QuantAsylum QA400 and QA401

Hmmm... this is getting beyond my pay-grade and job description.

Sounds to me that the lower the driver's output Z, the less noise?
Maybe?

The part is rated in specs at only ~110dB dynamic range, despite the 24bit aspect. So, there is something going on there, which well may be this specific problem it would seem.

The other thing that occurred to me is that the "magic" 1ufd cap is sort of like a single pole compensation. Perhaps a split pole or two pole affair might provide a degree of more betterer results?

Sure would like to kill those spurs, cap'n!
 
Bear what it come down to is there are several sources of noise that can be causing the spurs. Digital ingress is one. Loading from the ADC conversion is another which Demian described earlier. External noise entering with the signal is yet another and then there is simply over driving the ADC. That over drive may below the maximum Vpp indicated in the data sheet. What's on the data sheet is nominal.
 
Member
Joined 2004
Paid Member
1) 24 bit conversion at audio frequencies is still beyond or technology. Maybe possible at 500 Hz. This SOTA device will do it at 50/second: http://cp.literature.agilent.com/litweb/pdf/5965-4971E.pdf LTC claims 8000/sec but I'm not sure that's 24 bits at 8000/second. Definitely not 44.1 KHz sample rates. Most are at 15/second.
2) 32 bit conversion of anything is pure fantasy.
3) If you do a high resolution FFT you can always pull spurs out of the noise. The plots I posted earlier show the spurs crawling out of the noise.
4) If the spurs really bother you an easy fix is to reduce the resolution so the noise comes up to hide the spurs. . .

The trick in the AK5394A datasheet offsets the zero to move away from some of the processing issues. It could make sense to add a milliVolt of offset to try to reduce the Zero level junk. Pick the right opamps and it comes for free.
 
Disabled Account
Joined 2012
The trick in the AK5394A datasheet offsets the zero to move away from some of the processing issues. It could make sense to add a milliVolt of offset to try to reduce the Zero level junk. Pick the right opamps and it comes for free.

Suggest a mod/schematic to the QA400 and we will try it. Sounds like a very good solution - if it works well.

Thx-RNMarsh
 

mkc

Member
Joined 2002
Paid Member
Hi Demian

Set the system to 48 KHz, take a measurement. Shut it down and restart the app. See if it changes at all. Do the same with 192.

Here is my plot. From left.

1. Start application and set settings to 48KHz.
2. Close application and re-plug the QA400 HW. Start application and make Measurement at 48 KHz
3. App still running. Change 48KHz to 192 KHz. Make Measurement.
4. Close application and re-plug the QA400 HW. Start application and make Measurement at 192Khz

3 and 4 differ a lot even the settings didn't change between these 2 Measurements.

I agree. There must be a SW bug at play here.

Mogens
 

Attachments

  • QA400 10102013_1_48KHz_after_boot.png
    QA400 10102013_1_48KHz_after_boot.png
    29.8 KB · Views: 370
  • QA400 10102013_1_48KHz_restart_app.png
    QA400 10102013_1_48KHz_restart_app.png
    29.7 KB · Views: 370
  • QA400 10102013_1_192KHz_after_48KHz.png
    QA400 10102013_1_192KHz_after_48KHz.png
    27.8 KB · Views: 371
  • QA400 10102013_1_192KHz_restart_app.png
    QA400 10102013_1_192KHz_restart_app.png
    28.7 KB · Views: 363

mkc

Member
Joined 2002
Paid Member
Here is the same exercise with Pre-release 0.61. From left.

1. Start application and set settings to 48KHz.
2. Close application and re-plug the QA400 HW. Start application and make Measurement at 48 KHz
3. App still running. Change 48KHz to 192 KHz. Make Measurement.
4. Close application and re-plug the QA400 HW. Start application and make Measurement at 192Khz

At least they are consistent. But spectrum not as "clean" as some of my others.

Mogens
 

Attachments

  • QA400 10102013_rev061_192KHz_restart_app.png
    QA400 10102013_rev061_192KHz_restart_app.png
    28.6 KB · Views: 111
  • QA400 10102013_rev061_192KHz_after_48KHz.png
    QA400 10102013_rev061_192KHz_after_48KHz.png
    28.7 KB · Views: 110
  • QA400 10102013_re061_48KHz_after_boot.png
    QA400 10102013_re061_48KHz_after_boot.png
    29.7 KB · Views: 105
  • QA400 10102013_rev061_48KHz_after_192KHz.png
    QA400 10102013_rev061_48KHz_after_192KHz.png
    29.8 KB · Views: 364
Ok, ok, so there may be more than one cause/source of the spurs. Some may be inherent to the internal workings of the chip.

So, the next step, if there is to be a next step, is to try a few things and see if some or any or all of the spurs can be effected for the better by external means.

Seems like the first one, and easiest is to drop that input impedance down by 10x or 100x and see what that does. IF that is sufficient, then making up an ultra low distortion buffer to drive the input (at least on paper) ought to be amongst the easiest things we can do.

The VCOM thing remains unclear to me... is it not providing the offset that we are talking about?? Being buffered by an opamp? No? If not, then what is the purpose of the opamp to the 100Kohm "input" resistors?

signed,

remaining confused
 
Ok, ok, so there may be more than one cause/source of the spurs. Some may be inherent to the internal workings of the chip.

So, the next step, if there is to be a next step, is to try a few things and see if some or any or all of the spurs can be effected for the better by external means.

Seems like the first one, and easiest is to drop that input impedance down by 10x or 100x and see what that does. IF that is sufficient, then making up an ultra low distortion buffer to drive the input (at least on paper) ought to be amongst the easiest things we can do.

The VCOM thing remains unclear to me... is it not providing the offset that we are talking about?? Being buffered by an opamp? No? If not, then what is the purpose of the opamp to the 100Kohm "input" resistors?

signed,

remaining confused

The VCOM provides a common bias voltage for the ADC. It shifts the input AC from ground center to 1/2 the the Vpp input swing. This is necessary because the ADC is power from a single supply. Vcom is also the ADC's Vref. I can only guess at what the inner circuit looks like. On other industrial ADC this ref voltage is coupled by a resistor so RC filtering can be arranged externally by adding an external cap. With the Crystal this cap effects the noise shaping including some of the spur.

What's in the AKM data sheet adds a differential spread to the bias and moves the input bias slightly away from common. This also reduces spurs at the cost of dynamic range.

The op amp is a buffer to isolate the Vcom from stray reactance which would effect the noise shaping.
 
Member
Joined 2004
Paid Member
MKC-
I would be suspicious of the clean plots. They are too good to be real. The dirty ones are probably more real. With windowed measurements there are other potential artifacts that can surface from timing errors etc. Do you set similar results with more averages?

Have you tried the melt noise option? It will show synchronous stuff that won't go away.
 

mkc

Member
Joined 2002
Paid Member
Hi Demian,

MKC-
I would be suspicious of the clean plots. They are too good to be real. The dirty ones are probably more real. With windowed measurements there are other potential artifacts that can surface from timing errors etc. Do you set similar results with more averages?

Have you tried the melt noise option? It will show synchronous stuff that won't go away.

I have not tried the melt noise option. I agree that it is suspicious. In any case, I don't think this is a HW error in my device, so I think there must be a SW bug here. But what is the truth, I can't say. I think the Quantasylum boys must look at this.

Mogens
 
The VCOM provides a common bias voltage for the ADC. It shifts the input AC from ground center to 1/2 the the Vpp input swing. This is necessary because the ADC is power from a single supply. Vcom is also the ADC's Vref. I can only guess at what the inner circuit looks like. On other industrial ADC this ref voltage is coupled by a resistor so RC filtering can be arranged externally by adding an external cap. With the Crystal this cap effects the noise shaping including some of the spur.

What's in the AKM data sheet adds a differential spread to the bias and moves the input bias slightly away from common. This also reduces spurs at the cost of dynamic range.

The op amp is a buffer to isolate the Vcom from stray reactance which would effect the noise shaping.

Ok, so we want to sit the input at 1/2 Vpp. Vcom puts it dead bang on. Ok.

Now, we can filter with something *other than* a cap.
Like an inductor.

Anyhow, I think that it would be a darn good idea to look at the waveform is that is sitting on these pins...

The aim would be to try to determine what is effecting what.
For example *if* the injection of Vcom is actually re-injecting the spurs into the input, that leads one to think of ways of correcting that, including flipping the phase for example.

But if the spurs are entirely an internal generation, then I guess one has to live with the limitation and go for a decent notch filter when one needs to see a bit "deeper"...

Btw, where is the "noise melt" option sitting on 1.06?
 
Ok, so we want to sit the input at 1/2 Vpp. Vcom puts it dead bang on. Ok.

Now, we can filter with something *other than* a cap.
Like an inductor.

Anyhow, I think that it would be a darn good idea to look at the waveform is that is sitting on these pins...

The aim would be to try to determine what is effecting what.
For example *if* the injection of Vcom is actually re-injecting the spurs into the input, that leads one to think of ways of correcting that, including flipping the phase for example.

But if the spurs are entirely an internal generation, then I guess one has to live with the limitation and go for a decent notch filter when one needs to see a bit "deeper"...

Btw, where is the "noise melt" option sitting on 1.06?

Phase cancellation is not so easy to do. You really have to be zero beat with that for it to be effective. It's like tuning a notch filter.
 
Bear the ADC loads the input during a conversion which in turn loads the driving circuit. This can cause noise during conversion which is mixed with the signal and can generate spurs. If Crystal is using switched capacitor technology at the input then this very likely a cause.
I found in the past - deep, dark past - that including something like an LT-1010 within the opamp loop reduces problems where the load impedance suddenly changes, like when charging a cap. It's really single ended Class A, with adjustable standing current, so there are no cross-over challenges.

Perhaps using something like that for the input stages as well as maybe for the VCom driver might help.

Of course, this makes it even harder to run directly from USB power. Maybe the compromise could be made into a feature...
 
Member
Joined 2004
Paid Member
Here is the latest version of the schematic. I simplified it since the gains will be fixed at 6 dB in and out, which translates into around 3V, enough to drive most consumer power amps to full output. it will use 1/4" TRS balanced connectors in and out. Also it will use a simple charge pump and share the ground with the QA400 since the new version will fit inside the QA400 box.

The latest version of the layout should be done in a week or so.
 

Attachments

  • QA400 interface gen 2 -2.zip
    26.1 KB · Views: 164
Tried to run the output of the QA osc into the VP7725 analyzer section.

NG.

Did someone say there was a way to make the QA osc run continuously??
If so, that's what is needed.

On a related topic, the notch.

As long as the fundamental is removed by a sufficient number of dB, are there any concerns about *also* removing the spectrum *below* the fundamental as well as the fundamental?

Also WRT the notch in general, what about induced ripple in the passband (outside of the stopband, of course)?

If we can live with say <1dB ripple in the passband I can think of a good fast way to accomplish the HP function with very high slope. You can probably do as well.

_-_-bear