Harmonic distortion measurements

[troplin]

Measuring is fun, so I've measured the harmonic distortion of my new S.E.X. 3.0 and was a bit surprised about the results.
I've always thought that tube amps have mostly even order (musical sounding) harmonics, while SS amps tend to have odd (harsh sounding) harmonics.

In my measurements however, the odd harmonics seem to dominate up to 1.5 kHz, where the 2nd order harmonics takes over.
Are those measurements realistic or could that be a flaw/error in the build?

Also, while measuring I've noticed that I can hear the sweep sound from the amp itself. Is that normal?

For the measurements I've used:
- Room EQ Wizard on Mac, Log sweep with -12dbFS
- Behringer UCA202 Audio Interface for Line In (~27kOhm, max 2.0dBV) and Line Out (max 2.0dBV), Input and Output levels both set to maximum.
- S.E.X 3.0, RCA Input, Headphone output, Volume just in the middle (seems to be almost exactly 0dB).

Tobias

[Paul Joppa]

The plot makes no sense to me. I'm not familiar with the software or the non-Bottlehead hardware, so I can't guess what's going on. But the plot shows the fundamental being expressed in percent? Normally, distortion is a percentage of fundamental level, so the fundamental would always be 100% of the fundamental.

I had a 30-year career in acoustical engineering, and I can assure you that 9 times out of 10, the first experiment has an unanticipated flaw which gives garbage results. Getting the experiment right is almost always more difficult than doing the experiment once you do get it right.

Incidentally, this is one of the dangers of experimental science. If the result makes no sense, you figure out why and try again. If it does make sense (usually meaning it matches expectations) then you tend to publish. It may still be wrong, but the experimenter has stopped looking for flaws.

"Nobody knew reality would be so hard."  :^)

[troplin]

The plot makes no sense to me. I'm not familiar with the software or the non-Bottlehead hardware, so I can't guess what's going on. But the plot shows the fundamental being expressed in percent? Normally, distortion is a percentage of fundamental level, so the fundamental would always be 100% of the fundamental.

While the fundamental is in the legend it not actually shown in the graph because of the zoom level I chose, sorry about that.
The non-Bottlehead hardware is just an external sound card DAC/ADC. I've measured that too, it has much lower distortion. So that shouldn't influence the measurement noticeably.

[mcandmar]

Run a loopback cable from the output to input of your equipment and run the test again.  That will tell you how meaningless your data is.

I have played a lot with RMAA to measure frequency response and distortion figures and getting a repeatable reading every time you setup the equipment is the hardest part.  Even now having meticulously written down all the levels and settings i still only trust it for simple A B comparisons, i don't take the results as absolute fact.

[troplin]

Run a loopback cable from the output to input of your equipment and run the test again.  That will tell you how meaningless your data is.

That's what I did, actually:
iMac --(USB)--> UCA202 --(RCA/RCA)--> SEX --(TRS/RCA)--> UCA202 --(USB)--> iMac
Or did you mean something different?

I'm still convinced that the setup and measurements are meaningful. They are also repeatable.

[troplin]

For comparison, this is what it looks when I take the S.E.X out of the loop, i.e. just:
iMac --(USB)--> UCA202 --(RCA/RCA)--> UCA202 --(USB)--> iMac

(since you wanted an A/B comparison)

[troplin]

I had a 30-year career in acoustical engineering, and I can assure you that 9 times out of 10, the first experiment has an unanticipated flaw which gives garbage results. Getting the experiment right is almost always more difficult than doing the experiment once you do get it right.

Actually, I'm already past the first garbage measurement.
At first, the UCA202 audio interface was in monitor mode (direct input->output), which of course caused nasty oscillations in the amp.
But I've noticed and corrected that already before posting here. 

[Paul Joppa]

Thanks for the clarification; now I see I was looking at the THD and thinking it was fundamental.

What are the input and output voltages? (I mean the actual input voltage at the grid of the driver triode)? I ask because the dominance of second harmonic at high frequencies and third harmonic at low frequencies suggests some kind of overload or clipping.

[troplin]

Paul, I'm not sure if I understand you correctly:
So, for the input voltage I'd measure pin 3 to ground and for the output pin 7 to 5 with the multimeter set on AC.
All this while providing an input signal similar to the one I used in the measurement. A log sweep is probably not the best input, what frequency would be best?

Is my understanding correct?

EDIT:
For the reference, all the measurements above are for the right channel.
Just to be sure I've taken the same measurements now also for the left channel and they look exactly the same.

[Paul Joppa]

Sorry for the short reply, it's been a really busy week for me.

When I said input, I meant the grid that is attached to the level control output - that would be pin 10 of the tube. Pin 3 is the output stage, not the driver stage.

When I said output, I meant the speaker terminals.

If your software has an oscilloscope function, you can just look at the output waveform, with a sine-wave input - use a low frequency such as 100Hz, since that's where the peculiar results are strongest. At small input voltages, you should see a sine wave. As you near the maximum input,the waveform will become a little asymmetrical, showing even order distortion. With greater input voltage, it will get flattened at both positive and negative peaks - this is full-fledged clipping, with lots of odd harmonics.

[Paul Birkeland]

Another way of asking for the same information is to add a DVM set for AC voltage across the speaker terminals where you are measuring (there should also be a load resistor there equal to the impedance setting you're using and rated for at least 10W).  Set your software for FFT with %THD available as a value, and crank the signal up till you get 10% THD and record the voltage at the DVM (as suggested, using 100Hz will ensure compatibility with your meter).

Otherwise, you could push a SEX amp to make 5 Watts and get a lot of THD, but you'd be clipping the amp pretty hard.

[troplin]

Sorry for the short reply, it's been a really busy week for me.

There's nothing apologize. Enjoy your weekend!

When I said input, I meant the grid that is attached to the level control output - that would be pin 10 of the tube. Pin 3 is the output stage, not the driver stage.

When I said output, I meant the speaker terminals.

Thanks for the clarification, my vocabulary is still a bit limited but it's getting better.
I'll do some additional measurements then.

If your software has an oscilloscope function, you can just look at the output waveform, with a sine-wave input - use a low frequency such as 100Hz, since that's where the peculiar results are strongest. At small input voltages, you should see a sine wave. As you near the maximum input,the waveform will become a little asymmetrical, showing even order distortion. With greater input voltage, it will get flattened at both positive and negative peaks - this is full-fledged clipping, with lots of odd harmonics.

It doesn't have live oscilloscope function, but I can look at the recorded signal from a log-sweep measurement.

Another way of asking for the same information is to add a DVM set for AC voltage across the speaker terminals where you are measuring (there should also be a load resistor there equal to the impedance setting you're using and rated for at least 10W).  Set your software for FFT with %THD available as a value, and crank the signal up till you get 10% THD and record the voltage at the DVM (as suggested, using 100Hz will ensure compatibility with your meter).

Otherwise, you could push a SEX amp to make 5 Watts and get a lot of THD, but you'd be clipping the amp pretty hard.

I think I get this, but in my measurements above there's barely any load. The line-in has an input impedance of 27kΩ and a maximum input of 2dBV (~1.26Vrms). That means the means the maximum power taken from the amp without clipping in the ADC would be about 0.06 mW if my computations are correct.
So in that setup there should be no clipping, right?

Just for my understanding, is there a difference between clipping because of high input voltage and clipping because of high power consumption or is it the same mechanism?

In any case it seems that I should try out some other software with live analysis.

[Paul Joppa]

I get 0.2 watts - RMS volts squared, divided by 8 ohms.

As PB says, to get realistic measurements it's usual to load the 8-ohm output with an 8 ohm resistor.

Clipping is a matter of driving the am beyond its capabilities; for SETs that is usually excess signal voltages somewhere in the amp.

[troplin]

I get 0.2 watts - RMS volts squared, divided by 8 ohms.

As PB says, to get realistic measurements it's usual to load the 8-ohm output with an 8 ohm resistor.

Sorry to bother you with stupid questions, but where did you get that 8 ohm from?
I've plugged the amp output directly into the line input of the sound card which has 27KOhm input impedance, no speaker or other load involved.
Is that 8 Ohm something internal to the amplifier?

Of course this is not a realistic listening setup but OTOH with virtually no load attached I had (naïvely?) expected only minimal distortion. I thought that would be a good baseline measurement.

Anyway, I'm not giving up yet so I'm going to order that 8 Ohm resistor and repeat my measurements.

[Doc B.]

You connect an 8 ohm resistor - I would suggest one rated for at least 5W in this case - across the speaker binding posts, to emulate the load of a speaker.