Opamp survey

Status
Not open for further replies.
Jump to Latest
V

Voegelchen

Member
Joined 2010
aca-vogel.de
#1
#7
"Newer" devices

Hi folks, thanks for reading and comment it.
Like written in the pdf, for most "new" opamps, there are many diagramms given in the datasheet, all done in the same way my tests were done.
The only reason to do it on the "old" devices was catching up comparable THD figures because they are missing in the datasheets.
Perhaps, I have some luck and there exist an OPA627 sample on my workspace 🙄

Hint: The OPA627 has a laser trimmed input stage, so things work there almost ideal. That's the reason why it's so good and why it's so terrible expensive. Compared to the other devices, especially the LM4562, it give almost no benefit regarding THD, IMD and noise.
[EDIT]
Something the datasheet don't tell you: The noise spectra.
If anything about noise is given, it is a number at 10 Hz and at 1kHz or a diagramm about that, but rarely there are any specs about the spectral density. Only in the very special (and old ?) low noise parts like OPA27, a noise spectrum or better a scope trace is given.
So noise differs by hearing, especially in the range above 1kHz (where you can hear it, below 1 kHz the tweaters doesn't radiate anything and the amplitude below 1kHz is too low).
There are some "hints", never written to paper, that some very modern opamps like LME4xxx have some sort of popcorn noise, especially if they warm up. Thats fully different from thermal Johnson noise.

Perhaps I should start a study about it 🙂
 
Last edited:
#8
The datasheet curve of the NJM4580 shows the gain and phase of an amp set up for 40dB closed loop gain, it's *not* the open loop gain plot. You need to extrapolate the true open loop gain from that.

Above 50kHz or so the closed loop gain begins to fall apart and turns into the open loop gain asymptotically as there is no more gain left to hold up the 40dB. At 100kHz the gain is about 37dB so the conclusion is that the open loop gain at 100kHz is about 40dB, following the typical 20dB/decade slope until it flattens at a lower frequency when the open loop gain tops out at 110dB (as given in the main table of the spec sheet).

Hence your values and conlusions for the NJM4580 are definitely wrong.

The open loop gain is around 60dB at 10kHz, which is pretty much the typical value for an opamp of that type (dominant single pole) and a unity gain bandwidth of 4Mhz or so....

Same thing for the NJM2068.

Please update your document to correct this rather severe misinterpretation of the mfgr's graphs.
 
Last edited:
#9
Open Loop misinterpretation

Thanks for your advice, you're absolut correct. The tables were updated.....

Corrected Version:
View attachment Audio Opamps, fact, no myths and THD measurement results.pdf

Yeah, NJR is a bit strange with their datasheets :scratch:...
Noise values given as a RIAA amplifier output voltage 😱... nobody else do that...
Open Loop Gain as Amplifier gain 😕 ....
THD as number with gain and THD as graphs also with gain... but only one gain and one voltage :scratch:

Others do better, even in the 2000++ years, more and more datasheets contain more (easy readable) curves...
 
#12
abraxalito said:
George do you have any ideas why OPA627 sounds so good? Can't be its THD measurements can it?
Click to expand...

One possible theory is that the JFET input stage is dielectrically isolated from the substrate, rather than junction isolated. So, there is no modulation of the parasitic drain to substrate capacitance when there's common mode signal applied to the amp, which happens with a non-inverting stage.

Again, just a theory, but that form of nonlinear capacitance is something I find to be pretty annoying, and very hard to avoid unless you use inverting stages.
 
#13
I too have a hunch that dielectric isolation has something to do with it because my all-time favourite sounding opamp (HA-5222) is built on a DI process. But its bipolar and I'm skeptical that the reason is to do with non-linearity of device capacitances. I tend towards the view it may have something to do with isolation from substrate noise, just speculation on my part though.
 
#14
abraxalito said:
I too have a hunch that dielectric isolation has something to do with it because my all-time favourite sounding opamp (HA-5222) is built on a DI process. But its bipolar and I'm skeptical that the reason is to do with non-linearity of device capacitances. I tend towards the view it may have something to do with isolation from substrate noise, just speculation on my part though.
Click to expand...

A bipolar input pair isn't so different than an implanted JFET pair. A bipolar input pair collector is made by diffusing onto a junction isolated well, which has a capacitance to the substrate, just as the JFET drain is diffused onto a junction isolated well. While the substrate is fixed at V-, the other side of that capacitor is the collector (or drain), and it forms the same sort of nonlinear, parasitic capacitive load.

What makes this a problem is that in many circuits that try to maximize first stage gain, the drains/collectors are loaded into high impedance nodes, and this exacerbates the effects of the nonlinear parasitic capacitance.

A great answer to this is a low impedance cascode collector/drain load, which minimizes the voltage variation on the drain/collector, which minimizes the capacitance change of this nonlinear capacitor, while still allowing the first stage to provide the gain it's supposed to provide.

Amps like the AD797 that use a folded cascode, but do not use DI devices, can still avoid this nonlinear substrate capacitance problem, because the collectors/drains are loaded into a low Z, which minimizes the voltage across this nonlinear capacitor.

I guess I'm suggesting that yes, this nonlinear load is a problem, but it can be solved with DI device isolation, or by clever cascode loading. My 'belts and braces' approach is to try to find amps that get this right, and then use them only as inverters, so the problem never gets expressed.
 
#15
A/B switching the input of bootstrapped supplies for non-inverting config isolates the input capacitance effect in listening tests / measurements. I tried with opamps that have a bad reputation for this like TL071 but found it very hard to here any difference... may have been masked by some bigger problems of these chips.
 
#16
Hard to tell what was going on without seeing the circuit at hand... if I wanted to provoke input impedance distortion, I'd go for lowish supplies and provide external buffering for the output (so the notoriously wimpy output stage doesn't play a role), and a nice big impedance imbalance of course. I wouldn't expect distortion to jump out right away, given that it should be dominant 2nd... might take some IMD testing.
 
#17
Thanks for all this work, Voegelchen.
I had looked at the datasheets but the different input impedences of the noise specs, and the different units quoted, had me confused.
I have equipment with 4 of the ICs you tested and ear results using speakers parallel yours:
4558- hissy fuzzy
33078- okay. (replaced 4558 in RA88a mixer)
4580 - okay, capable of driving long twisted pair lines from sound booth to stage amps. .02% HD in spec book for system. (Peavey Unity 12 mixer)
lm4562 - okay, pricey
So numbers & ear results match up. NE5532 was not in stock when I was looking around to replace the 4558's, nor was 2068. 5532 would probably require more parts and board space in the power supply than the two 8v zeners and two resistors I'm using to supply 4 dual op amp packages.
In addition to above, equipment with NJM4560 sounds okay, Swapping with 33078 makes no difference through my speakers. (CS800s amp, SP2=XT speakers)
 
Last edited:
#20
Monte McGuire said:
A bipolar input pair isn't so different than an implanted JFET pair.
Click to expand...

Not true a JFET in a standard planar process has a back gate junction which is a non-linear capacitance. DI has real oxide hence in the ideal case no non-linearity. Some oxide isolated processes have an isolation that can develop a small depletion region and can have some residual non-linearity.

The non-linearity with high unmatched source resistance is almost entirely due to input gate capacitance.
 
Status
Not open for further replies.
Top Bottom