Echo Mona Audio Interface RMAA Result

[chenxuwen]

Dear All,

May I share the RMAA result of my Echo Mona? I prefer this one to my ASUS Xonar D2 which has 115db dynamic range. (Will post its RMAA result if interest is showed, if it is OK and if I am posting at the right place.)

It is still working nicely today, but I cannot understand why its THD+N is not good enough.
I plan to mod the AK4393 to AK4396. If you could provide some inputs on these two things, that would be great!

The test was done using XLR cable at 24Bit 96Khz.

Frequence Response:

Noise Level:

Dynamic Range Test:

THD+N:

IMD:

IMD Swept:

Crosstalk:

The Summary:

 

[AnalogSteph]

It is still working nicely today, but I cannot understand why its THD+N is not good enough.

At up to +22 dBu, the output can deliver substantial analog level. Circuitry might be single-ended and not using the greatest opamps by modern standards. Also, the rise in noise floor indicates independent PLLs for ADC and DAC clock, or at least somewhat more phase noise than what you'd ideally want. Heaven knows whether that's fixable but looking into chip datasheets certainly can't hurt. In any case you're looking at some reverse-engineering there, basically the output and input stages.

While you're in there, check XLR pin 1 (TRS ground) routing for AES48 compliance.

Oh, wait, were you using an XLR --> XLR cable? That would be using the mic preamp, which by the looks of it will take 20 dB less, or +2 dBu max. (Distortion ought to be limited by the input side then.) For line-level loopback you need XLR(f) --> TRS, for the TRS portion of the combo jacks.

In order to make sure what is limiting you, you really want an external attenuator / amplifier in the loop that qualifies as sufficiently transparent. You could use one of the fancy balanced headphone amps that are so common these days, like maybe a Sabaj A20h with an output adapter cable (e.g. 4.4 mm Pentaconn to 2x XLR).

I plan to mod the AK4393 to AK4396.

Not worth the hassle I think. The converters aren't really your problem. (If you absolutely insist, the AK4395 has a better digital filter, so that's the one I would be going for. Make sure DVDD is within spec though. Only the AK4393 can also operate at 3.3 V. The Motorola 56301 DSP they've got in there is a 3.3 V job for sure.)

Here's another AK4393-AK5393 combo for comparison:

Behringer MODs DCX Upgrade Review

This is a review and detailed measurements of an upgrade for the venerable Behringer Ultra-Drive Pro DCX2496 by Behringer MODs. An upgraded unit was sent to me for testing. The upgrade itself costs £138 in UK and £115 outside UK. The board fully upgrades the existing DAC board: I like the...

www.audiosciencereview.com

 

[Sir Sanders Zingmore]

THD+N is not good enough

-91.8dB is almost certainly inaudible (and therefore “good enough”). It may not be state of the art, but does it really matter?

 

[sam_adams]

It is still working nicely today, but I cannot understand why its THD+N is not good enough.

ENOB=15. Just shy of CD quality.

 

[wwenze]

Spec:

D‑A converters: AKM AK4393 24‑bit multi‑bit delta‑sigma 128x oversampling.
Input dynamic range: 115dB A‑weighted.
Output dynamic range: 116dB A‑weighted.
Frequency response: 10Hz to 22kHz (±0.25dB).
THD + Noise: <0.002% (20Hz to 22kHz).

Sure seems good enough to me and certainly close enough to measurements.

 

[chenxuwen]

Thanks for your input. I now see the problems now.
Putting together the input of @wwenze and @AnalogSteph, the THD+N is limited by the following factors:
1. The input INA134 at the mic stage. Its sensitivity might be the biggest factor as I need compressed air cleaner on the input ports to reach this RMAA result.
2. The output balance driver DRV134 which makes the singled ended output balanced.
3. There is indeed a PLL inside the TCXO I modded. (Actually, the output OP AMPS have been modded to LME49722 while the ADC input OP AMP after INA134 are OPA1642)

As I mainly uses the unbalanced outputs to poweramp and headphone output, so I guess I should let this THD+N problem go then.

The AK4393 is supplied by 3.3V digital now, so for a drop in replacement, I guess it would have to be AK4396 which I already got 5 pcs. (I actually have 3 pcs of AK4395 then realized the problem.) There are also much less fake/remarked AK4396.

Could you give me further advice for a low phase noise clock with size smaller or equal to DIP8? (I have sockets for the clocks so changing them is easy, and I would like see the difference.) But please, no Crystek or things more expensive than that, I cannot afford that.

I have to work with what I have 1st. (But that Focusrite Scarlett 4th Gen's spec does look sexy, if its measurement can be at that level. At the same time, my Titanium HD and Xonar D2 measure better, I still prefer this Echo Mona...)

 

[AnalogSteph]

Could you give me further advice for a low phase noise clock with size smaller or equal to DIP8?

What kind of setup is the unit running stock? Is there a crystal inside the external unit or is everything derived from the oscillator we see on the PCI card? The crux of the matter is that (low-frequency) phase noise common to both ADC and DAC would cancel out to a large degree in loopback measurements, so there must be sort of a jitter problem getting the clocks from point A to point B.

From my experience with CMOS VCOs in a past life, a quiet power rail will go a long way. Same goes for inverter-based XOs (HC04), although those should never have concerning levels of phase noise for audio applications anyway. Basically, everything that is banging between power and ground all the time tends to have poor PSRR. It used to be a major problem for Class D amplifier as well until designers found ways to introduce feedback.

 

[chenxuwen]

What kind of setup is the unit running stock? Is there a crystal inside the external unit or is everything derived from the oscillator we see on the PCI card? The crux of the matter is that (low-frequency) phase noise common to both ADC and DAC would cancel out to a large degree in loopback measurements, so there must be sort of a jitter problem getting the clocks from point A to point B.

From my experience with CMOS VCOs in a past life, a quiet power rail will go a long way. Same goes for inverter-based XOs (HC04), although those should never have concerning levels of phase noise for audio applications anyway. Basically, everything that is banging between power and ground all the time tends to have poor PSRR. It used to be a major problem for Class D amplifier as well until designers found ways to introduce feedback.

Thanks for helping me.

The stock and now have two dip 8 size vco (tcxo now), one 11.2896 and one 24.576 inside for the clock. The powersupply is from one switch powersupply, lc filter, then lm2940+2990 (analog and digital seperate), then another 3.3v regulator. All capacitors have been recapped.

 

[AnalogSteph]

The stock and now have two dip 8 size vco (tcxo now), one 11.2896 and one 24.576 inside for the clock.

???

I think you don't actually mean VCOs but rather integrated crystal oscillators (these 4-legged can jobs), right? Looks like a typical arrangement with 11.2896 for 44.1 kHz relatives and 24.576 for 48 kHz ones. There may not be any PLL action going on at all here.

A TCXO is rather gilding the lily for an audio application, nobody really cares about + or -100 ppm of fixed offset (pitch inaccuracy) or the few ppm of thermal drift that a regular XO will exhibit. (A quartz watch that's 1 second per day fast is only 11.6 ppm off.) Some DACs in the 800+ USD class have them but it's really more of a gimmick. It's a different story when you are trying to build a communications receiver that's supposed to remain rock-steady at tens of MHz or even higher (<1 ppm of drift). If you want <20 Hz of drift for SSB on the 2 m ham band, that's <0.12 ppm!

For all I care, this area could well be left stock.

then another 3.3v regulator

And the oscillators run on that? What kind of regulator is it? Are both running all the time?

You wouldn't happen to have some photos somewhere, so that we have a better idea of what's going on in there...?

The other very critical supply that can cause dynamic range degradation would be ADC / DAC Vref. Where is that connected? It generally needs to be very clean and low impedance, up to the point where the best designs are using filtered low-noise voltage references buffered by premium low-noise opamps to generate it, or you might find a whole array of capacitors to keep impedance low over a large frequency range, far more elaborate than what the AK4393 or AK5393 datasheets suggest. An LM2940 is moderately low noise at best and would require extra cleanup. I hope there's some extra passive filtering for Vref, which would make it easier to split it if need be.

(Actually, the output OP AMPS have been modded to LME49722 while the ADC input OP AMP after INA134 are OPA1642)

What were they originally? What's the circuitry (schematic) looking like? Subpar distortion performance can be a sign of high-frequency oscillation and may go unnoticed with an oscilloscope.

So far all the parts listed should be capable of better than -100 dB distortion performance. Not sure where all that 2nd harmonic is coming from. Possibly either the microphone preamp circuitry (what is it? some INA217 job?) or the ADC driver, or the output stage.

 

[chenxuwen]

???

I think you don't actually mean VCOs but rather integrated crystal oscillators (these 4-legged can jobs), right? Looks like a typical arrangement with 11.2896 for 44.1 kHz relatives and 24.576 for 48 kHz ones. There may not be any PLL action going on at all here.

A TCXO is rather gilding the lily for an audio application, nobody really cares about + or -100 ppm of fixed offset (pitch inaccuracy) or the few ppm of thermal drift that a regular XO will exhibit. (A quartz watch that's 1 second per day fast is only 11.6 ppm off.) Some DACs in the 800+ USD class have them but it's really more of a gimmick. It's a different story when you are trying to build a communications receiver that's supposed to remain rock-steady at tens of MHz or even higher (<1 ppm of drift). If you want <20 Hz of drift for SSB on the 2 m ham band, that's <0.12 ppm!

For all I care, this area could well be left stock.

And the oscillators run on that? What kind of regulator is it? Are both running all the time?

You wouldn't happen to have some photos somewhere, so that we have a better idea of what's going on in there...?

The other very critical supply that can cause dynamic range degradation would be ADC / DAC Vref. Where is that connected? It generally needs to be very clean and low impedance, up to the point where the best designs are using filtered low-noise voltage references buffered by premium low-noise opamps to generate it, or you might find a whole array of capacitors to keep impedance low over a large frequency range, far more elaborate than what the AK4393 or AK5393 datasheets suggest. An LM2940 is moderately low noise at best and would require extra cleanup. I hope there's some extra passive filtering for Vref, which would make it easier to split it if need be.

What were they originally? What's the circuitry (schematic) looking like? Subpar distortion performance can be a sign of high-frequency oscillation and may go unnoticed with an oscilloscope.

So far all the parts listed should be capable of better than -100 dB distortion performance. Not sure where all that 2nd harmonic is coming from. Possibly either the microphone preamp circuitry (what is it? some INA217 job?) or the ADC driver, or the output stage.

Again, thanks for helping me.

Here is a photo of the board of the external box. After the LM2940-5V, there are reg1117-3.3 for the 3.3V supplies. The lm2940s are currently replaced by LT1963S.

The input circuit is not a INA217/THAT1510, it is 4 INA134s on the right side of the photo under the black heatsink. The two big heatsinks covers the 2 AK5393s.

The AK4393s on the left do not have any large caps nearby, two 22uFs are there for them.

 

[chenxuwen]

Dear All,

I have done the AK4393->AK4396 Mod. I measured the result with a 2i2 gen 4 and realized many things.

1. The ADCs in my Echo Mona are acually 2 pieces of AK5383 instead of AK5393, so its self mesurement showing 110db dynamic range is at its limit. Moreover, I have found the vref capacitor for the ADCs is a single 0.1uF cermaic, I believe this is where the S/(N+D) of the AK5383 got limited. I am going to replace the AK5383 with AK5393 and those 4 0.1uF ceramic by 220uF ceramic (Due to the technology these 20 years, there are 6.3V 220uF ceramic SMD capacitor in 1206 package now.)

2. AK4396 is indeed quieter than the original AK4393 and can be measured. People and AKM were not lying or saying marketing hype. You will see in the RMAA coming up.

3. The 2i2 gen 4 is indeed very potent!!

Here are the RMAA results:

The following noise level and the dynamic range should be limited by 2i2's input which is spec-ed at 115dB.

Notice the peak level, dB FS. The best I have measured with AK4393 was at around -75, see the original RMAA measurement, the current MONA self loop measurement also show this change. So in this sense, AK4396 is indeed quieter than AK4393.

THD+N is now at -95dB and sort of solved my concern at the begining. The 2 devices do not have clock connections, so THD+N and IMD should be tested higher than the sum of the two.

Compared to the Mona self measurement, maybe the stereo crosstalk of the input of 2i2 is weaker than the Mona. Hence the stereo cross talk and the IMD swept tone have totally different signature from the original one.

 

[AnalogSteph]

The following noise level and the dynamic range should be limited by 2i2's input which is spec-ed at 115dB.

This can be circumvented by (digitally) reducing the output level, turning up input gain as needed, and some manual calculations. You have some super low-noise inputs at your disposal after all (down to about -127 dBu(A) for the microphone part). That should be enough for output noise to be vastly dominating input noise, which is what you want.

Should you not have a digital volume control, you can always save the test waveforms in WAV format and use an external audio editor to modify them or an external audio player. (This is also how I have tricked RMAA into accepting recordings below -10 dBFS which are staunchly refused otherwise.)

Notice the peak level, dB FS. The best I have measured with AK4393 was at around -75, see the original RMAA measurement, the current MONA self loop measurement also show this change. So in this sense, AK4396 is indeed quieter than AK4393.

This would seem to be correlated with lower 1/f noise - note how much its onset has shifted to lower frequencies. Makes sense for a process improvement.

It's obviously not terribly relevant in the audible sense, but by professional standards is still nice to see.

Compared to the Mona self measurement, maybe the stereo crosstalk of the input of 2i2 is weaker than the Mona. Hence the stereo cross talk and the IMD swept tone have totally different signature from the original one.

It's the relatively basic THAT 626x series input preamp in the 2i2 that's limiting the swept IMD and distortion performance. It's much the same as what the MOTU M2 uses. The 4i4 model would have come with an extra line input that circumvents this problem.

 

[chenxuwen]

This can be circumvented by (digitally) reducing the output level, turning up input gain as needed, and some manual calculations. You have some super low-noise inputs at your disposal after all (down to about -127 dBu(A) for the microphone part). That should be enough for output noise to be vastly dominating input noise, which is what you want.

Should you not have a digital volume control, you can always save the test waveforms in WAV format and use an external audio editor to modify them or an external audio player. (This is also how I have tricked RMAA into accepting recordings below -10 dBFS which are staunchly refused otherwise.)

This would seem to be correlated with lower 1/f noise - note how much its onset has shifted to lower frequencies. Makes sense for a process improvement.

It's obviously not terribly relevant in the audible sense, but by professional standards is still nice to see.

It's the relatively basic THAT 626x series input preamp in the 2i2 that's limiting the swept IMD and distortion performance. It's much the same as what the MOTU M2 uses. The 4i4 model would have come with an extra line input that circumvents this problem.

I just got confused by

E-MU 0204 in 2023, RMAA measurements, and some modifications

Well, I want to tell you a story about E-MU 0204 and how I tamed this wild beast from early 2000s in 2023. It all started back in 2018, I was building my amplifier on TPA3255EVM and I really needed a device that I could use for measuring as well as for normal daily tasks 24/7, and for connecting...

www.audiosciencereview.com

in which the noise level test is obviously worse than mine. But it has a better Peak Level, dB FS.

Could you explain that? Is RMAA's algorithm wrong?

By the way, how do you think of my AK5383 -> AK5393 with 0.1uF vref caps changed to 100uF ceramic project? I will also change all the input potential meters.

 

[AnalogSteph]

I just got confused by

E-MU 0204 in 2023, RMAA measurements, and some modifications

Well, I want to tell you a story about E-MU 0204 and how I tamed this wild beast from early 2000s in 2023. It all started back in 2018, I was building my amplifier on TPA3255EVM and I really needed a device that I could use for measuring as well as for normal daily tasks 24/7, and for connecting...

www.audiosciencereview.com

in which the noise level test is obviously worse than mine. But it has a better Peak Level, dB FS.

Could you explain that? Is RMAA's algorithm wrong?

0204 has an AK5385 ADC, i.e. a newer model. I wouldn't be surprised if even a substantially newer-production AK5383 were to perform better than yours, assuming they weren't all made from the same run of silicon (as the improvement in 1/f noise would be a matter of process).

As an aside, the 1/f may also be added by preceding opamps, although things should be well-controlled in the typical bipolar types like 4580s or 5532s that I'd expect to find here.

By the way, how do you think of my AK5383 -> AK5393 with 0.1uF vref caps changed to 100uF ceramic project?

Be careful. Keeping impedance low over a large bandwidth is no trivial matter. You almost certainly need multiple capacitors in parallel if you want to do it right, while taking care that the parasitic Ls and Cs don't interact to form parasitic resonances causing impedance peaks. I would try 100-220n + 2.2-4.7 µF ceramic (or tant, >10 V) + a conventional 22-100 µF 10-16 V (normal/lowish ESR, radial through-hole) electrolytic tacked on in parallel. You may be able to place a 100n directly at / on the ADC pins if small enough, then you've got a space for the medium-sized ceramic, and the big guy is least critical in terms of placement.

BTW, I would imagine that a 100µ 6.3V in 1206 would be using some pretty crummy dielectric with a major voltage coefficient. Do by all means check the datasheet, you might find that it's down to a mere 20 µF at 5 V.

Note that there can be some differences in optimum drive circuitry between ADCs (e.g. CS5361 and CS5381, where distortion may be higher without changes), but since AK5383 and AK5393 were specifically designed to be very close you may have better luck there.

 

[chenxuwen]

0204 has an AK5385 ADC, i.e. a newer model. I wouldn't be surprised if even a substantially newer-production AK5383 were to perform better than yours, assuming they weren't all made from the same run of silicon (as the improvement in 1/f noise would be a matter of process).

As an aside, the 1/f may also be added by preceding opamps, although things should be well-controlled in the typical bipolar types like 4580s or 5532s that I'd expect to find here.

Be careful. Keeping impedance low over a large bandwidth is no trivial matter. You almost certainly need multiple capacitors in parallel if you want to do it right, while taking care that the parasitic Ls and Cs don't interact to form parasitic resonances causing impedance peaks. I would try 100-220n + 2.2-4.7 µF ceramic (or tant, >10 V) + a conventional 22-100 µF 10-16 V (normal/lowish ESR, radial through-hole) electrolytic tacked on in parallel. You may be able to place a 100n directly at / on the ADC pins if small enough, then you've got a space for the medium-sized ceramic, and the big guy is least critical in terms of placement.

BTW, I would imagine that a 100µ 6.3V in 1206 would be using some pretty crummy dielectric with a major voltage coefficient. Do by all means check the datasheet, you might find that it's down to a mere 20 µF at 5 V.

Note that there can be some differences in optimum drive circuitry between ADCs (e.g. CS5361 and CS5381, where distortion may be higher without changes), but since AK5383 and AK5393 were specifically designed to be very close you may have better luck there.

Thanks. Thanks for letting me know about the Ceramic capacitor voltage coefficient, I did not know it is that severe.

I am gonna use 25V 100uF 1206 with X5R. Would that be a better option then? Beside, it is only 0.1uF in there now, even it goes down to 20uF at 5V, it still more than the 10uF required in spec.

By the way, at this place, I am inclined to have one single capacitor to avoid hanging capacitors as the factory used only 1 0.1uF ceramic. If there was a conventional electrolytic there, things would be much easier.

My two options are now: 1206 25V 100uF X5R ceramic or 1206 10V 100uF Tan (much more expensive and polarized hence need extreme careful), which one do you think would be better?