Here's an experiment. Take some copper foil tape and stick a slice over the trace on the QA board leading into the driver op amp. Drive the foil with a low Z copy of the input signal and see what happens to the distortion. Then try a low Z copy of the output signal from the driver op amp. 'Driven shield.' This will wipe out any stray capacitance.
This is a snip of a string of emails from QA's Matt Taylor during the beta testing of the QA400. The pics are out of order from the dialog but you can piece it together.
On 1/6/2013 10:30 PM, Matt Taylor wrote:
Hi David,
The CS4271 and CS4272 are two variants of the same part and in fact might be the exact same die. One is single ended, the other differential. We’ll do a fully differential product at some point, and we wanted to keep just one part in inventory.
Below you can see the 4271 and 4272 suggested input. Note the 4271 just ties the negative inputs (what they call VQ2 and VQ3) to VCOM. And then the 4271 has precisely 6 dB less DR. The noise floor is the same, but the total amplitude is half because it’s single ended. So the numbers make sense.
Why I think this might be inherent in the CS ADC is because it seems odd that only sub 100 Hz noise would be present on the BNC ground and not present in the Vcom. How could a layout or routing be so selective over just an octave? It make no sense to me. If the ground was sketchy, then the impact would run well beyond 100 Hz. Now, it’s fully possible that we could get rid of this via a dramatic placement shakeup, and we’ll explore that for a future product.
From your other mail, yes, agree on negative supply. But we really wanted to hit that magic $200 point and that required some compromises. Another product in the future will pull out all the stops…relay attenuators, fully differential inputs, full isolation…
More after the pictures
Also agree that EMU doesn’t have this issue to the same extent. But they still have a rising noise floor. They use an AK5385. Below is EMU, cal’d to -15 dB. Red is with nothing plugged in and the blue is with a 68 ohm short.
And here’s my stock QA400 with both 68 ohm load and open circuit.
Also, take a look at the EMU in loopback at -15 dBV from its rear output to its rear input. And compare that to the QA400. Not even close J
On 1/6/2013 10:30 PM, Matt Taylor wrote:
Hi David,
The CS4271 and CS4272 are two variants of the same part and in fact might be the exact same die. One is single ended, the other differential. We’ll do a fully differential product at some point, and we wanted to keep just one part in inventory.
Below you can see the 4271 and 4272 suggested input. Note the 4271 just ties the negative inputs (what they call VQ2 and VQ3) to VCOM. And then the 4271 has precisely 6 dB less DR. The noise floor is the same, but the total amplitude is half because it’s single ended. So the numbers make sense.
Why I think this might be inherent in the CS ADC is because it seems odd that only sub 100 Hz noise would be present on the BNC ground and not present in the Vcom. How could a layout or routing be so selective over just an octave? It make no sense to me. If the ground was sketchy, then the impact would run well beyond 100 Hz. Now, it’s fully possible that we could get rid of this via a dramatic placement shakeup, and we’ll explore that for a future product.
From your other mail, yes, agree on negative supply. But we really wanted to hit that magic $200 point and that required some compromises. Another product in the future will pull out all the stops…relay attenuators, fully differential inputs, full isolation…
More after the pictures
Also agree that EMU doesn’t have this issue to the same extent. But they still have a rising noise floor. They use an AK5385. Below is EMU, cal’d to -15 dB. Red is with nothing plugged in and the blue is with a 68 ohm short.
And here’s my stock QA400 with both 68 ohm load and open circuit.
Also, take a look at the EMU in loopback at -15 dBV from its rear output to its rear input. And compare that to the QA400. Not even close J
Attachments
So, a low distortion buffer to do this job?
Don't want to induce any nasties, I would guess.
The real issue is the semiconductor junctions are nonlinear caps. Tricks like Cascoding reduce this effect. Unfortunately its hard to get a large common mode input range with a cascoded input so I don't think you will find it in may opamps if any. I posted a circuit that could reduce the effect here: http://www.diyaudio.com/forums/equi...dc-design-ak5394a-other-adcs.html#post3708537 however its a lot of parts to dot he task. The capacitance of the traces on the PCB are a small part of the issue.
So, at low levels the capacitance of the input device starts to look big?
But isn't the variable capacitance very small and therefore more or less "out of band"??
Or does the high input Z make it not so out of band?
I am at a disadvantage here, not being up to speed on this. Sorry.
But isn't the variable capacitance very small and therefore more or less "out of band"??
Or does the high input Z make it not so out of band?
I am at a disadvantage here, not being up to speed on this. Sorry.
Last edited:
Here is the issue. All semi junctions are voltage variable capacitors. If your load changes capacitance say 1000 parts per million between zero volts and 1 volt and your source impedance is .001X the impedance of the capacitance the modulation will change the signal level during the cycle to show as harmonic distortion at -120 dB (1 PPM). That is the level we are discussing here. Change the source Z to something low and the cap change becomes small enough to not show. However if the opamp is tied to a 10K pot the lowest source Z would be 2.5K unless you are at one end of the pot. In this case the modulation of the capacitance could be quite significant.
Then theoretically if there were a second derivative capacitance changing along with junction capacitance the load would be balanced.
Another technique could be to add a capacitance parallel to the junction capacitance and then cancel it with a derivative signal.
Kind of like what Scott did with the 797.
Another technique could be to add a capacitance parallel to the junction capacitance and then cancel it with a derivative signal.
Kind of like what Scott did with the 797.
Last edited:
Lowest Zo is best -
In most, if not all, cases the output Zo from a source is set to a "standard" value with a series R.... 50 ohms or 75 or 600 ohms, for example. For lowest distortion, those "standard" values need to be forgotten about and reduced to milli-Ohm Zo of just the opamp or output stage. Often just a jumper across the standardizing Ro resistor can help.
It has other benefits as well..... reduction of emi/rfi pickup, for one. However, the OPS needs to remain stable under various test loads and current limited for self protection while keeping the lowest Zo.
This is what I also do within audio components.. reduce the Zo of each connected piece of audio equipment on the recording and the playback side of the process. (then check for stability etc). It reduces many of the interfacing issues that occure with a high(er) Zo. BUt it isn't a substitute for a design that is stable etc and with low Zo from the R&D stage.... from the design beginning.
Thx- RNMarsh
In most, if not all, cases the output Zo from a source is set to a "standard" value with a series R.... 50 ohms or 75 or 600 ohms, for example. For lowest distortion, those "standard" values need to be forgotten about and reduced to milli-Ohm Zo of just the opamp or output stage. Often just a jumper across the standardizing Ro resistor can help.
It has other benefits as well..... reduction of emi/rfi pickup, for one. However, the OPS needs to remain stable under various test loads and current limited for self protection while keeping the lowest Zo.
This is what I also do within audio components.. reduce the Zo of each connected piece of audio equipment on the recording and the playback side of the process. (then check for stability etc). It reduces many of the interfacing issues that occure with a high(er) Zo. BUt it isn't a substitute for a design that is stable etc and with low Zo from the R&D stage.... from the design beginning.
Thx- RNMarsh
Last edited:
The distortion added goes up further with increased source impedance.... At source generator Zo = 3.3K the thd+n increased about .0003% But the real trouble for the Qa400 was still the many added spurious tones and harmonic products. Especially 2H which was all but gone from the source but the QA400 indicates 2H of significant level as well.
While doing this --> I learned the input Z of the QA400 at 1KHz is only 8K Ohm (Spec is 10K). This is too low for many equipment measurements. I'd like to see ac input Z of 100K.
Thx-RNMarsh
While doing this --> I learned the input Z of the QA400 at 1KHz is only 8K Ohm (Spec is 10K). This is too low for many equipment measurements. I'd like to see ac input Z of 100K.
Thx-RNMarsh
Last edited:
They must have a shunt resistance in there. The op amp operates in non inverting mode and the VCOM resistor is 100k. What's changing the Z.
It's been a long time since I've had the cover off.
Tomorrow I'll open it up and rip some components out of input.
Last edited: