Nested feedback/ composite amplifier opamp requirements(unexplained results)

[JohnYang1997]

Hi everyone.
Recently, I have been doing a composite amplifier with tpa6120a2. It's a great chip and i have had pretty good result so far.
However these is something I can't explain so far.

I have had a chance to measure the distortion of the original jds el amp and did modification with opa2211/opa1612 and had higher distortion with low impedance load at high frequency 5-10khz.
As we all know opa2211 has by far the best spec (not sure about ad797 but still).

Here is the problem. I used opa1612 to control the tpa6120 and had higher distortion at high frequency as well. When recently opa2156 came out, I tried them in the same circuit, the thd is impressive at 0.00009% at 6500hz. That is better than opa1612 as what I remembered. But two days ago I had some sort of discussion in another thread about cmrr etc. That got me thinking, opa1612 has super high CMRR across the frequency range up to about 100khz and very high open loop gain at 10khz. Why it made the measurements worse than with opa2156?

Today, I am thinking, probably it's just in my mind, maybe circuit had some issues or measurement equipment wasn't right, or I got fake opa1612 etc. Just now, I retested with new opa1612 and opa2211. The distortion of the amp into 250mv 8 ohm load at 6500hz with opa1612 and opa2211 is 0.00026%(they are pretty much the same chip) but when I measured with opa2156 I get 0.00009-0.0001% less than a half distortion at 6500hz (maybe even lower because that's about the limit of my measurement rig).

Could someone give me a explanation?
I assume it's because of fet input or something related to that. But cmrr should be what matters, right?
The parameters I took considerations are:
open loop gain especially at around 10khz
CMRR especially at around 10khz
THD of the opamp

Things probably matters but I don't have connections yet:
input impedance
low input/ source impedance of my circuit
470ohm Rf in the nested feedback network

Please, someone give me some insights, thanks in advance.

 

[JohnYang1997]

Another side information: yesterday I measured jds atom which has njm2068 wrapping around lme49600 instead of lme49720. The performance under load is pretty much the same as opa2156 and is better than what I measured from el amp. Njm2068 doesn't have jfet input nor have super high loop gain or anything.
Summing of information:
njm2068 and opa2156 lower distortion
lme49720 and opa1612/opa2211 higher distortion

 

[SIY]

Schematics would be nice.

edit: Also, at these levels, which are pushing measurement capability, exact layout of circuit and test probing is quite critical.

 

[JohnYang1997]

Schematics would be nice.

edit: Also, at these levels, which are pushing measurement capability, exact layout of circuit and test probing is quite critical.

They are pretty much like this. 49600 is basically the circuit of atom and el amp.

 

[JohnYang1997]

Should I probably move the thread to audio electronics and hardware subforum?

 

[SIY]

Are you really running these at 18V, pushing the max specs (and hence maximizing thermal modulation)?

Same chips for each measurement? I mean the chip itself, not chip type.

No stoppers or inductors at the op-amp input?

Investigated changes in layout? Removing the potentiometer?

 

[JohnYang1997]

Are you really running these at 18V, pushing the max specs (and hence maximizing thermal modulation)?

Same chips for each measurement? I mean the chip itself, not chip type.

No stoppers or inductors at the op-amp input?

Investigated changes in layout? Removing the potentiometer?

Not really just convenience for simulation. I am actually using +-7.4v for this atm.

 

[JohnYang1997]

Tried a few opa1612 before had pretty much the same results. Today the opa2211 and opa1612 performs identical.
No stoppers or inductors in the input for tpa. The atom has another gain stage work as a input buffer.
The layout is pretty much optimal.
Don't think disabling pot would do much.

 

[SIY]

"Pretty much optimal." Many sins can be hidden there, especially when you're trying to rationalize ppm. Likewise potentiometers,, even with a few nanoamps of bias or offset current being modulated.

At ppm, it's not just the circuit, it's the physical implementation. And details like stoppers for the 49710. All *can* make a difference at those levels.

 

[solderdude]

I don't see any decoupling of the opamps. This (and routing of the traces) is kinda critical.

 

[JohnYang1997]

"Pretty much optimal." Many sins can be hidden there, especially when you're trying to rationalize ppm. Likewise potentiometers,, even with a few nanoamps of bias or offset current being modulated.

At ppm, it's not just the circuit, it's the physical implementation. And details like stoppers for the 49710. All *can* make a difference at those levels.

Well of course I understand the importance of implementation and pcb physical routing. The potentiometer is 1kohm pot. And the feedback loop of tpa6120 itself is optimal just as shown in the data sheet. The opamp is just a few mm away from tpa6120. The global feedback loop is fly wired point to point directly. I did optimized everything I could here.

Same thing happens to jds's design too. Not just mine. Due to physical structure lme49600 implementation is more optimal than tpa6120(due to nc pins at the bottom).

Btw I just tested njm2068 with tpa6120. The result is really close to opa2156 probably a hair higher distortion at about 0.00009%-0.00013% which is still half of that of opa1612.

Well even if the implementation is not optimal, why aren't they all perform non-ideally but opa2156 and njm2068 is better? I still don't know why either.

 

[JohnYang1997]

I don't see any decoupling of the opamps. This (and routing of the traces) is kinda critical.

2.2uF is used for opa1612 and 10uF is used for tpa6120. I don't think I have to show these kind of stuff right?

 

[SIY]

I don't see any decoupling of the opamps. This (and routing of the traces) is kinda critical.

Indeed. At ppm, physical implementation becomes important, as well as things like component choice (e.g., VCR) and details like stoppers and decoupling. If that's waved away, then the idealizations of the models become less relevant.

 

[JohnYang1997]

Indeed. At ppm, physical implementation becomes important, as well as things like component choice (e.g., VCR) and details like stoppers and decoupling. If that's waved away, then the idealizations of the models become less relevant.

Ok can we skip my amplifier. Let's talk about jds el amp and jds atom. Basically the only difference is the opamp. And atom performs better.
About the implementation here is the comparison.
Also can we just think that the performance should be like that and actually talk about why opa1612 lme49720 isn't that great for this application. Not only I have measured this type of results. On lme49600tsbd datasheet, ti showed much higher distortion at 16ohm and higher frequencies.

 

[SIY]

Since I'm unfamiliar with these products, I'll bow out then. But leaving you with the warning that if you're trying to parse ppm (for reasons that escape me), "basically" doesn't cut it. Devil's in the details.

 

[solderdude]

PCB layout, ground planes, capacitor choice, compound caps or not, power supply, wire routing, component tolerances.
Even with the exact same components you can get (small) differences in performance.
I think it's all moot when the numbers are all reasonably close together.
0.0005% of the same type of distortion and 0.0001% may be measurable and give nicer distortion and SINAD values but in practice don't make any difference.
These number differences can easily come from layout and component choices.

Of course there is nothing wrong with trying to design for the best performance possible.

 

[JohnYang1997]

PCB layout, ground planes, capacitor choice, compound caps or not, power supply, wire routing, component tolerances.
Even with the exact same components you can get (small) differences in performance.
I think it's all moot when the numbers are all reasonably close together.
0.0005% of the same type of distortion and 0.0001% may be measurable and give nicer distortion and SINAD values but in practice don't make any difference.
These number differences can easily come from layout and component choices.

Of course there is nothing wrong with trying to design for the best performance possible.

Questions is the opposite. The board is the same, but using different opamps. You can argue that 2068 perform exactly the same as opa2156 but the difference between opa1612 and opa2156 is really big.
Also about component tolerances, i tested a few opa1612 before and today both opa2211 and opa1612 performed just identical. And I swapped opa2156 the left and right channel, the performance is consistent. So I think we can ignore the component tolerances.

 

[solderdude]

When all remains the same but different results on different days could point to RF influence or measurement errors.

 

[777]

The answer is very simple, it's all about bandwidth of the driver. You must to over compensate for OPA1612. Now you have a burst distortion at high frequency. Have you a very good oscilloscope or spectrum analyzer ? Your power supply is liniar or smps type ? What does a step look like ?

 

[JohnYang1997]

The answer is very simple, it's all about bandwidth of the driver. You must to over compensate for OPA1612. Now you have a burst distortion at high frequency. Have you a very good oscilloscope or spectrum analyzer ? Your power supply is liniar or smps type ? What does a step look like ?

Two problems.
1, the circuit is nested feedback/composite. And it has an output inclusive feedback. Openloop gain up to 10khz is identical to no compensation but still stable.
2, opa2156 works much better than 1612. I did tests other opamps. The pattern is not very clear. Some does go at level of opa2156 like opa189, opa1652, opa828, njm8080, njm2068. And the others, like 5532, opa211, opa1612, 4562, ad797 aren't as good.
There isn't simple pattern that cmrr or openloop gain or anything that is consistent. What's somewhat consistent is that fet input low distortion opamp from ti generally work great.