• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required as is 20 years of participation in forums (not all true). There are daily reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!

Topping D90LE - SINAD measurement (with E1DA Cosmos APU Notch)

Rja4000

Major Contributor
Forum Donor
Joined
May 31, 2019
Messages
1,823
Likes
2,181
Location
Liège, Belgium
In my previous measurements, I was peaking around 122dB SINAD measurement, with a direct ADC connection

Using E1DA Cosmos APU Notch, (and some lessons learned from last REW editions),
I was finally able to achieve a more relevant SINAD measurement of the D90LE

Measurement is done with Topping D90LE AES input, 48kHz, right Channel, 5V mode

Output of the DAC goes into E1DA Cosmos APU, 1kHz Notch, +6dB (pre-notch)
ADC is RME ADI-2 Pro fs R, 19dBu range, Mono mode
The RME is also used to send the test signal to the D90LE through AES.
Software is Virtins Multi Instrument 3.9.6.1


2022-06-27 19_07_22-Greenshot Cropped.png



Now, there are some tricks to know to achieve such a measurement !

First, we need, of course, to calibrate the chain to compensate for the notch.

Then you need to adapt the way the measurement software is computing the fundamental:
Indeed, you'll realize soon that the "foot" of the 1kHz peak in the spectrum anlyzer is wider than if you connect the ADC directly, like in the following picture.

Skirt zoom.png

This has a direct consequence:
SINAD is typically computed by comparing the Fundamental to the total signal level.
This is usually done by taking into account the width of the FFT window that is used to compute the levels for each frequency.
Unfortunately, this doesn't work here, since the fundamental "spreads" over a much wider area.
This results in a much higher Noise level, including a good part of the 1kHz signal in the noise.

Then comes AES-17 (2015), which, for practical reason, I guess, introduced a "workaround":
To measure THD+N (or SINAD), measure the fundamental over 1/2 octave below the fundamental frequency and 1/2 octave above.
For a signal at 1kHz, that means, from 707Hz and 1414Hz.
Noise is computed on the remaining frequencies only (from 20-707Hz and from 1414-20000Hz)

This is what REW is implementing with its new "AES-17 standard notch" option :
(The very last release of REW, v5.20.9, even allows to display the frequency range taken into account for the fundamental. That's genious !)

The "standard" way :
The selected region for the fundamental is the greyed frequency range.

2022-06-27 19_31_39-Analog 1 (1) on hdspfire at 48 kHz, 32.768-point Spectrum, Dolph-Chebyshev...png


And then the "AES-17 notch" way,
where the selected region spreads from 707Hz to 1414Hz:

2022-06-27 19_30_44-Analog 1 (1) on hdspfire at 48 kHz, 32.768-point Spectrum, Dolph-Chebyshev...png



This last way should be more compliant to AES-17 2015 standard
And to Audio Precision @amirm is using.

Virtins Multi-Instrument doesn't support this "AES-17 Notch" workaround natively.
But it is very flexible, so I was able, with the kind author's help, to measure and display SINAD and SNR, computed from the RMS levels in the different bands.
We then reach the exact same SINAD value than with REW.


So we have a 126.4dB SINAD measurement
 
Last edited:

Matias

Major Contributor
Forum Donor
Joined
Jan 1, 2019
Messages
3,593
Likes
6,777
Location
São Paulo, Brazil
Amazing, thanks for sharing!
 

linear diver

Member
Joined
Mar 6, 2022
Messages
20
Likes
20
Location
Colorado
Thank you so much for this write up! I just got my APU but I haven't started using it it. The timing is perfect :)

I also have a Cosmos ADC ... any reason you used the RME instead? (Edit: For the AES output?)
 

Moto

Active Member
Joined
Jan 19, 2021
Messages
140
Likes
30
Can you please detail your physical/connection setup for this measurement. That would be helpful for others trying to duplicate these tests. Thx
 
OP
R

Rja4000

Major Contributor
Forum Donor
Joined
May 31, 2019
Messages
1,823
Likes
2,181
Location
Liège, Belgium
Can you please detail your physical/connection setup for this measurement. That would be helpful for others trying to duplicate these tests. Thx

Sure.

Measurement connections are as follows:

RME ADI-2 Pro fs R as the output interface from the measurement software
RME AES Out > D90LE AES In (using the RME-included short AES cable)
D90LE Right channel XLR > E1DA Cosmos APU Notch input (usual balanced connection from XLR to 1/8" TRS jack), using a cable I had laying around.
APU Notch Pre-amp switch set to +6dB
APU notch output to XLR cable (home made)
XLR to Y cable XLR to feed both inputs of RME ADI-2 Pro fs R
RME analog input set to M/S mode (left channel contains an average of the 2 inputs, lowering ADC noise by a few dBs)

The APU notch compensation curve is computed using REW "measurement" as 192kHz, with the exact same path.
Procedure may be found here.
Note that I DID NOT remove values between 997Hz and 1003Hz like Archimago is doing,
but I flatten compensation below 20Hz and above 20kHz (although this doesn't impact measurement).

Does this answer your question ?
 
Last edited:
OP
R

Rja4000

Major Contributor
Forum Donor
Joined
May 31, 2019
Messages
1,823
Likes
2,181
Location
Liège, Belgium
Thank you so much for this write up! I just got my APU but I haven't started using it it. The timing is perfect :)

I also have a Cosmos ADC ... any reason you used the RME instead? (Edit: For the AES output?)
Well...

First reason: I own the RME ADI-2 Pro fs R.
This is an awesome device, first for sound (!!!)
and then also for measurements.

It has some serious benefits over the E1DA Cosmos ADC:

First, the E1DA ADC has a very low input impedance. Even worse when you want to use it in mono mode.
As you can see by comparing the 2 measurements above (the second, with 122dB SINAD, being done with the E1DA ADC), this may limit the output level of the DAC.
This may even become kind of a torture test for the device you want to test.
Of course, you mitigate this issue with the APU, anyway (APU has a 47k ohm input impedance. E1DA ADC has around 840 ohm input impedance for 4.5V range in mono)

Second, for some measurements, you'd better have both ADC and DAC in perfect clock sync.
This is especially true for Frequency response measurements, where some specific signal may be used if that's the case.
The E1DA ADC doesn't have a clock in or out. So this becomes more... random.

Third reason: with the E1DA ADC, I have to use ASIO4ALL, which is... random.

And then, the RME is doing everything perfectly well. And, for most cases, at least as well as the E1DA ADC. Sometimes better.

I'm also not 100% sure yet of all results with the E1DA ADC
(as an example, I see some clue of IMD at intermediate level in mono mode. To be true, this is because noise is so low, so not really a concern)

So, in my case, I was using the E1DA ADC only on specific occasions: to get the ultimate last SINAD dBs the RME couldn't reach.
That means measuring devices with SINAD >110dB or so.
Which is the exception for me.

Here, with the APU, I may use the RME and get this last level of information. And even more.
(I get the exact same measurements for the D90LE wih the Cosmos ADC, as an example)
Limits are pushed way beyond what I'll want to measure.

Actually, I should now be able to even test the RME itself.

Bottom line:
If you can afford it, the RME, especially with the APU when required, remains the top choice for measurerments, IMO.
On top of it -or before all- this is an incredible machine to play music !!!

If not, the E1DA ADC is a very good choice. As long as you're aware of its limitations.
(But that's true for any "state of the art" equipment. Even for the APX555B Amir is using)


By the way:
Here the RME ADI-2 Pro fs R
DAC tested, Mono output (Y cable to sum both channels), 13dBu range
AES-17 2015 Notch, APU notch +6dB

2022-06-27 22_14_00-Greenshot.png
 
Last edited:

Moto

Active Member
Joined
Jan 19, 2021
Messages
140
Likes
30
@Rja4000 thx for the response. On your apu notch output to xlr did you short the ring snd sleeve at the trs plug or just leave tip to 2, ring to 3 , and sleeve to 1 since ring and sleeve are both grounded in the apu?
I’m struggling with noise which is why I ask.
 
OP
R

Rja4000

Major Contributor
Forum Donor
Joined
May 31, 2019
Messages
1,823
Likes
2,181
Location
Liège, Belgium
@Rja4000 thx for the response. On your apu notch output to xlr did you short the ring snd sleeve at the trs plug or just leave tip to 2, ring to 3 , and sleeve to 1 since ring and sleeve are both grounded in the apu?
I’m struggling with noise which is why I ask.
I used TRS, so connected exactly as if the output was balanced.
It's not, but then "cold" is connected to ground in the APU itself.

I power the APU from a powerbank.
 

IVX

Major Contributor
Forum Donor
Joined
Mar 3, 2019
Messages
1,140
Likes
2,462
Location
South of China, SHZ area
I totally agree, Cosmos APU is a kind of 30x THD microscope, hence, a high-quality ADC is not required anymore, even so-so ADC is entirely enough. Any cheap audio interface like Behringer, Focusrite or Motu is ready to outperform APx555 for the THD+N/sinad or Dynamic range measurements with APU. I believe, does exist such integrated into a laptop audio input which handles THD+N -100db, hence, being cooperating with APU it will outperform APx555(residual THD+N -124..-125db).
PS: even 16bits ADC with APU provides THD+N -122db ;)
2022-06-28_07-34-57.jpg


PS: I forgot to mention that regarding SNR and DR Cosmos APU is 1000x "microscope", so, absolutely any ADC(including 16bits) with APU will outperform APx555b about 5x times.
A few bytes of philosophy:
The only key factor before you become the kingpin of the fine measurement of audio is understanding of the involved physics. Any instrument is only a continuation of our brain, when you buying the best ever microscope, screwdriver or solder iron it doesn't help if you don't know what to do next. For sure, Cosmos APU dramatically reduces your hobby-related financial risk down to the range of $100, from a ridiculous $30000, and actually, from $50K for the APx555b in China. i.e. you can try APU easily and not refuse your wife about the new Evoque. However, regardless of $100 or $30000, remember that without understanding it is wasted money ;)
 
Last edited:
OP
R

Rja4000

Major Contributor
Forum Donor
Joined
May 31, 2019
Messages
1,823
Likes
2,181
Location
Liège, Belgium
To make an analogy, it's like with a camera.

You may have a 45Mpixels camera with a set of wide aperture high resolution pro zooms, say 16-35 2.8 + 24-70 2.8 + 70-200 2.8. This is a very polyvalent and still extremely performant tools set.
That's the RME.

Then you have a cheaper, very sharp, fixed focal length 85mm opening up to 1.4 or 1.2. This will give you awesome portrait shots, it's even a bit sharper, but less polyvalent.
Say that's your E1DA Cosmos ADC.

Finally, you have a fixed very long focal length, like the 800mm f/11 Canon launched some time ago. Relatively cheap, if you compare to the pro zooms, sharp, and will give you awesome details of distant small birds, even on a lower resolution camera.
But you can't focus at less than 4m, it has a fixed aperture, and it requires some use to be able to aim accurately at the bird you want to capture.
That's your APU.

In comparison, the Audio Precision is like having all that together, and way more: the 5.6 version of the 800mm, a 11-24 zoom, and all the best intermediate wide aperture fixed focal lenses.
But on a camera 5 years old, so "only" resolving, say, 36Mp. State of the art 5 years ago. Still, in most cases, having the exact ideal lens at hand will give better results than compensating with resolution.

You may do 99% with the first set of lenses.
Adding the second may bring you significant additional money, if you're into studio or wedding photography.
And if you can't afford the full zoom set, you probably will be able to earn enough to live with just this one, if it fits your needs.
The third one is a very specialized lens, but gives you results impossible to achieve with the others.
A combination is what you'll probably need.

Finally, the AP fits all needs very competently.
But at a price...

(I know. Analogies are dangerous and misleading. But that's to give a general idea.)
 
Last edited:

IVX

Major Contributor
Forum Donor
Joined
Mar 3, 2019
Messages
1,140
Likes
2,462
Location
South of China, SHZ area
AP's engineers seem old but not stupid, they made the variable notch-filter for all audio frequencies, and there are tonnes of compromises. However, the performance level they reached is doesn't look like worth 30-50K anyway, because APx555b isn't capable to measure a cheap Cirrus Logic DAC like CS43131(Meizu HiFi DAC $13 noise 500nV(A) vs 700nV(A) of APx555b). $5000-10000 is fine, or give me something outstanding for $30000. If I was AP CEO, I would offer the same APU device in the huge and heavy case for $10000 to stop that shame ;) BTW, do you know that my AP SYS2522 20 years old analyzer has less residual noise than APX555b? Try to find the analogy from the camera-lense world for that if you can :D
 
Last edited:
OP
R

Rja4000

Major Contributor
Forum Donor
Joined
May 31, 2019
Messages
1,823
Likes
2,181
Location
Liège, Belgium
I know, from my job, that pro-level costs seem completely crazy, from an amateur point of view.

But that's because the productivity factor plays a major role in the professional world: specialized work costs may very quickly go way over the hardware or license cost.

An AP is a pro tool.
A high level specialist may spend some time building a set of test scripts (that's what Amir did, I think), and then a set of less specialized technicians may run the script for QA 24/7, with 100% reproducibility (within a tolerance), in different facilities accross the world.

With a device like your Cosmos range, an amateur, if spending enough time, may achieve similar individual results, or even better.
But reproducing the same level of productivity and coherence would require investing a very significant amount of time that, if you have to pay for it, may quickly overcome the AP's price.

And, at the end of the day, it will still not be accepted as a reference, since it's not performed with an AP, and AP is a de facto standard on the market. At least, you'll have to cover the costs to prove that your procedure gives coherent results with AP results.
 
Last edited:

IVX

Major Contributor
Forum Donor
Joined
Mar 3, 2019
Messages
1,140
Likes
2,462
Location
South of China, SHZ area
I think only hobbyists may need to prove to see the same result APx555 vs something else but 5-6x times less noisy. For an expert all that is clear at the first glance ;)
FYI, I worked with SYS2722 for about 9 years and that machine about 6 times per year forgets all internal settings due to EEPROM issues. We sent that to the local AP service to repair SYS2722. each time $1500, until I call AP USA and ask - what a f**k is this?? Next, they gave me the EEPROM_loader.exe and I "repair" the analyzer by myself with the same rate around 3-6 times per year. My SYS2522 I did repair 3 times already. P -professionalism, right? ))
 
OP
R

Rja4000

Major Contributor
Forum Donor
Joined
May 31, 2019
Messages
1,823
Likes
2,181
Location
Liège, Belgium
How come the base of the fundamental is larger with the APU ?

It's a question that I ask myself since I observed the fact

Here are some clues:

The APU is introducing a narrow notch, to cut frequencies around 1kHz by around 30dB
(more like 29.4dB in my case)

2022-07-03 18_22_26-REW V5.20.9_Notch.png



We may think that the noise in the frequencies around 1kHz will be decreased by the same amount.
But that's NOT the case in this case (with the D90LE attached as a source).
On the contrary, the noise actually increases around 1kHz


2022-07-03 18_07_05-Greenshot Noise plot 200dB.png


Note that in the above measurement, the notch is NOT compensated for.
If we did, things would be even worse (the following measurement is pure noise, without signal)

2022-07-03 18_06_45-Greenshot Noise plot 200dB Compensation.png


Of course, the noise level is VERY low.
Look at the vertical scale.
But we try to measure a device that has a THD+N level (way) below -120dB.
So can we neglect this ?


Let's zoom in a bit
I changed the scale from 200dB (what I use for SINAD measurements) to 50dBr (relative to peak)

Without compensation.
We see a raise by around 7dB around 1kHz.

2022-07-03 18_07_05-Greenshot Noise plot 50dB.png


With compensation (no signal at all, just the noise)

2022-07-03 18_06_45-Greenshot Noise plot 50dB Compensation.png


So the noise is coming from the APU notch.
If not, it would look differently.

Here is an example of a noisy device (a mic preamp with 60dB gain !) with the APU without compensation.
(Scale is 60dBr)

This is more in line with what I expected.

2022-07-03 18_49_55-Greenshot Preamp.png



And before you ask :
No, the cause is not the ADC nor the connected DAC.
I checked


So what should we do with that ?

What we see is that the APU itself introduces some (very low) noise in the measurement.
And this noise has a specific frequency profile.
Is this noise negligible ?
Not so sure.
Probably not, if, as I think, it's high enough to cause the enlargement of the base of our fundamental.

So maybe we should take that into account in our compensation curve ?
Compensate for the fundamental and the harmonics, but less so (or not at all) for the noise ?
As the preamp noise shows, that's not that simple.
And we want to measure the DAC, not the APU noise.

Well.
AES-17 way of computing noise by removing 1/2 octave below and above the fundamental helps here, of course.
But is that the most accurate measurement ?

Food for thoughts...
 
Last edited:
OP
R

Rja4000

Major Contributor
Forum Donor
Joined
May 31, 2019
Messages
1,823
Likes
2,181
Location
Liège, Belgium
AP software doesn't ignore H1+/-40% noise as REW does but in fact, Ok matches vs AP's THD+N.
View attachment 216254
Thanks Ivan
As I said above, this AES-17 notch method will most probably help making things more equal.
As you confirm, it does.

Is that the plot of the AP notch and HPF/LPF?
Can you evaluate the width of the frequency range the AP is using for the fundamental (H1) measurement ?


What do you think about the APU noise profile ?
As I wrote, level is very low, so it is only a question for measuring device with extraordinary low noise.

Does the AP also see this enlargement of the 1kHz peak with the notch ?
 
Last edited:

IVX

Major Contributor
Forum Donor
Joined
Mar 3, 2019
Messages
1,140
Likes
2,462
Location
South of China, SHZ area
The theoretical APU THD+N's Noise-related limit is <-134db, I did mention that in the discord. The practically achieved number is -132db, I've no sine source better yet. I just finished the Matlab simulation of the THD+N with an ideal 1kHz sine + real APU noise with exactly the same notch+20-20kBW as AP2700 SW does. As you can see, a noise-limited THD+N value is [email protected]@10Vrms(APU Notch gain 0db) i.e. about 10db better than APx555b ;)
2022-07-04_16-21-37.jpg
 

IVX

Major Contributor
Forum Donor
Joined
Mar 3, 2019
Messages
1,140
Likes
2,462
Location
South of China, SHZ area
Oh, I have to mention that my ADC sensitivity was 2.7V which corresponds to about 8.8Vrms APU input at 1kHz(+20db gain -30db notching = 10db or 3.16x times), hence, 10Vrms input has about 1db room. The correction should subtract .5 to 1db from the THD+N simulation above i.e. the limit is around -135.5db.
 
OP
R

Rja4000

Major Contributor
Forum Donor
Joined
May 31, 2019
Messages
1,823
Likes
2,181
Location
Liège, Belgium
Oh, I have to mention that my ADC sensitivity was 2.7V which corresponds to about 8.8Vrms APU input at 1kHz(+20db gain -30db notching = 10db or 3.16x times), hence, 10Vrms input has about 1db room. The correction should subtract .5 to 1db from the THD+N simulation above i.e. the limit is around -135.5db.
Amazing.
In another life, I'll start playing with Mathlab.

By the way, it seems your theoretical noise profile is matching the one I'm seeing.
Or did I get it wrong ?
 
Top Bottom