Topping D10s + E1DA Cosmos ADC as a measuring system

[pkane]

@pkane : Paul, how would I synchronize REW from 2 PCs? I need REW to make THD and THD+N vs. level and frequency measurements.

Don't think you can do sweeps with REW across two instances on different computers, you're right. This should be doable with Multitone, but I'll have to think about it

 

[somebodyelse]

Wow! That really is a bargain! But it appears to be only full-speed, so may not work with some DACs and at higher sampling rates.

It's similar to the other widely available ADuM4160 based isolators. We can hope they'll step up to the ADuM4165/4166 soon, or do something similar with TI's ISOUSB211. Until then TI's eval board is ~£50, a bit less than AD's equivalent.

 

[IVX]

Guys,
a) you can try any old and battery-powered laptop to isolate DAC or ADC to avoid GND loops.
b) If you are more a practical DIYer than hig-hend-perfectionist, you can try to use neighbors USB ports placed closely on the PC's mother-board i.e. reduce the GND loop effect.
c) does exist very inexpensive USB HS isolator which works with DACs at least https://aliexpress.ru/item/10050019....store_pc_groupList.8148356.25.30113767EIB9Hs
PS: Pavel, tell me pls why don't believe my Cosmos ADC AUX input?
It works up to 43Vrms and utilises 2pcs in serial(!) a high quality Viking MELF 0207 resistors.

 

[linear diver]

This is why I use two PCs: one driving the DAC the other (battery-powered laptop) -- the ADC.

The magic ingredient! I get much better results when I run the D10s off my old Mac laptop. It is also fun to see the effect of AC and DC power. I have not seen much about using the Cosmos with Mac. At first glance, it kinda works, but I need to investigate further.

For any serious work, you will need the USB isolator. I gave a link and that unit works with Topping D10s.

I lucked-out on this one, but I have a hunch the USB isolation issue will continue to rear its ugly head. How does the USB-ISO compare to the Topping HS01 for measurement purposes? I live in the US, so I might get the Topping faster.

Please also check this image
https://www.audiosciencereview.com/forum/index.php?attachments/e1da_dividers-png.211407/ carefully, the RCA - XLR cable between D10s and Cosmos must be shielded twist pair where the -sig wire from RCA body to XLR pin 3 serves to subtract interference voltage across the cable shield due to ground currents.

Here are my latest results. My Cosmos was set at 4.5V input sensitivity. I fabricated the divider (thanks!) and have a XLR y-cable, so I will be playing around with that.

so please take it serious. It is a learning process that is never ending.

OMG! This is so true. We are talking on the order of -150 dB here! That's what makes it fun. Thanks so much for the help!

 

[linear diver]

b) If you are more a practical DIYer than hig-hend-perfectionist, you can try to use neighbors USB ports placed closely on the PC's mother-board i.e. reduce the GND loop effect.

Hi Ivan! I guess I'm both a practical DIYer AND a big-headed-perfectionist

So far, I have not noticed a coorlation between ports and PC-PC performance, but I'll keep an eye on it.

What about using the Cosmos with a Mac ( an old one 2015 ... Yosemite)? I tried it and it sort of worked, but I don't trust it.

 

[pma]

PS: Pavel, tell me pls why don't believe my Cosmos ADC AUX input?
It works up to 43Vrms and utilises 2pcs in serial(!) a high quality Viking MELF 0207 resistors.

It is nice, but:

1) I need input range between 10V and 43V as well. In the PM-A4 measurement, the input range of the ADC is about 16.93V, with a divider. This covers my low power (50W/4ohm) amps.
2) If something has to be burnt it should be the external divider and not the internal SMD parts
3) I do not like the tiny input TRRS connector, as well as I do not like any 3.5mm and 2.5mm jacks.

 

[pma]

b) If you are more a practical DIYer than hig-hend-perfectionist, you can try to use neighbors USB ports placed closely on the PC's mother-board i.e. reduce the GND loop effect.

I am very practical engineer and my measuring loop to a big power amp is usually 2m + 2m = 4m. So any games with the "short tracks" are pointless. USB isolator or balanced input with only 1 ground point are the cure. The goal is not only to measure DAC-ADC loop .

P.S.: the option of balanced input without COM wire does not work well, just tested. USB-ISO remains recommended.

 

[pma]

So one more, with detailed test conditions. It is really not a 16-bit measurement, as @IVX was suspecting in another thread. SINAD (16Hz-22kHz BW) is 111.4dB, and this is very close what @amirm has measured with Topping D10s at lower Fs=44.1kHz (112.24dB), in fact it tells that the two methods and systems give almost same result. And, the THD distortion components measured with Cosmos are lower than those in Amir's review, the reasons can be only speculated. The measurement was done at 0dBFS Topping D10s output, which resulted in -3.7dBFS E1DA Cosmos input. Please take into account 250 ohm Cosmos input impedance in mono 2-channel mode at 2.7V input range. This is an engineering discipline and everything has to be correctly stated and described.

Topping D10s USB DAC and Bridge Review

This is a review and detailed measurements of the Topping D10s USB DAC and S/PDIF plus Toslink bridge. It was kindly sent to me by the company for testing. The D10s has a retail cost of US $99. Let me start by confessing that the original D10 was one of my favorites with its cute orange...

www.audiosciencereview.com

 

[AnalogSteph]

P.S.: the option of balanced input without COM wire does not work well, just tested. USB-ISO remains recommended.

Huh. How exactly did you test this? I suspect that since ADC input impedance is low in general, the same may be true for common mode, so the slightest bit of imbalance in source impedances could screw up effective CMRR.

Mind testing common-mode input impedance (I won't be surprised if it's no more than 2x differential mode) and best-case CMRR (i.e. 2 and 3 tied together)?

No problem, I use usual resistor dividers, both SE and balanced configuration, at low impedance (not to add resistor noise), like 1k / 100R and appropriate switch setting on E1DA.

I don't actually think you'd need fundamentally different attenuators for balanced and SE.

This is what I have previously suggested for a speaker measurement setup without galvanic isolation:

It can take both kinds of input. Hand-match R1 = R3, R2 = R4, and physical construction should actually be squashed vertically for minimum loop areas. When using galvanic isolation, results may be improved by an optional OUT- to SHIELD connection due to reduced common-mode signal levels.

 

[pma]

Huh. How exactly did you test this? I suspect that since ADC input impedance is low in general, the same may be true for common mode, so the slightest bit of imbalance in source impedances could screw up effective CMRR.

Mind testing common-mode input impedance (I won't be surprised if it's no more than 2x differential mode) and best-case CMRR (i.e. 2 and 3 tied together)?


I don't actually think you'd need fundamentally different attenuators for balanced and SE.

This is what I have previously suggested for a speaker measurement setup without galvanic isolation:

It can take both kinds of input. Hand-match R1 = R3, R2 = R4, and physical construction should actually be squashed vertically for minimum loop areas. When using galvanic isolation, results may be improved by an optional OUT- to SHIELD connection due to reduced common-mode signal levels.

Hi, yes, I have been using the same divider arrangement for years, so nothing new. It worked well with Focusrite 2i2 and I also designed an interface on the same principle and described here

Balanced Analog Audio Measuring Interface for the Soundcard

Balanced Analog Audio Measuring interface for the soundcard I was tired to solder divider diy patches every time I needed to measure power amplifiers with a soundcard, and also by investigating how to avoid groundloops and mains related interference spectral lines. So I decided to design and...

www.audiosciencereview.com

But - the interface degrades parameters of E1DA Cosmos considerably. I also tried the divider only (same as yours) but the result without COM (and no USB-ISO) is much worse than with USB-ISO, speaking specifically about combination with E1DA Cosmos ADC. So it is not an option, to me, in this case.

Please note I know exactly what you speak about.

 

[pma]

This is what I have previously suggested for a speaker measurement setup without galvanic isolation:

Hi, yes, I have been using the same divider arrangement for years, so nothing new. It worked well with Focusrite 2i2 and I also designed an interface on the same principle and described here
But - the interface degrades parameters of E1DA Cosmos considerably.

Here is the comparison of all 3 interconnection methods with DAC on USB1 and ADC on USB2 of the same PC, no other connection to mains PE earth. This means that USB1 and USB2 share the same PC ground.

1) Green is usual direct interconnection, no isolation. USB1==>DAC--SE link cable==>Preamp--balanced out cable with COM==>ADC==>USB2
2) Orange and red have USB1 isolated. USB1+USB-ISO==>DAC--SE link cable==>Preamp--balanced out cable with COM==>ADC==>USB2
3) Blue is direct interconnection, but ADC input only to +IN and -IN pins, shield connected with COM at only 1 end, same as in @AnalogSteph suggestion. The GND loop is thus interrupted. USB1==>DAC--SE link cable==>Preamp--balanced out cable without COM==>ADC==>USB2

One can see that the result of method (3), blue, is worse than (2) with USB isolator. Please note blue spikes below 1KHz and also some HF spikes that are not present in the orange plot (2).The method (2) gave the best results re distortion and noise.

 

[AnalogSteph]

At least (3) seems to be a whole lot better than (1) already, making the difference between a measurement that's clearly compromised and one that's actually decent and fairly usable, so there's that... but an effective ~22 dB of CMRR still is Not Great absolutely speaking.

There are only two things I can think of that could be responsible for this:
1. You've still got shield current running on the unbalanced connection between DAC output and preamp input. In this case there should be a difference to be seen if the interface is explicitly grounded, or simply used on the battery-operated laptop instead (which is another way of providing galvanic isolation, of course). Another telltale sign for this would be seeing a difference with the USB isolator on the DAC side vs. the ADC side.
2. The balanced connection may be slightly imbalanced. In this case a trimmable version of the balanced attenuator would be worth a shot (It would take 2 trimmers I think, one for either R1 or R3 and the other for either R2 or R4). Could be a bit of a pain to adjust though.

 

[pma]

I think I found why the result in the setup #3 (2 wire balanced input, 1 COM point of USB1 and USB2, no loop, no USB isolator) is inferior and worse than I have expected.

@IVX : Ivan, I have measured the CMRR of E1DA Cosmos ADC and it is not very good, like 52dB (range 2.7V, mono, 2-channel parallel). How well is the input divider impedance matched? Below please see the measurement result, blue is input level and green is output level with both +IN and -IN driven simultaneously (shorted). Generator output impedance was 20 ohm, so the possible mismatch to be related to 20 ohm.

 

[IVX]

pma, opamp's feedback resistors are .5%, input resistors are 1%&.5%. My bench Cosmos ADC also has CMRR of about -54db@1kHz for a stereo mode.
Viking offers a lead time of 3 months for .5-.25% resistors, for example, if I don't replace Viking in May for Chinese .25% 0207, I would wait for Viking up to the end of Jul..
BTW, these .25% Chinese .25% MELFs in Cosmos APU show me CMRR -120db@1k, however, I didn't test many units for that.

 

[linear diver]

This is what I have previously suggested for a speaker measurement setup without galvanic isolation:

Hi @AnalogSteph ! Do you mind if I ask what software you use to make diagrams and run simulations? I'm getting started with my new Cosmos ADA and Topping D10s and I need to update my software 'arsenal' ... just REW for now.

 

[pma]

Measurement of OPA1622 - perfect! Cleaner than D10s-E1DA loop. (OPA1622 EVM has 80dB CMR)

 

[AnalogSteph]

Measurement of OPA1622 - perfect! Cleaner than D10s-E1DA loop. (OPA1622 EVM has 80dB CMR)

The advantages of the OPA1622 EVM are twofold:
1. 0.1% tolerance resistors
2. Higher input impedance than the Cosmos ADC in all ranges from 1.7 Vrms to 10 Vrms (~4 kOhms rather than 640 ohms to 3.48 kOhms), reflecting in higher common-mode input impedance as well.

Around 50 dB of CMRR is actually a fairly typical result for this sort of input stage topology with normal quality parts. Designing for high common-mode rejection in balanced audio inputs discusses how resistor tolerances affect CMRR. The table will tell you that even 0.1% resistors are only good for a worst-case CMRR of 54 dB, so you have to hope that they're actually tracking much closer than that in each batch (the hobbyist is best advised to hand-match critical resistor pairs with a multimeter, having a bunch of cheap 1% Yageos to sift through may make more sense than just a few expensive 0.1% tolerance parts in this regard).

For very high practical CMRR, you need both well-matched resistor pairs and a high ratio of common-mode input impedance to source impedance mismatch. You could just use higher resistor values, but this obviously comes with a noise penalty. One thing you can do to avoid that is introducing unity-gain buffers, the INA1650 architecture shown in the above document is a typical example. Douglas Self has been using this topology for years to good effect. One downside is that the input cannot take any voltage outside the supply rails, though this often is a tradeoff you can live with.

The classic Bill Whitlock article, Design of High-Performance Balanced Audio Interfaces, is a must-read in this context. (The Rod Elliott article is a good introduction and jumping-off point.) It also introduces an even more advanced method of increasing common-mode input impedance through common-mode bootstrapping, commercially implemented in the THAT1200 series of receivers.
The T-type input topology (R1, R2, R5 in Fig 9) can also be used with conventional buffered inputs and enables a high common-mode input impedance while differential mode input impedance can remain modest, so an open input is not overly prone to hum pickup. (For example, 2x 4k7 + 1x 470k. If you do the math, this has more than twice the common-mode input impedance compared to using a traditional two 470k resistors.) The same trick will also minimize the effect of mismatch in input capacitors on CMRR (so you could put in e.g. some 2n2s and "defuse" them with 100p or 47p in series to COM).

About 52 dB of CMRR would have been fine had they actually been available in practice, but 1 kOhm of input impedance in a basic balanced input stage means that the affair is very sensitive to CMRR degradation due to source impedance mismatch.

Hi @AnalogSteph ! Do you mind if I ask what software you use to make diagrams and run simulations? I'm getting started with my new Cosmos ADA and Topping D10s and I need to update my software 'arsenal' ... just REW for now.

This particular diagram was, believe it or not, drawn in LTspice (line width = 2, background white, grid off). If all you've got is a hammer and such. I do not doubt for a second that you can get far more suitable software with prettier output if all you need it for is drawing schematics, like Kicad's editor or something. LTspice will do in a pinch though.

 

[IVX]

AnalogSteph, my AP SYS2522 has 2 "trim-pots"(one is actually the cap) to fine adjustment of CMRR down to -100db+, and I have to say that's doesn't really help to avoid the GND loops effect. The best way is always isolation, battery-powered laptop as an example or $40 USB HS ISO for the DAC.

 

[AnalogSteph]

@AnalogSteph, my AP SYS2522 has 2 "trim-pots"(one is actually the cap) to fine adjustment of CMRR down to -100db+, and I have to say that's doesn't really help to avoid the GND loops effect.

This would have to be looked at in more detail (what sort of setup etc.). As stated above, you may be able to trim theoretical CMRR to astronomical levels in a synthetic test (XLR 2+3 shorted, signal applied between that and 1), but that doesn't necessarily mean that common-mode input impedance is high enough to reliably get anywhere near that in real life when any amount of source impedance imbalance is present. In the AP SYS-2522, each leg is specified with 100k||185p typ, common mode should be seeing 50k||390p.

Doesn't the AP have galvanically isolated inputs and outputs though (on I²S level)? That's what I was thinking anyway (isn't that one of the key benefits of a legit dedicated audio analyzer?). It would be nigh-on impossible to take meaningful unbalanced loopback measurements otherwise. For this reason it may actually be beneficial to establish exactly one explicit ground connection in the measurement setup, so that ground is actually near earth potential rather than floating at part of the mains voltage. I would expect some mains leakage through the AP's power supply section, but no more than a few 10 VA of mains transformer worth (or even less if a shield winding is present).

 

[IVX]

AnalogSteph, no )) I believe if APx555b was isolated from its USB they would ask not $30K but $300K ))