Big test of DIP dual op-amps for audio: LM4562, NE5532, OPA2134, TL072 and LM1458

[LTig]

The possible snags here are used DAC/ADC quality and music for the test files and the actual performance of gear used to compare the test files at home.
If the quality is lower than that of the test files what is one actually comparing when the technical differences (that are guaranteed to exist) are masked by lesser reproduction gear.

Data sheets of low distortion opamps usually contain specially designed circuits with much reduced loop gain ( e.g. by a factor of 100 or 1000) to measure distortion with standard equipment as the best opamps are better than an AP in low gain circuits.

 

[solderdude]

Indeed... so they can publish the acquired/derived data in data sheets to show performance.
The problem with testing opamps using sound cards and even AP without specialized pre-amps, notch filters etc. is that one may be looking at limits of the test gear as well as actual signals.
For that reason it makes sense to publish a loop-back of the measurement gear to see where the limits are.

 

[Salt]

The problem Amir described is not caused by the op-amp used but rather the entire amp and the limits of the entire circuit.
Especially switching devices (like class-D) can easily have/cause measurement errors when looking in the microvolts and volts levels of measurement signals happening at the same time.

It is a DAC to ADC circuit which is Under Test here, and same test circumstance for all OPA ...

 

[solderdude]

It is a DAC to ADC circuit which is Under Test here, and same test circumstance for all OPA ...

We won't know until the loopback results have been published.
The best op-amps may well be near, or even below, test equipment limits.

In case of Amir's tests it is very obvious that the entire amp is the limiting factor here and not the AP test system.

 

[restorer-john]

Some years ago I carried out similar tests but all at 1kHz and into a 2.7k load. I still have most of the ICs so I can carry out the same measurements as PMA (although possibly my ADC might be a limiting factor.
I also have the Signetics opamps as mentioned by @Zek

OPAMP tests into a 2k7 load

OPAMP tests into a 2k7 load

gtkc.net

Wonderful, but what about the loopback of the measurement system itself? We've got D/A distortion, A/D distortion, DUT distortion and the various effects of reinforcement and cancellation of harmonics all at play.

I do love the look of win7/Vista though- God I miss that beautiful UI.

 

[restorer-john]

@solderdude posting at the same time.

 

[Gorgonzola]

For whatever much or little it may be worth, atm audio tested these several op amps in context of designing their I/O buffer for the Hypex and Purifi amp modules:

• LM4562
• LME49720
• MUSES01
• Burson Audio V5
• Sparkos Labs SS3602
• Sonic Imagery Labs SIL994

The produced the harmonic spectra for each of these that may be found HERE.

Unfortunately the only overlap with @pma's testing was the Lm4562.

 

[audio_tony]

Wonderful, but what about the loopback of the measurement system itself? We've got D/A distortion, A/D distortion, DUT distortion and the various effects of reinforcement and cancellation of harmonics all at play.

I do love the look of win7/Vista though- God I miss that beautiful UI.

I'll try and post the loopback (along with additional tests) in the coming week.

 

[Sokel]

That is the point of the Douk thread.
However, the test here is about measuring performance limits in simple audio circuits using some well known op-amps.

Most 'audiophiles' don't care about performance and want to hear recordings of op-amps.
The possible snags here are used DAC/ADC quality and music for the test files and the actual performance of gear used to compare the test files at home.
If the quality is lower than that of the test files what is one actually comparing when the technical differences (that are guaranteed to exist) are masked by lesser reproduction gear.

My opinion for testing with music is that it would have to be done in lab conditions and controlled with suitable gear.

For technical measurements the whole ear and reproduction gear is not needed but that's what audiophiles are more interested in.

Here's an attempt with music, OPA is LME49720 at the mid-high part of an x-over, so take that at account too, useful is probably >300Hz.
It's more of a sanity test and a question of legitimacy, so thoughts about it.
Test conditions are visible, results are both at dBFS and percentage were levels have to be taken into account too.

Make whatever you want from it:



Conditions



dBFS




Percentage

Edit:the importance of level is really interesting, see how the percentage plot looks like the mirrored image of the track's FR.

 

[xanalog]

Thank you, @pma,

...and OPA2134, TL 072 (JFET input).

Ships with Topping D10 and

>> @Eiffel replies >> …OPA 2134 is just a 2132 who failed the frequency test at the factory, but it's still able to go up to 100kHz.

<$5...

 

[solderdude]

For whatever much or little it may be worth, atm audio tested these several op amps in context of designing their I/O buffer for the Hypex and Purifi amp modules:

• LM4562
• LME49720
• MUSES01
• Burson Audio V5
• Sparkos Labs SS3602
• Sonic Imagery Labs SIL994

The produced the harmonic spectra for each of these that may be found HERE.

Unfortunately the only overlap with @pma's testing was the Lm4562.

The Sparkos Labs SS3602 ans Sonic Imagery Labs SIL994 outperform the DIP versions.
All are well below audible thresholds for music (at least at the super easy to perform well 1kHz sine wave).
These and some other modern op-amps as well as discrete ones is what the (swapping) public is mostly interested in but they would rather read how much wonders they do for the sound with different colorful descriptions for each device. That's what the public really wants.

Multitone, S/N ratio numbers and FR would have been welcome. FR would be determined by the output stage without doubt.

 

[KL....]

For whatever much or little it may be worth, atm audio tested these several op amps in context of designing their I/O buffer for the Hypex and Purifi amp modules:

• LM4562
• LME49720
• MUSES01
• Burson Audio V5
• Sparkos Labs SS3602
• Sonic Imagery Labs SIL994

The produced the harmonic spectra for each of these that may be found HERE.

Unfortunately the only overlap with @pma's testing was the Lm4562.

Thank you and the following is a little from there/HERE :=)

• Input buffers….
• Buffers are electronic circuits with unity gain (which means that voltage level is the same at both the input and the output) that provide electrical impedance transformation from one circuit to another, with the aim of preventing the signal source from being affected by whatever currents the load may be produced with.
• The main element of the buffers is the operational amplifier, whose main characteristics are the very high input impedance and the very low output impedance. These features, together with the rest of the passive components of the circuit, prevents a large amount of energy from being extracted from the source, eliminating high disturbances caused by the overloads (Edit/Me: also called Swamping). In audio circuits, this translates into greater signal integrity, preventing high frequency details from being lost and resulting in a signal with more “color and warmth”. In short, a buffer makes the Hi-Fi system more pleasant to listen to.

Using Sherlock Holmes Principle…. This implys, for the Douk A5, that the circuit (which includes the swapable opamp/NE5532) needs to feed the ClassD circuit with compatable impedance, voltage, and current for the ClassD circuit to perform at its (absolute) best/highest capability…. this is reasonable, isn’t it? So is the feeder circuit (the circuit which includes the swapable opamp/NE5532) doing this? As suggested, well no, not until we know the best/highest capability of the ClassD circuit.... that is reasonable, isn’t it?

Thankfully, @Amir has provided (accurate) measurements for us which provides a measure of the Douk A5 as sold, which is fabulous, isn’t it? Nonetheless, the Douk A5 allows for opamp swamping and even encourages it with the Lid to provide easyer access, doesn’t it? Douk A5, by doing this, could be suggested as reckless in that they have not provided guidelines and circuit diagram, for the Swaping process, have they? Even a disclaimer, ‘Swaping compatable opamps is permissable but you do so at your Risk’, it could be suggested, will encourage those to provide consideration to/with the opamp swaping process, wouldn’t it?

It could be suggested that the following (and more than these) needs to be considered when swaping….

• Will the opamp be compatable with the circuit where it is being placed?
• When placed, will the resulting circuit be less or more compatable with the circuit (classD) that it is feeding, and by the unit/circuit feeding this circuit. If more compatable/fabulous expect better N/D measurements, if less compatable (instability/swamping/overload) expect worse N/D measurements…. this is a reasonable analogy, isn’t it?
• Just applying the above requires considerations to the circuit, the opamp, the unit/circuit feeding this circuit, and the classD/circuit being fed by this circuit, doesn't it?

 

[pma]

Thanks, great contribution.

What power supply voltage have you standardized on using? Gain compression would be a useful measurement.

The NE5532 was designed into the Focusrite Studio Console by Rupert Neve and was popular with many other professional recording equipment makers.

It would be fun to test some of the boutique op amps, though they may have problems with part-to-part variation. Your test may be the first proper test that part has experienced. It would also be fun to test a 741.

The NE5532 was released in 1979, linear audio circuitry has made almost no progress since then.
1458 is a dual version of 741, so it is covered in the tests in post #1.
Power supplies are regulated +/-15V, standard LM7815/7915 circuitry. There is a good bypassing to the ground plane, see the photo in post #1.
Just to add, second channel of the test rig is to test with -1x gain and noise gain of 86dB. The measurements have not been posted yet with this option. They are probably of not much interest of the audio user.

@pma, thank you for testing, nice surprise with the NE5532, isn't it?
From these tests, do you expect the discrete opamps (Sparkos, Burson, Sonic Imagery) to have lower performance than the NE5532?

I checked the Sparkos datasheet and I find it incomplete, not so well defined as TI, AD, LM, LT datasheets. There seem to be some good and some poor parameters, like very high input bias current.

Well done Pavel.

Results are THD only. How about absolute noise figures for each device (20-20k WTD or UNWTD)?

John, I will measure the noise separately. I assume it will just confirm the datasheet parameters. I Can make separate measurements of THD and noise with good resolution, but I cannot compete in THD+N with the @amirm AP.
Edit: John, the noise of all tested op-amps (except for 1458 and TL072) in the +6dB test rig is so low that I am not able to measure it reliably. I will need to use my +30dB measuring amplifier and it will take some time.

Can you show a loopback ... to get a feel of the measurement floor. ?

This is a good and logical question, but the answer is not simple. If you check my test circuit in post #1, you will see that load impedance is 1kohm. It is not so unreasonable, because the SOTA preamps like Topping Pre 90 have balanced input impedance of 2kohm, which makes exactly 1kohm load per one branch of the balanced output of the signal source.
The distortion of the op-amp DUT is thus affected by this 1kohm load and now it depends on op-amp output stage design and crossover distortion. To make long story short, below is a comparison THD vs. frequency at 1.76V output voltage for the loop (Topping D10s + E1DA Cosmos ADC 4.5V range) compared to the same system with OPA1622 buffer inserted into the loop. This buffer has 2kohm input impedance, and we can see that HF distortion is now slightly reduced due to lighter load to D10s output.

Now, below is the comparison of THD vs. frequency of the loop (D10s-Cosmos) and the +6dB test circuit (post #1) with the LM4562.
Just one more note: I have some stock of LM4562 genuine parts (Mouser, I do not use AliExpress or similar channels) and the distortion results are consistent. I have also checked LM49720 vs. LM4562 and there is no difference.

So - now we can see the degradation of distortion in the test circuit with +6dB gain. Just as expected.

The test results also lead to a conclusion that NE5532 output stage has more capabilities to drive low impedance loads than the newer LM4562, which reflects in HF distortion.

All the measurements done with 96kHz sampling and 45kHz effective bandwidth.

 

[pma]

Results are THD only. How about absolute noise figures for each device (20-20k WTD or UNWTD)?

I have just measured the noise, using my +30dB measuring amplifier, because the noise of DUT opamps was too low to be measured directly. As a reminder, test circuit is shown in post #1. it has gain +6dB and 10k/10k feedback resistors (important info).

Total 20Hz - 20kHz unweighted noise
LM4562: 3.03 uV
NE5532: 2.70 uV
OPA2134: 3.06 uV
TL072: 4.75 uV
1458: 7.22 uV

Please note that feedback resistors are 10k/10k, this makes 5k seen as a noise resistance. The result of LM4562 is worse than NE5532 due to the fact that input current noise density is 1.6pA/rt(Hz) compared to 0.6pA/rt(Hz) of the NE5532. Again, as in the case of MM preamps, we can see dominance of the current noise above voltage noise, if the feedback total resistance or generator impedance is above 1kohm.

I can conclude that NE5532 is, to me, better universal op-amp design than LM4562/LME49720. The later needs lower values of resistors in the FB net and also lower generator impedance to get any advantage. Almost same applies to OPA1611/12, which has very low voltage noise, but not low current noise. That is the reason why Topping and other manufacturers are pushed to use low input impedance of their preamps, if they are oriented to the lowest distortion.

 

[audio_tony]

@pma the graphs in your OP are THD vs output level, which means the linearity of your DAC is also going to influence the THD figures, not so?

You are measuring from 0.005 vRMS to 2vRMS. Assuming your DAC outputs 2vRMS @ 0dBFS - this means the starting point is approx. -52dB - so how linear is your DAC at such a low output?

Surely it would be better to measure THD vs frequency? Or perhaps limit the minimum level to -6dB or so instead of -52dB?

*unless you are using an autoranger which maintains the DAC output at a steady 0dBFS?

 

[Cbdb2]

For whatever much or little it may be worth, atm audio tested these several op amps in context of designing their I/O buffer for the Hypex and Purifi amp modules:

Not worth much. Thers this BS to start. "However, despite of all these advantages, they are practically devoid of any sonic signature so they require an external input buffer to tuning in a “warm house sound”, customizable according to tastes and requirements."

And get this, $300 for a tiny pair of empty PCBs. And if you want them to add $5 worth of components that will be $450. Add 2 $50 unnecessary discrete opamps and that will be $750.
Is it any surprise there best measurements are for the most expensive option. They probably thru out the NE5532 measurements.

 

[Cbdb2]

I can conclude that NE5532 is, to me, better universal op-amp design than LM4562/LME49720. The later needs lower values of resistors in the FB net and also lower generator impedance to get any advantage. Almost same applies to OPA1611/12, which has very low voltage noise, but not low current noise. That is the reason why Topping and other manufacturers are pushed to use low input impedance of their preamps, if they are oriented to the lowest distortion.

Hey Pavel, great job, a couple questions.
Just came across this from TI, a new zero crossover OA.

https://www.ti.com/lit/ab/sboa181b/sboa181b.pdf?ts=1742833198766

Is this just there name for a class A OA? Got me wondering if you ever measured the difference biasing a OA into class A (with a resistor from output to the positive rail) makes.

 

[DualTriode]

There are questions if it matters which type of op-amp is used in the audio amplifier and my answer is that it depends on circuit schematics and op-amp parameters. Some of those parameters are obvious from datasheets, but others may be quite hidden and not clear from the first view. For the test I have chosen well known dual op-amps LM4562, NE5532, LM1458 (bipolar input) and OPA2134, TL 072 (JFET input).
LM1458 is one of the first op-amps usable for audio and it is dated at sixties of the previous century. NE5532 was then a big big improvement in audio op-amps, introduced by Signetics company in 1979. LM4562 appeared in 2006 as an improvement in noise and supposed to have lower distortion than NE5532.
OPA2134 was introduced by Burr-Brown in the nineties as a lower noise and lower distortion improvement of TL072.
As far the datasheet and popular info.

When making a choice of the op-amp, we always need to know the application circuit. For the test I have chosen quite common non-inverting amplifier circuit with +6dB gain and 10k feedback resistors.

View attachment 438490

The test rig is driven from a DAC with low, 20 ohm output impedance, and loaded with 1 kohm ADC input impedance. This is quite difficult load, but as the output voltage did not exceed 3V, all the op-amps are deep in the area of allowed output current.

Measurements

For the start, I have chosen measurements of THD vs. output voltage at 1kHz and 10kHz and measurements of 13+14kHz CCIF IMD vs. output voltage. The plots are shown below:

View attachment 438493

View attachment 438494

View attachment 438495



We can see that the LM1458 is simply unusable for audio, even at low output voltage and that we can assume it will have its own sound signature. Both JFET opamps, OPA2134 and TL072 have higher distortion than LM4562 and NE5532, in all 3 tests. The result of the old NE5532, that is slightly better than LM4562, is a small surprise to me, I have expected the opposite result.

In all cases, I have tested at least two samples of each op-amp type, to prevent random results. All the samples of the same types behaved identically within the measurement repeatability.

More tests can be done, suggestions welcome. But I will wait for the forum feedback.

All the measurements done with 96kHz sampling and 45kHz effective bandwidth.

Note: for loopback test please go to post #33

Hello,

Thanks for the interesting thread.

I am a bit puzzled by limiting the Op-Amp output voltage to 3 volts. What is the reasoning?

The typical Op-Amp with 15 volt + / - rails will output close to 10 volts 10 volts + / -.

Thanks DT

 

[audio_tony]

Hello,

Thanks for the interesting thread.

I am a bit puzzled by limiting the Op-Amp output voltage to 3 volts. What is the reasoning?

The typical Op-Amp with 15 volt + / - rails will output close to 10 volts 10 volts + / -.

Thanks DT

I think it's to limit distortion as the OP is driving a 1k load, which some of the OPAMPS won't be too happy with at large voltage swings.

 

[pma]

Hello,

Thanks for the interesting thread.

I am a bit puzzled by limiting the Op-Amp output voltage to 3 volts. What is the reasoning?

The typical Op-Amp with 15 volt + / - rails will output close to 10 volts 10 volts + / -.

Thanks DT

- in a typical line stage opamps work up to 2Vrms. Balanced 4Vrms are a difference between OUT+ and OUT-, both again up to 2Vrms.
- for higher level I would need to add gain stage behind the DAC, which would lead to increased noise and distortion.