Big test of 13 op-amps in inverting configuration with gain = -1 and non-inverting configuration with gain = +1

[pma]

More measurements, now with gain = +2, op-amp under test is followed by BUF634 high speed unity gain buffer (inside FB loop), output load is 1kohm. Several op-amps are tested and as a reference overlay there is always the result of OPA627 + BUF634.

I have repeated this measurement in the same setup, but now with gain = +4. Output voltage is 8Vrms then, so the large signal non-linearity is enhanced, and lower loop gain plays its game as well. The results of op-amps are similar as before, with one notable exception. I have added AD797 to the test group, which could not be done before due to stability issues at low gain. AD797 is again compared to reference OPA627 and the AD797 is a clear winner.

Measurement is done with 90kHz BW and reveals excellent linearity of the AD797 at higher and high frequencies, but also at low and mid frequencies. This is completely in conformance with results of Samuel Groner:

https://www.nanovolt.ch/resources/ic_opamps/pdf/opamp_distortion.pdf

Analog Circuit Design · Samuel Groner · Resources · IC OpAmps

Again, this thread is not only about audibility (which would be a subject of extremely well conducted DBT tests), but about best technical parameters.

 

[antcollinet]

It is a myth that op-amps make no difference

It is a myth that people are saying they make no difference.

What people are saying that is that op amp rolling is a pointless exercise** - which is correct when one or more of the following is true:

• Device performance is dominated by the power stage such that better op amps cannot improve it. Typical of many or most power amps. Certainly typical of the low cost chip amps where op amps are socketed to allow rolling.
• The design of a device has already selected the best op amp to work with that circuit.
• A device is already audibly transparent with the designed op amp - in which case no audible improvement can be made.

One or more of these is true in nearly all cases where op amp rolling is carried out. IE with any half way competent design.

You should also note people frequently point out that while no audible improvement can be expected - it is quite possible to make the performance worse by swapping op amps. IE - of course op amps can make a difference.


** They might use the words "op amps will make no difference" as a short hand under the specific operating conditions. (Context is everything)

 

[solderdude]

Not to mention the aspect @pma always mentions... not all op-amps are suited and may need better/different decoupling or require a miniman amount of gain.

Pavel knows all of this and merely shows all op-amps when used 'properly' do have different performance. He also mentions audibility is another thing.

The latter is what the rather silly 'op-amp rolling' is all about. It isn't about 'technically improving' anything but rather fooling around with 'tech' and 'listening for sound quality improvements' in a highly uncontrolled and subjective manner.

 

[Scytales]

Many thanks to PMA to share the results of this experiments! Highly appreciated. Kudos to the done efforts!

May I suggest another experiment to all that are both interested in and able to implement it: proceed with similar comparisons using suitably chosen inter-modulation test signals.

Inter-modulation distortion products, when test signal is suitable, fall in the audio band and are notoriously more objectionable than pure harmonic distortion products, especially when the latter fall outside the audible range. Thus, IMD tests may be more enlightening for the layperson.

 

[DonH56]

IMD directly correlates to THD so no real need for a separate test. Explaining how and why IMD matters is a worthy subject; here is a start:

Harmonic and Intermodulation Distortion

This post provides a quick comparison of how a distortion percentage relates to dynamic range in dB. That is, if an amplifier has 1% distortion, what does that mean? How soft (or loud) is 1%? Figure 1 shows a 1 kHz voltage signal with 1% second harmonic distortion (HD2) added. The spectral...

www.audiosciencereview.com

 

[wwenze]

He's got a point tho: We've always said "oh THD above 10kHz is irrelevant because 10kHz * 2 = 20kHz"

So even tho the THDs of 19kHz & 20kHz signals are outside audible range, the 1kHz IMD is inside audible. And since IMD directly correlates to THD, that implies we need to know the THD at 20kHz if we want to avoid the separate test.

 

[pma]

Yes. Though the relation between THD20k and CCIF 19+20k IMD is quite complex and not straightforward.

https://www.researchgate.net/publication/277870128_A_Comparison_of_Nonlinear_Distortion_Measurement_Methods

as we have more circuit related sources of distortion. However, THD20k may indicate to circuit susceptibility to HF intermodulation like from switching supplies and sources.

 

[DonH56]

He's got a point tho: We've always said "oh THD above 10kHz is irrelevant because 10kHz * 2 = 20kHz"

So even tho the THDs of 19kHz & 20kHz signals are outside audible range, the 1kHz IMD is inside audible. And since IMD directly correlates to THD, that implies we need to know the THD at 20kHz if we want to avoid the separate test.

Yeah, and even worse because the 1 kHz difference tones (at 18 and 21 kHz) are related to the third-order harmonic term so 60 kHz is also important. But it very much depends upon the circuit to determine the amplitude of the terms.

 

[Scytales]

DonH56, I acknowledge what you said.

But to infer inter-modulation distortion from a purely harmonic distortion analysis requires an intellectual effort the laypersons are not aware of (and they are some caveats, as said by PMA), whereas a straightforward IMD test requires no such effort and let the reader to simply see on a graph the kind of performance he can expect in the audio band. That's the only reason I made my suggestion.

 

[pma]

whereas a straightforward IMD test requires no such effort

It is good to do both tests. Sometimes (no big exception) CCIF 19+20kHz is rather related to THD 1kHz than to THD 20kHz. Like below, same DUT measured.

THD 20kHz, even THD 10kHz here is much higher than CCIF IMD 19+20kHz limited to 20kHz BW. There is no correlation. That's why they (Hypex, Purifi) measure CCIF 19+20kHz and NOT THD 20kHz. But please note that plots here above are from class AB amplifier, not from class D. One needs to make hundreds or thousands of experimental studies to understand the nuances. No cook-books here.

 

[Lars Risbo]

keep in mind that the feedback loop shapes all the distortion components. having high loop gain up to 20k ensures suppression of all audio IMD components. HD for a 20kHz tells not much about the audible range.

 

[pma]

Most probably it does not, but ....

Reducing measuring range to 20kHz (or so) makes "nicer" plots even in case of 19+20kHz CCIF IMD measurements, as it cuts second order additive distortion component, f1 + f2

For the amplifier with distortion rising with frequency, as shown in my previous post, we get much nicer IMD result (with BW limited to audio band), because f1 + f2 component is simply cut off. Then, it may look (looks in this case) like this:

I know you know, but this may not be a case for majority of readers here.

BW limited (24kHz) 19+20kHz IMD distortion is high (or a "tough" test) in 2 cases:

1. slew rate of complete amplifier is low (regardless class of operation)
2. class D amplifier has inductors in the output LC filter with a non-linear core and this non-linearity is not well corrected by the NFB. This is a case of many cheap class D amplifier and tough challenge for filters outside NFB loop.

Please note I am not speaking about audibility, for good reasons, not to slide to speculations level.

Similar explanation is valid for 3rd order distortion components -

 

[AnalogSteph]

At the end of the day, <20 kHz performance is what we as humans really care about. 10/20 kHz THD is handy for simulating conventional (AB) amplifier types as it tends to have good correlation with IMD performance in this case. Class D is another kettle of fish entirely in this regard, and may have much better in-band IMD performance than the THD would suggest (as the above example shows).

 

[Lars Risbo]

the difference between the ‘fish’ is that AB amps typically have only 1st order control loops (dominant pole) whereas modern class D has high order loops. A memory less nonlinearity (ie zero order loop) has a mathematical relationship between harmonics distortions and IMD. However, that gets more complex when there is feedback. Doing IMD tests therefore makes sense.