the QA402, QA403 and QA404. You can see the 2H are up quite a bit on the QA402.
QA402 THDN: -88.78 dB
QA403 THDN: -105.67 dB
QA404 THDN: -106.84 dB
You will for sure need a notch to measure.
Here is QA404 at that level (single ended) being driven by QA480 (-112.73 THDN):
Thx @matt. My qa402 loopback is -89.7db thd+n.
Shouldnât I be getting much better numbers from the Topping E70 based on what you get with the qa480 oscillator?
I did try both balanced and se from the E70 into the qa402 with very similar results. I also tried 24db and 30db which showed somewhat improved thd but of course worse thd+n. At the 24db input setting I got thd -99db and thd+n -95db.
I also tried reducing the E70 output by 1, 2, and 3db also with just somewhat improved thd, but nothing like your plot.
I will run the Cosmos apu into the qa402 as well.
Hi @Moto, so your loopback figure is where it should be. Which QA480 into QA402 figure are you referring to? That was probably done with a notch since the QA480 has a notch. Do you have a notch you can try?
I do. The thd+n numbers you showed in this thread were very similar to the ones in the doc on your website I referenced in the first post and you said that was without a notch. Thats what had me perplexed.
Here is the E70 into the Cosmos APU into the QA402. 24db attenuation gave better results on harmonics than 18db attenuation by a wide margin and only a few 10ths worse SNR. I can see that if the QA402 were in mono that the thd+n would be -125db ( as in the mono mode Cosmos ADC)
instead of the -122db we see here. So the results are comparable to the ap555x.
Thanks for sharing, @Moto. The chipset guys are bonding channels to get the noise down. But the cross correlation lets you do the same with lots of measurements. So, if you bond two HW channels to get the noise down (aka mono mode), thatâs akin to two cross correlation measurements. And then 4 cross correlation measurements would be like having 4 bonded conversions. But cross correlation can obviously go much higher than bonding if you have enough time.
I ordered a Topping E70. Those numbers from that DAC pretty much smash a 1 kHz dedicated analog oscillator like the QA480. Plus, you get any frequency 
Looks like the days of analog oscillators are finally coming to a close.
@matt, the Cosmos ADC switches to mono in software.
The change in that adc using mono rather than stereo seems much larger than going from 1 to 2 averages in cross correlation. Iâll do some more investigation though.
Given the performance of some of the new dacs, getting pc mirroring back into q40x would be great.
Did some testing. With the E70 going directly into the Cosmos adc, thd+n is improved by 2db when going from stereo to mono using their software switch in Windows. This fits with what should theoretically be a 3db reduction in noise going to mono.
However when I run the E70->Cosmos apu notch->Cosmos adc, there is virtually no difference between stereo and mono. Hmmm. Why?
The improvement in thd+n when using cross correlation in QA402 going from 1 to 2 acquisitions is only about .2 db.
I also took another look at the aes17 notch in REW when using the Cosmos apu notch filter.
Using the Cosmos apu notch and E70, thd -136.9db and thd+n -118.5. If I add the aes17 notch in REW, thd -136.6 and thd+n -125.8. So a whopping 7db improvement. Hmmm.
Other measurements of the E70 on an AP555 are thd+n -123.5. Another measurement with an AP555 was thd+n -123db. Some comments are that this is around the limit of the AP. These are all at the 4vrms 0dbfs setting of the E70. There is also a 5vrms setting which I havenât tested yet.
So it seems like there should be some compensation for the wide skirt when using a 1k notch filter, but is the aes17 compensation too much?
As a last note, why is there so little widening of the skirt when using the Cosmos apu 1k notch in the QA402 plot vs the Cosmos adc plot in REW using the same 128k fft?
In all cases, summing wonât buy you anything once youâve hit the limit of the DUT.
As a last note, why is there so little widening of the skirt when using the Cosmos apu 1k notch in the QA402 plot vs the Cosmos adc plot in REW using the same 128k fft?
Can you share a plot? Are the same windows being used?
For instance this is 128k. The others I have with REW were 64k. Yours above with the wide foot on the QA404 was at 64k. I am in a different city from my QA402 and QA403 so canât retest now.
Hi @Moto, Ah, OK I get your question now.
The QA480 notch (and also the notch used in the Measuring a Modern Class D amp thread) is a very low-Q notch (Q=1). Which means it has significant contributions at 2k and 3k. The reason for this was reduce sensitivity and to avoid falling deep into the notch as room temperature changes. But it also means that compensation has to happen at 2k and 3k (and 4k) for the notch.
Here is the balanced notch being swept. Note the 9 dB at 2 kHz.
So, if you go through the process of sweeping your QA480 notch, and then apply the correction while still doing the sweep, your sweep will appear as below (note this is zoomed in on y axis). This is also why itâs so important to correctly identify the frequency of the test toneâbecause very tight âfiltersâ are being placed around the harmonics.
In the response, note how narrow the peaks areâthis is where the frequency selectivity comes in and also why the skirts are non-existent. And this side-steps the various problems with skirt widths, AES bandwidths, etc.
This is the cosmos apu notch which is much steeper around 50.
So this should lead to a wider foot than the qa480?
So this should lead to a wider foot than the qa480?
If the notches were applied equally, then the QA480 notch would have the wider skirt because it is the lower Q. But the compensation applied by the QA40x software (in the file->Export->notch weighting) creates the very narrow âfiltersâ you see in the picture above. Look in the exported weighting file created by the QA40x software. From 0 Hz to perhaps 970 Hz or so, there is no gain modification. So the skirt of the notch is ignored. And then at 1 kHz youâll see perhaps 40 dB of gain applied (a negative number), and then at 1.03 kHz itâs back to 0 dB of gain applied.
Maybe one way to think about it is the QA480 hardware is a low-Q filter, but after processing the HW/SW âsystemâ becomes an very high Q filter: That is, thereâs no gain adjustment at 950 Hz, but 40 dB of gain adjustment at 1 kHz. And as a result, no skirt.
Thx. I see. What causes the wide skirt in your earlier post of the qa480 driving the qa404?
Because itâs an analog oscillator?
Sorry. Should have realized that.