• Welcome to ASR. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!

Amplifier Op-amp Rolling Part 2 [Video]

I agree about certain tube gear with intentional distortion levels (in the percent range) and undesirable high and frequency dependent output resistance that interacts with the impedance of a speaker. You could call that deliberate 'voicing' and there are people that prefer that 'sauce' pored over everything.
But op-amps that only will differ slightly in distortion in inaudible amounts above 1kHz, is not 'voicing' at all but a marketing strategy and makes use of how the human brain 'works'.

So this my take on all this Op-Amp rolling stuff:

Yes much of it is psychological, first comes Denial then Anger ......

Real measurements are important. Speaking of Op-Amps the only place I found numbers that stood out to me was rt Hz kind of sound floor measurements for high gain noise measurements of riaa Pre-Amps.

The OP the brought up voicing in general.

Personally, on my bench I have built more than several vacuum tube and jFet line level (pre-amp) circuits designed for voicing. See the plot below in the pdf. I still have that AP 2522 analyzer in the cage.


Note that this is not a power amplifier with variable output impedance that will interact with the speaker impedance.

Thanks DT
 
Last edited:
I would love to see any op-amp having that distortion profile and 0.5% distortion (+ a lot of IM which is not nice sounding).
One will find none of those.
This is the whole thing with this op-amp rolling endeavor in low gain circuits.
They are all better than 0.01%, even the crappy ones so there is no 'voicing' swapping op-amps nor will there be 'goodness' or 'lesser sound quality'.
This is the crux with this op-amp rolling to achieve 'better sound quality'.
 
Sure, if we're limiting the discussion strictly to opamps used as mere buffers or amplifiers like you said, then you're right. But I'm commenting on amirm's post, and I'm talking about the broader picture of 'voicing', including transformers and tube amps.
I have yet to see any company make playback gear with tubes and transformers claim they are voicing anything. They will instead, tell you how they have achieved higher fidelity and transparency. Voicing is a crutch they lean on when told their devices are full distortion and noise, or that they are not as good subjective as someone else's product.
 
Personally, on my bench I have built more than several vacuum tube and jFet line level (pre-amp) circuits designed for voicing.
And yet you have assiduously avoided any validation. Just more claims.
 
And yet you have assiduously avoided any validation. Just more claims.
My validation is the measurements that I have posted.

This afternoon I completed the assembly of a HP-22 PCB purchased from neurochrome. The PCB makes a nice PRE-Amp / OP-Amp test project.


The volume pot was replaced with bare wire jumpers. The output load is a Vishay RN65 1/2 watt 0.1% resistor, very low added noise and distortion.

The gain was set at ~ 10dB's (9.92 times). The input voltage was 1.00V.

I tested 2 OP-Amps, LM4562 nd OPA1656.

LM4562
SINAD at117.8 dB's
THD+N 0.000154%

OPA1656
SINAD 118.5 dB's
THD+N 0.000118%

The OPA1656 measures a little better plus has what appears to be significantly lower noise.
Is there a difference in sound between them? Did not listen, do not know.
The conventional wisdom tells us that listening tests are not required for OP-Amps. I am making no claims about any sound difference.

The OPA 1656 is becoming my preferred OP-Amp because of the reduced noise. See the plots below.

FFT Spectrum 4562.PNGFFT Spectrum OPA1656.PNG

Thanks DT
 
Last edited:
You made audibility claims. Once again, zero evidence.


Yes I could hear differences for Tubes and Jfets at Elevated line level outputs.

My posted measurements do the job well enough.

There is also a lack of audibility evidence supporting the no difference of audibility claims between/among Op-Amps. The posted measurements do the job well enough, perhaps with the exception of noise.
 
There is also a lack of audibility evidence supporting the no difference of audibility claims between/among Op-Amps. The posted measurements do the job well enough, perhaps with the exception of noise.
You can fix that by posting a recording through both op-amps and let people listen (pma also made such a blind test).

But ... that would not be the final word... you need to invest in several of the 'discrete' op-amps and post audio files of those too.

In my younger years I built a 'double' pre-amp.
One circuit used a single TL062 (even worse specs than the TL072) and another the, then hugely popular and 'best sounding', OP27 per channel.
The version with the OP27 had film caps paralleled with silver-mica (also was the FAD), compound decoupling, 1% metal film resistors and a well filtered power supply and was 'dual mono'.
The TL062 had 5% carbon film, 4 electrolytic (polar) caps in series with the signal all combining to the same capacitance and ceramic disc caps in feedback (they were said to have poor sound quality), simple 7805/7905 power supply and 100nF ceramic disc decoupling caps.
A switch on the back to select outputs.
The goal was to 'show' interested people what the difference in SQ was.
It worked great. Everyone including me could easily hear the difference...
After a 'demo' I accidentally left it in the 'worst sounding' position and used my stereo 'normally' and enjoyed the sound of the 'good pre-amp' for months.
Then, after a few months I found out my error... it was on the worst sounding position.

It triggered me to start testing blind. The audible difference was not detectable. Sighted testing was the cause of the heard differences.

Later on I had a Technics integrated amp compared to my own power amps. Could hear a difference, even in level matched blind testing.
Took out the tone generator and scope. Found out that the tone control dials in the center position did not lead to flat FR.
Used the tone generator to make the response flat (pots no longer in clicked center position).
blind tested again ... no more audible difference.

Then I got an old tube amp.. defective... got all the parts out. Build a cheap 4-transistor power stage into the chassis. Connected the heaters so the tubes glowed.
Let people believe they listened to either SS and the 'tube amp' and all but one person thought the 'tube amp' sounded best. That one person hated tubes.
Everyone could clearly hear the warmer 'tube sound' (where there was only SS)

Moral of the story ... do not test sighted IF you want to know the truth. Only test blind and level matched.
 
@solderdude : Yeah, it is not easy to prepare a valid test with 2 identical channels. Working on it, by accident. With coupling caps, 2 channels with AD797 and 2 ADC channels of E1DA Cosmos I am able to get this matching:

FR_2xAD797.png


2 channels are driven by identical signal sweep. One output is a reference and the other one is plotted against it. They are measured and recorded simultaneously. BTW, if you plot only one channel without a reference, all you see is a FR error and ripple of DAC/ADC filters. The small error seen here above is due to non-equivalence of coupling capacitors values and it comes exclusively from Cosmos ADC, not from the test rig.

Now if I change one of the AD797 for TL071, all else being equal, I get this:

FR_797-071.png


The biggest deviation is above 20kHz, but some tiny one is at lowest frequencies as well.

Interestingly, if you try Deltawave on these resulting files, it increases the audible error, compared to direct subtraction of these simultaneously recorded files.

P.S.: the difference in time stamp is a result of date of creation of the image, not the time between measurements. The data were collected in an immediate sequence.
 
Last edited:
Yes I could hear differences for Tubes and Jfets at Elevated line level outputs.
If you repeat an assertion enough times, it becomes true. Right?
 
@solderdude : Yeah, it is not easy to prepare a valid test with 2 identical channels. Working on it, by accident. With coupling caps, 2 channels with AD797 and 2 ADC channels of E1DA Cosmos I am able to get this matching:

View attachment 448167

2 channels are driven by identical signal sweep. One output is a reference and the other one is plotted against it. They are measured and recorded simultaneously. BTW, if you plot only one channel without a reference, all you see is a FR error and ripple of DAC/ADC filters. The small error seen here above is due to non-equivalence of coupling capacitors values and it comes exclusively from Cosmos ADC, not from the test rig.

Now if I change one of the AD797 for TL071, all else being equal, I get this:

View attachment 448168

The biggest deviation is above 20kHz, but some tiny one is at lowest frequencies as well.

Interestingly, if you try Deltawave on these resulting files, it increases the audible error, compared to direct subtraction of these simultaneously recorded files.

P.S.: the difference in time stamp is a result of date of creation of the image, not the time between measurements. The data were collected in an immediate sequence.
Yep, nulling will show a difference despite the error being merely 0.01dB so far below any audible thresholds but in a null it would show as approx -60dB. That would appear to be bordering on audibility thresholds for certain types of errors but not this particular one.
Listening tests would give more clarity (no audible difference but measurable) when tested blind.
 
Last edited:
  • Like
Reactions: MAB
Yep, nulling will show a difference despite the error is merely 0.01dB so far below any audible thresholds but in a null it would show as approx -60dB.
This exactly:

Code:
DeltaWave v2.0.13, 2025-05-01T15:14:58.5387055+02:00
Reference:  SD_BC_797.wav[L] 10034247 samples 96000Hz 32bits, mono, MD5=00
Comparison: SD_BC_071.wav[L] 10034247 samples 96000Hz 32bits, mono, MD5=00
Settings:
    Gain:True, Remove DC:True
    Non-linear Gain EQ:False    Non-linear Phase EQ: False
    EQ FFT Size:65536, EQ Frequency Cut: 0Hz - 0Hz, EQ Threshold: -160dB
    Correct Non-linearity: False
    Correct Drift:True, Precision:30, Subsample Align:True
    Non-Linear drift Correction:True
    Upsample:False, Window:Hann
    Spectrum Window:Kaiser10, Spectrum Size:262144
    Spectrogram Window:Hann, Spectrogram Size:4096, Spectrogram Steps:2048
    Filter Type:FIR, window:Hann, taps:8192, minimum phase=False
    Dither:False bits=0
    Trim Silence:True
    Enable Simple Waveform Measurement: False

Discarding Reference:  Start=0s, End=0s
Discarding Comparison: Start=0s, End=0s

Initial peak values Reference: -2.834dB   Comparison: -2.831dB
Initial RMS values Reference: -23.331dB   Comparison: -23.329dB

Null Depth=63.598dB
Trimming 0 samples at start and 0 samples at the end that are below -90.31dB level

X-Correlation offset: 0 samples
Trimming 0 samples at start and 0 samples at the end that are below -90.31dB level

Drift computation quality, #1: Excellent (0µs)


Trimmed 34570 samples ( 360.104167ms) front, 27326 samples ( 284.645833ms end)


Final peak values Reference: -2.834dB   Comparison: -2.833dB
Final RMS values Reference: -23.291dB   Comparison: -23.291dB

Gain= 0.0025dB (1.0003x) DC=-0.00003 Phase offset=-0.000209ms (-0.02 samples)
Difference (rms) = -74.33dB [-90.01dBA]
Correlated Null Depth=78.28dB [82.97dBA]
Clock drift: 0 ppm


Files are NOT a bit-perfect match (match=10.83%) at 16 bits
Files are NOT a bit-perfect match (match=0%) at 32 bits
Files match @ 49.9754% when reduced to 12.95 bits


---- Phase difference (full bandwidth): 0.16956184363114°
    0-10kHz: 0.16°
    0-20kHz: 0.15°
    0-24kHz: 0.16°
Timing error (rms jitter): 341.5ns
PK Metric (step=400ms, overlap=50%):
RMS=-91.1dBr
Median=-91.9
Max=-85.3

99%: -86.03
75%: -89.57
50%: -91.88
25%: -94.64
1%: -110.52

gn=0.999711624773979, dc=-3.28622887668772E-05, dr=0, of=-0.0200377509677133

DONE!

Signature: 750f3c085d695399e32a596e2e684ee7

RMS of the difference of spectra: -151.415520133218dB
DF Metric (step=400ms, overlap=0%):
Median=-54.6dB
Max=-42.3dB Min=-62.7dB

1% > -61.36dB
10% > -58.13dB
25% > -56.89dB
50% > -54.65dB
75% > -52.28dB
90% > -51.03dB
99% > -43.98dB

Linearity 26.3bits @ 0.5dB error
 
If you repeat an assertion enough times, it becomes true. Right?

Enough of your absence of evidence fallacy.

if you say something is blatantly false the proof is on you.
 
Enough of your absence of evidence fallacy.
Burden of proof. You make the claims, and if they're offered without evidence, repeatedly, and justified with transparently specious reasoning like that, then they are presumably false claims.

Provision of evidence for unlikely claims is YOUR responsibility, and you have ducked it again and again. It's not up to me to "prove" that you're a bullshit artist.
 
Now to open the ultimate can of worms.... can we do a tube rolling experiment?
There's lots of them out there. Start with Morgan Jones's analysis of ECC82, 6SN7, and several other types of small signal triodes.
 
My version;
:cool:


JSmith
 
My version;
:cool:


JSmith
Haha...it's possible for this song to go through AI and change "rawhide" to "op-amps" but there could be some IP implications.
 
Back
Top Bottom