Another big opamp listening test
Some time ago I decided to build an AB box that would allow to connect two power amplifiers to same speakers by A-B switching. The AB box has had 2 independent link signal paths with 2 Alps potentiometers and opamp buffers, to match gain differences between the amplifiers under test. It also has had 2 relay switched power paths to link 2 stereo amplifiers with 1 pair of speakers. The AB box is built into quite well shielded metal Al box with top and bottom covers made from 3mm Al metal sheet. Clean stabilized power supply with TL431, opamp and 2 transistor regulator is used.
The link stage path is as follows:
Fig. 1. The test circuit
As a secondary use, this AB box may be easily utilized for opamp testing, provided that IC1 and IC2 positions are occupied by different dual opamps and 2 Alps pots are set at the same volume. So, I decided to make an opamp blind test for myself, after having made many opamp tests by IC swapping in the past (like 15 years ago), when "clear differences" were audible in uncontrolled tests. Frankly, after those 15 years my current audio system is much better tested and thoroughly measured, compared to that one I had 15 years ago.
So I had put opamps into sockets IC1 and IC2, matched volume of both paths within 0.1dB and started A-B testing. And - nothing was heard, both paths sounded the same. I was unable to tell the difference. So I dug out the oldest dual opamp I still have, the Tesla MA1458 (LM1458 equivalent, i.e. dual LM741) from about 1980, and put it in the test against LM4562. Again - almost nothing. With some music material, I thought I heard some difference, but was unable to confirm it by a valid AB test protocol.
Based on this experience, I decided to prepare the opamp test with 4 music files recorded through 4 different dual opamps. This test was then published at one of our local forums and at diyaudio.
The opamps tested were:
aa - Tesla MA1458 (LM1458 equivalent, dual LM741, 0.5V/us)
bb - National Semiconductor LM4562 * (see note at the end of this page)
cc - SGS Thomson TL072
dd - Burr Brown OPA2134
The D/A ==> A/D audio chain parameters
Let's show measurements of the D/A ==> A/D audio chain used and let's keep it as a black box, to prevent any possible negative or positive bias that would correspond to "names". 0dBFS = 168mVrms.
Fig. 3. THD at 1kHz (48kHz/24bit)
Fig. 4. CCIF IMD 13+14kHz (48kHz/24bit)
Fig. 5. THD vs. frequency at -6dBFS level (96kHz/24bit)
Opamp measurements
First, I made measurements with all opamps in the AB box to make sure there are no irregularities, additional noises or oscillations. All the opamps performed well, in agreement with their datasheet specifications. Distortion measured was negligible, below 0.001% even for the MA1458 (0.0008% at 1kHz), slowly rising above 3KHz to reach 0.01% at 20kHz for the MA1458. This is definitely inaudible. The recording signal level was quite low, 0dBFS = 168mV. This allowed at least some measurable distinction in parameters, and it was the noise, S/N ratio. With LM4562, the signal to noise of the whole D/A ==> opamp ==> A/D audio path was S/N = 113dB/1V/20kHzBW, unweighted. With MA1458, it was S/N = 107dB/1V/20kHzBW, unweighted. This related to 168mV was 97.5dB (LM4562) and 91.5dB (MA1458) for the whole test chain including D/A and A/D conversion. Still enough for 16bits with LM4562 and with a small noise degradation with MA1458. At least certain possible audible difference.
Measurements with LM4562 in the test chain
I have repeated the same set of measurements as for the sole D/A ==> A/D chain (Fig. 3, 4, 5) with the test circuit inserted into the chain, in fact exactly the conditions as during recording of the test files. The results for LM4562 are as follows:
Fig. 6. THD 1kHz with LM4562
Fig. 7. CCIF IMD 13+14 kHz with LM4562
Fig. 8. THD vs. freq with LM4562
We can see that the LM4562 buffer inserted in the D/A ==> A/D chain is doing almost nothing, no change of the parameters.
Amplitude response
The difference in amplitude response was unmeasurable for the 4 opamps, in audio band, however it was detectable when measured by a wideband noise up to 6MHz, high above the audio band.
Fig. 9. Wideband noise test of amplitude response
The music sample
I tried AB tests with opamps on many different music samples, from pop to classical. The indistinguishableness was the same for all genres, so I decided to use a sample with transients and clear woman voice, Rickie Lee Jones - Night Train. The original recording is of 1978, released 1979, appeared on CD in 1990 (from analog source). The CD track was ripped to the PC and then used as a source file for this test. Sent from the sound card to the test circuit (see Fig. 1.) and then digitized to 48/16 format wav (see below). This was done with all the 4 opamps. The volume was the same during all the recordings and after the files were recorded they were time matched and aligned to the same starting sample and file length. Such files were put as aa, bb, cc and dd to the forum test.
Average RMS value of the music sample used was -25dB. Peaks at about -4dB. The most silent part of the music sample was about -60dB RMS. Please see it in the image.
Fig. 10. Most silent part of the music sample
Because the RMS power in the most silent part was about -60dBFS, I decided to save disk and web space and used only 16-bit A/D conversion. This reduced S/N of the D/A ==> A/D chain to 93dB, but still enough for this music sample, as microphone path noise and analog source noise in the original recording is much much higher than that of the 16-bit dithered quantization.
File distinctions
All the files under test have the same average RMS level and total RMS level. Their length is always 2 490 060 samples. There are some small differences between maximum sample value and minimum sample value for the different opamps, which is mostly a result of small time/phase shifts between the samples of the files.
If we make a difference file between the "worst" opamp file aa (MA1458) and "best" opamp file bb (LM4562), using audio diffmaker software , we get the noise-like file, with no music content, with sample values up to +/- 5 LSB.
From http://pmacura.cz/optestdiffiles.zip you may download all 4 diffmaker difference files for all the 4 opamps used. Not all for all op amp combinations, but they all resulted in noise only.
Fig. 11. aa - bb
Spectral analysis shows some line frequency residuals and noise, not a trace of any music signal
Fig. 12. aa - bb spectrum. The only difference, except for noise, is in mains components, they come from the MA1458 poor PSR.
To evaluate possible audible difference by listening, the aa - bb difference file was amplified by 26 dB and you can download it from
http://pmacura.cz/diff1_26dBamplified.zip
In case you hear anything but noise, please let me know.
If we make the same diffmaker process with files from the 2 best opamps used, bb and dd, we get now noise again, which is mostly between +/- 3 LSB.
Fig. 13. bb - dd
So with the best opamps used in the test, we got just a little less noise as a difference between the 2 files. In spectral analysis, now the difference is just pure noise
Fig. 14. bb - dd spectrum
Conclusion
OK, so what we got from this test? It seems that the only difference between the opamps, when used as unity gain buffers, put into a shielded Al box and operated at hundreds of mV is a small amount of noise. Otherwise there is no audible difference between them, though in uncontrolled tests performed by opamp swapping there were seemingly significant audible differences. Even if that said, it still makes sense to use high quality low noise parts. This test only shows that their audible influence is sometimes exaggerated, especially in low-gain link level circuits. And though I performed a lot of uncontrolled tests in the past, I do not rely on them anymore.
The key of successful use of opamps in audio is not to push them at the limits of their specs.
I would like to add that a proper PCB design, wiring and shielding and a good power supply is a must. If these criteria are not fulfilled, we may "hear" some differences that would reflect external conditions rather than opamp distortion.
*Note: The LM4562 was quite susceptible to air-coupled mains line frequency noise, namely multiples of 100Hz. The cure is to place the PCB with this opamp in a shielded metal box and to connect the box with signal ground. Never use unshielded box neither with this op amp, nor with another audio circuits.
Pavel Macura, June 2017