Has anyone tried measuring amplifier distortion?

[Saurav]

I thought I'd give it a shot with my DIY tube amp. I'm wondering if anyone has tried anything similar, and if there are any tips / tricks / gotchas to doing this.

I decided to use REW, because it's free, and the current version has a 'spectrum' tab that can measure THD+N. Set up the generator for a 1kHz sine wave, disconnected the amp from the speaker and hooked it up to a 7.5ohm resistor, ran alligator clips from the resistor leads to the input RCA cable.

The first thing I discovered is that the signal generator in REW can be a little squirrely. Since the speaker wasn't connected, I didn't have an easy way of telling if the signal was on, or what frequency / level it was playing it. I have a cheap DMM and I was using that, set to VAC, to check for signal. As it turns out, if I type in a frequency into the generator window, it apparently doesn't register until I tab out of that field. And sometimes the start/stop button would stop doing anything. A few times I had to restart REW to get the signal to stop or start. Also, at these low signal levels, the DMM leads pick up noise from the air, I think, so sometimes the signal reading would jump higher if the DMM lead slipped off the resistor. It probably didn't help that I had a bunch of unshielded alligator clip leads all over the place.

Then, my first measurement had an OK THD (0.4% or 0.04%, don't remember, almost all H2), but the THD+N was up at around 5%. The spectrum showed 60Hz and its harmonics. It took me a little while to figure this one out. The amp has directly heated triodes on the output, which have AC on the heaters. This adds 60Hz hum to the output, the DMM says the hum level is 5mV. The last time I measured it with a scope (years ago), I think I got similar numbers. Since I've always used this amp to drive a midrange and tweeter, I never bothered trying to reduce that hum. Anyway, with the laptop volume controls set to about half-way, and the REW generator at the default -20dB, I was sending about 10mV to the amp. If I trust the DMM, the amp's gain is around 3 (it's a 4W SET), and the output was around 30-40mV. So the level was clearly too low.

Then I decided to crank up the REW signal generator level. It goes up to -3dB, which sent about 300mV to the amp, and I had about 900mV output. The REW distortion chart showed about 3% THD before it said 'clip' and stopped updating. So, was it my amp distorting, or the soundcard input? I don't know.

So then I decided to try a voltage divider to reduce the input level to the soundcard. I added a 0.27 ohm resistor in series with the 7.5, and took the signal off that to send to the soundcard. This caused the output signal level to drop by 90% as soon as I connected the ground lead to the soundcard. Sometimes in my test gear also started 'singing', because I could hear the 1kHz tone near the laptop. This really puzzled me for a while, but I think I finally figured it out. I'd put the 0.27 ohm resistor on the 'hot' side of the output transformer, and connecting that to the soundcard was grounding it... and either shorting something out, or I don't know. Once I switched the 0.27 ohm resistor to the 'bottom', and made sure the soundcard ground lead connected to the output transformer ground lead, the singing stopped, and the voltage drop went away. I hope I didn't sink a lot of current into the soundcard and kill it.

Anyway, by this point I think I was finally set up to take some measurements, and my laptop rebooted. It was late, so I decided to listen to some music before going to bed. I don't think I hurt the amp, because it played music fine. I'll pick this up again tonight and see if I can get some numbers.

Can REW do the distortion/frequency charts that I see people posting from SoundEasy? Speaker Workshop says it can do that, but I've never been able to make it work. Assuming it can't, what signal should I be using for this measurement? 1kHz seems like a marketing standard for reporting THD. I'll probably also measure it with a 500Hz signal, since that's the XO frequency to the woofer so that the nominal 'lowest' frequency the amp is playing. I measured FR for the amp a couple of days back using SW, and it's flatter than I expected (I built the amp years ago and haven't messed with it in a long time). Channel balance is OK (one side is 0.4dB louder than the other). I should probably also try to measure cross-talk.

[ThomasW]

Chasw98 has a long thread on AVS where he measures a ton of amps, most are high powered ones for subs.

He's using hardwired test gear ........

http://www.avsforum.com/avs-vb/showthread.php?p=10689427#post10689427

[Saurav]

Thanks. That's a great resource. Way more elaborate than what I'm trying to do, but it gives me some ideas about what to look for.

[Amphiprion]

Try using ARTA. It should do exactly what you want pretty easily. I measured the distortion of my soundcard at 0.04% THD+N in loopback mode with it and it was very simple to use. It is free, you just can't save files in demo mode. Otherwise fully operational.

[Saurav]

Ah, I didn't know there was a fully functional free version. Thanks for that tip

Another question - is there a way to measure the noise floor of an amp? I tried a THD+N measurement with no signal sent to the amp (should have shorted the input, forgot about that). But then the distortion calculator decided the fundamental was the 60Hz hum, and so of course I ended up with 50% THD, and a very low overall signal level. Also, REW thinks the input is coming from a mic and the absolute SPL level has been calibrated, so when I'm measuring voltages like this, the SPL readings are pretty weird. Not sure about the best way to get around that.

[Amphiprion]

You are in Oregon, that is not too far from AudioPrecision, is it?

You will have to make a notch for the 60hz hum I think. But then you will get 120, 180, etc harmonics showing up at lower levels. If the source is the amp itself then there is no fix other than to notch it, if it is coupling into the system via unshielded cables or something then that needs to be attended too.

You could try a THD+N with a -3dBFS 60Hz tone. The program should notch it out itself for the THD+N measurement.

Just an FYI, ARTA calculates a THD and a THD+N figure separately. Maybe that will help you get around it.

[davey_m]

Also try Rightmark Audio Analyser:

http://audio.rightmark.org/index_new.shtml

There's a free edition too.

[Saurav]

You are in Oregon, that is not too far from AudioPrecision, is it?

LOL. I had to look them up. They're in the same city I work in

I have the free version of RMAA on that laptop, I totally forgot about it. I'll have to see what it can do.

FWIW, REW gives me a THD and a THD+N too, that's how I figured out that my 1kHz tone was too low in level, and the 60Hz + harmonics noise was messing things up. Hopefully I'll be able to get it working with my resistor divider tonight, and have the 1kHz tone a lot louder than the 60Hz noise.

[Dennis H]

Another question - is there a way to measure the noise floor of an amp?

If you calibrate a spectrum analyzer such as the one in ARTA with your voltmeter, you can read volts (or maybe it's dBV) right off the graph. That seems more useful than a single number as you can see which frequencies are creating the noise.

[mazurek]

I also use ARTA. I go without a resistive divider, but measure output levels before plugging in. In the manual, it shows schematics for an impedence testing jig that will protect against high amplifier output, I think a variant of that jig would work well for measuring amplifiers.

I'm done with preamps for now, as I never find any that measure better than the digitally attennuated DAC output, and sometimes find frequency response/distortion surprises on preamps, unbal->balancing devices, and mic preamps.

[Saurav]

I go without a resistive divider, but measure output levels before plugging in.

I started with that. I'm pretty sure my little DMM isn't true RMS, but it shouldn't be too far off with a single tone. It showed 900mV on the amp's output, and I'd expected the soundcard line-in should be good up to the 2V nominal. However, it threw the 'clip' warning and stopped. I guess, thinking in dB, 1V is only -3dB away from clipping, so I did let it get too close.

The reason I'm trying to go with higher overall signal levels is to get away from the 60Hz hum / noise floor.

[chasw98]

Saurav:
I think you are starting to see that using a computer based system to measure very low levels of distortion and very high ratios of signal to noise doesn't work that well. I have been looking for a way to do it with a computer for a long time. Even my older HP and Tektronix analyzers are good to .0007% THD and 120 dB S/N. I have not been able to or have heard of a way to get a computer to do that precise of a measurement. The computer will display the results and log results just fine, but as a measuring tool it falls short for low distortion/high signal to noise ratio measurments. And don't even ask me about PC based oscilloscopes!

[Saurav]

I see what you're saying. I'm mostly playing around here - I'm thinking of making some changes to my amp, so I wanted some kind of baseline to determine what difference the changes might make. It's a single ended tube amp, I doubt I need a measurement system that needs to go very low

[Saurav]

OK, ran some more tests with ARTA. After calibrating my soundcard and raising levels to just below clipping, the soundcard measured at 0.003% THD and 0.011% THD+N. This was with a stereo interconnect going from the USB soundcard's input to output, and switching the laptop from AC to battery didn't make any difference. So that's a fairly reliable number, I think?

Then I tried the amp, with the voltage divider. REW and ARTA both gave almost identical results, so that was encouraging too.

At 1kHz and 1V output (measured across the whole load, not just the voltage divider), the amp showed 0.44% THD / 1.47% THD+N (ARTA), and 0.38% THD / 1.62% THD+N (REW). Dropping the signal to 0.5V output brought the THD down to 0.22%. Raising the signal to 1.5V output showed 0.6% THD and 1.16% THD+N.

I ran some more tests at 500Hz, 2kHz, 5kHz, 10kHz... pretty much the same THD values.

I also noticed that the 60Hz and harmonics didn't change level as I changed the signal level. I think this means it's mostly coming in through the alligator clip leads, and that explains why THD+N goes up as I reduce signal level.

REW also prints the individual components of the THD value, so that was interesting to see as well. For the 1V output numbers:

THD 0.38
H2 0.37
H3 0.06
H4 0.001
H5 0.004
H6 0.001
H7 0.0005
H8 0.0004
H9 0.0006

Like I said, it's a home-brew SET, so it's dominated by H2, as expected. H5/H7/H9 are higher than H6/H8/(H10, I guess), which was slightly surprising.

The other signal level numbers followed a similar distribution.

Can someone help me interpret these numbers? First off, do they look trust-worthy? 0.4% maps to about -48dB. 0.22% is -53dB, 0.6% is -44dB. I took a look through some of the distortion charts on Zaph's and Brandon's sites. Most seem to have the fundamental at -10dB on the graphs, and the driver distortion is down in the -60dB range. Brandon's site has some spectrum graphs, and the THD readout for some random drivers I looked at is in the 0.3 - 0.5% range.

So... if I'm reading that right, this amp's distortion is on the order of some of those drivers? Or am I an order of magnitude off somewhere? I really had no idea what to expect... everyone knows SET amps have high distortion, but I'd never done any research into how high, or how it compares to drivers. I'm sure this is several orders of magnitude higher than any SS amp, or even a good PP tube amp.

Also, there's the issue of output power. 1.6V into roughly 8 ohms is roughly 0.3W, right. With a 95dB midrange, that's about 90dB/1m? Listening distance is ~2m (subtract 6dB), 2 speakers (add 6dB), so that's about 90dB at the listening position? If I have that math right, then that's louder than I usually listen. But I should probably check this with the SPL meter against the voltmeter. Is the 'marketing standard' to report output power as the level where THD (or really, THD+N) crosses 1%? I guess for tube amps that's all over the map, I'm checking Stereophile measurements as I type this.

The changes I'm thinking of trying are supposed to help reduce distortion in the driver stage, so now I have a way of seeing if it makes any difference at all. I'm sure I'll like the change regardless, especially if it takes a lot of work to put it in

[benchtester]

Saurav,

Looks like you have a pretty good tool there. :T

Your results correspond with what I would expect from a SET amp. You might want to look at some of Nelson Pass' write-ups on the Zen amps and others.

http://www.passdiy.com/articles.htm

He uses an AP to measure his changes and I believe the distortion scale is in percent.

[JonMarsh]

Saurav:
I think you are starting to see that using a computer based system to measure very low levels of distortion and very high ratios of signal to noise doesn't work that well. I have been looking for a way to do it with a computer for a long time. Even my older HP and Tektronix analyzers are good to .0007% THD and 120 dB S/N. I have not been able to or have heard of a way to get a computer to do that precise of a measurement. The computer will display the results and log results just fine, but as a measuring tool it falls short for low distortion/high signal to noise ratio measurments. And don't even ask me about PC based oscilloscopes!

The only way to get close to making that work in the real world is by using a sound card such as I have in my Praxis/LspCAD Emulator PC; an M-Audio 24/192. They don't make that card anymore, but the ProFire 610 outboard Firewire box has comparable resolution.

OTOH, you're usually better off for the money spent picking up a good used unit, like an HP8903 (that's what I use), unless you have a lot of other part time tasks for that firewire interface, as it's somewhat overkill for speaker measurements.

Another reason NOT to use PC based tools for measurements is the quality of the PC analog ground; with the stuff going on with the typical PC power supply and MB, line related noise on the analog ground from the PC will be a problem - even with an HP or AP analyzer using very quiet power supplies (electrically), you sometimes have to jigger your ground connection setup to get the cleanest measurements, depending on the grounding configuration of the DUT.

[chasw98]

OTOH, you're usually better off for the money spent picking up a good used unit, like an HP8903 (that's what I use), unless you have a lot of other part time tasks for that firewire interface, as it's somewhat overkill for speaker measurements.

That's what I have also. :T

[benchtester]

Saurav has been pretty clear about his goals, set up and calibration. His loop-back test shows that his test equipment contributes less than 1% of his device-under-test.

Personally, I am a little discouraged by the results. I would like to do the same thing; but my amplifiers are high feedback solid state and would need better loop back numbers.

Since he is using a USB sound card, I assume he can locate it away from the computer power supplies and such. I suggest upgrading to a linear power supply on the USB sound card. It is funny, I am using a USB sound card for speaker measurements and making/buying a better power supply never occurred to me until I starting thinking about Saurav's setup.

BTW: I have used linear power supplies (brick-style wall worts, <$20) from Jameco, and they significantly improved 60 Hz. rejection on sensors at work. I haven't checked if they are "audio quality".

[Saurav]

Hmm... I have the little M-Audio USB soundcard, and it's powered from the laptop, there's no wallwart connector on it AFAIK. That's why I disconnected the laptop power supply to see if any 60Hz noise was coming in from there, and it didn't really make any difference to the loopback spectrum. The soundcard itself was pretty far away from the laptop's wallwart. Did you mean a linear power supply for the laptop?

I didn't move any equipment out of the rack, so the amp had about 5' of speaker cable, the test resistor was at the end of that, and there were alligator clips connecting that to an RCA cable which went to the laptop. Definitely not a low-noise environment.

[ergo]

I also mainly use ARTA, but if you need also a THD vs output power measurement the Audiotester has a automated mode for that.

I also like that Audiotester allows selecting which harmonics you want to be included in analysis.. so one can even do separate graph for 2nd, 3rd, 4th etc

http://www.audiotester.de/

Ergo

[Dennis H]

Saurav:
I think you are starting to see that using a computer based system to measure very low levels of distortion and very high ratios of signal to noise doesn't work that well. I have been looking for a way to do it with a computer for a long time. Even my older HP and Tektronix analyzers are good to .0007% THD and 120 dB S/N. I have not been able to or have heard of a way to get a computer to do that precise of a measurement. The computer will display the results and log results just fine, but as a measuring tool it falls short for low distortion/high signal to noise ratio measurments. And don't even ask me about PC based oscilloscopes!

Chuck, check out the EMU USB 2.0 soundcards with mic preamp. Bill Waslo tried one with Praxis and his analog Audiopod was the limiter for noise and distortion. I think he measured something like .0003% distortion for the EMU. He said it's the first card he tested where it actually improved things to go up to 24 bits. Another nice thing is they get their power from the USB port even at 192K so, with a laptop running on the battery, you're isolated from line and power supply problems. At $150 full retail for the Tracker, it looks like a bargain.

http://www.libinst.com/EMU%200202%20USB.htm

http://www.emu.com/products/welcome.asp?category=610

[benchtester]

Hmm... I have the little M-Audio USB soundcard, and it's powered from the laptop, there's no wallwart connector on it AFAIK. That's why I disconnected the laptop power supply to see if any 60Hz noise was coming in from there, and it didn't really make any difference to the loopback spectrum. The soundcard itself was pretty far away from the laptop's wallwart. Did you mean a linear power supply for the laptop?

I have an EMU 0404 that is bigger and has a mic preamp. It uses a wall wart. I doubt a linear supply on the computer would help much, certainly not as good as running on the laptop's battery power.

But I think you are getting good results as it is. Did you check out Nelson Pass' reports?

[Saurav]

I looked through the folder and skimmed over the first 'distortion, harmonics, etc.' article. Haven't looked at any of the test results yet. It wasn't immediately clear from the article names which ones contained actual designs / measurements, and which ones were more theoretical discussions.

I spent some time on the Stereophile website last night looking at their measurements of tube amps. I don't think my numbers are *too* shabby (at least, just on the distortion front)

Ergo, I didn't know about that audiotester website / application. I'll give that a shot as well.

[benchtester]

I looked through the folder and skimmed over the first 'distortion, harmonics, etc.' article. Haven't looked at any of the test results yet. It wasn't immediately clear from the article names which ones contained actual designs / measurements, and which ones were more theoretical discussions.

I think you would find his Zen series interesting. It started with him putting out a simple SE Mosfet design. It got a following and a life of it's own. Over nine versions the distortion was reduced considerably.

[Saurav]

Thanks. That gives me a pointer to where to start reading.

[mazurek]

Saurav:

I have a "TrueRMS" Fluke DMM. I'm pretty sure it still doesn't have a flat frequency response. I just set the tone generator to 60Hz to check voltage level (and assume my amp has flat enough response that it won't go any higher at other freqs), which is what it was designed for I guess.

You're numbers are already great for higher harmonics. I bet you'll get better numbers at a couple of watts output, if you look at published THD+N versus voltage/power curves for op-amps, and other amps on the internet, they usually decline in percentage up to a certain sweet spot, and then start to go up exponentially.

Chasw:

I would disagree to some extent. I have an EMU1616M, and its line in to line out gives ~100dB signal to noise and THD. Its instrument/mic in will only give ~80dB SNR/THD. I've measured a Lynx Aurora ADDA, and it can do ~110-120dB SNR/THD. That's better than the quoted specs for a NI data acquisition card. A dedicated analyzer might be more usable and have better noise, but I'd want to input the data into a computer anyways. The other thing is that you can enjoy a high end sound card for playback, but I don't think an analyzer would be very enjoyable to listen to, everything sounds like "bleoop","chirp","shhhhhhh" : )

[Saurav]

So... something's out of whack. Here's what I tried:

* Checked the FR of my amp using SW, and it's reasonably flat between 1kHz and 60Hz. This doesn't require the DMM, and the soundcard's FR has been accounted for in the calibration.

* Checked the amp output on the voltmeter at full signal output from my laptop. 1kHz was about 2.5V out, 60Hz was a few mV higher. So the DMM would appear to be fairly flat as well...

* 2.5V @ 8 ohms = 800mW. At this signal level, the amp's distortion went up to about 1.1%

* With the same signal level, I hooked the speaker back up and measured it with the RS SPL meter at roughly 3'. It measured at 103dB.

So... no way that's 800mW. The mid is an Audax PR170M0, which says it's 100dB/W/m, but realistically, it's more like 95, depending on which measurements you trust (AFAIK, both graphs below are from the manufacturer):





Comments from other people who've used this driver also say it's around 95-97. Also, I'm measuring through the crossover, so there's some insertion loss there as well.

So... I vaguely remember setting up the amp's gain to roughly go to full output at the nominal 2V input. My laptop's ouput is maxed out, and my amp seems to be putting out something like 3-4W based on the SPL reading, but < 1W based on the DMM reading. Which do I trust? I even tried generating a sine wave in Audacity at full level and playing that back, and I get the same signal levels.

I think it's closer to 4W than 0.8W. That's max output (or even a little higher) for a 2A3 tube, so that would explain the increased distortion if the amp's gone into the soft clipping range. H3 also went up a little (H2 1%, H3 0.25%) - would that be a sign of clipping? And I just trust the SPL meter more at 1kHz... either the DMM is off for AC, or I'm mixing RMS / peak somewhere and getting some math wrong, or... I don't know.

I could play some full-level test tones through my DAC, or through a CD player I have sitting around, and see what the DMM measures for that output. 0dB FS (if that's the right term) should be around 2V, right?

[chasw98]

When you say that you are measuring 8 ohms, what are you using for a non inductive load? If you are not using a non inductive purely resistive load, then you may be putting out 800 mw at that frequency and then putting out 10 watts at another frequency.

[Saurav]

Sand-cast resistors from my stash, don't remember the source. Could be anything from PE/Dayton to Radio Shack to a local electronics surplus store to... Not sure how inductive they are, but I'd expect them to be mostly resistive? Also, I measured the output signal and the speaker SPL at the same frequency. It's possible the speaker load is very reactive at that frequency... but I'm not sure if that should give me higher power output or not.

[mazurek]

Maybe measure current in series with the driver. You've got everything else, its got to add up.

[chasw98]

Sand-cast resistors from my stash, don't remember the source.

Run a quick impedance sweep on your load resistors as if they were an actual speaker and see if the impedance is constant at all frequencies.

[Saurav]

I tried measuring the current. My Radio Shack DMM won't pass a 500Hz signal through it. It'll pass it when I use the 10A/unfused input, but that won't register a mA current level (and the range selection button doesn't do anything).

Longer version - I decided to try this with my stereo, not the laptop. A 500Hz tone is the highest frequency I have on my server's 'test tones' folder, so I used that. ROKU, DAC, preamp, active XO, tube amp, speaker. The active XO high pass would be down a few dB at 500Hz, but that doesn't matter, since I'm just trying to check the voltage/current/SPL at the driver.

So... voltage at passive XO input - 118mV, at driver terminals - 81mV, SPL at ~1m - 62dB. My measurements of this driver's impedance show 8.06 ohms, -10 degrees @ 500Hz.

I'm not sure exactly how this math works here. Let's assume the driver is 95dB/2.83V/1m. So 62dB output means I'm at -33dB from 2.83V, and 20 * log (81mV / 2.83V) = -31 (if I did that right), so that would roughly line up.

So I did something wrong in my previous measurement, I guess. I didn't re-measure the amp output after I switched from the resistor to the speaker. I'll do that, and measure the voltage at the driver as well.

I also realized that I don't remember the last time I replaced the batteries in the DMM, so I should take care of that as well.

Anyway, hopefully the math above is correct, which means at least in this test, my measurements are more or less in sync.

[owdi]

I measured my amp using Arta almost two years ago, the results are below. I forget what the power level was. I used 2x 16 ohm resistors in parallel for the load, and a 10k series/1k parallel resistor as a voltage divider.

Link to original thread at diyaudio.com



I also had a problem with 60hz hum, which I fixed by plugging my amp and computer into the same power strip.

Dan