measuring THD on audio circuits

[tsaG]

Hi!

Im building some audio power amplifier circuits of which I want to measure the THD. Right now, these are just Tube Amps so the THD will be around 1% at 1kHz. But it would be nice to be able to measure down to around 0.1% or even lower.

The first Distortion measurement device I found was the Keithley 2015 which costs around 350€ on ebay and looks pretty nice to me.
 It even can be used as bench top DMM (which I could use as well). The 205-P can even be used as a Spectrum analyser.

It also offers a serial port to output the data to the pc which is mandatory (The output to the PC is mandatory, not the serial port )

I do have a Siglent SDG2042X which has a THD of around 0.072%, if the testing device doesnt have a frequency generator.

Are there any other (maybe cheaper or better) devices you can recommend?

Thanks!

[Messtechniker]

good enough for most power amps:
Soundcard (see below) source distn. goes even lower than your siglent +
something like Arta or Audio tester.

[FrankBuss]

Better and industry standard, not cheaper, are the Audio Precision instruments. Sometimes you can get a used one old 2700 series for 2k bucks on eBay.

[Performa01]

The first Distortion measurement device I found was the Keithley 2015 which costs around 350€ on ebay and looks pretty nice to me.
It even can be used as bench top DMM (which I could use as well). The 205-P can even be used as a Spectrum analyser.

Go for it!

The Keithley 2015THD is a very capable and precise bench DMM, and it never hurts to have one.

The ability to measure THD in the range 20Hz to 20kHz is a bonus, and it is very easy to use.
The instrument also has a built-in low distortion signal generator.
Having an RS232 connector is a real plus too, as most instruemnts of that vintage only have GPIB - if at all.

Be aware that you will have to find (or write) a suitable application for communication with the instrument and data processing, as to the best of my knowledge there is nothing supplied from Keithley for it (at least not for free).
 

[TimFox]

Two classic but very usable -hp- instruments:

333/334:  THD measurement, auto-tuning, needs external low-distortion oscillator for source
339:  THD measurement, down to < 0.01%, with internal low-distortion oscillator and auto-scaling of THD.

Can be found on eBay.

[macboy]

I have a HP 8903B (not U8903B) which is an 80's era instrument that does distortion and other related audio measurements (SNR, AC and DC level, SINAD). Unlike the 33x series instruments, this one is fully automatic, has digital displays, and is PC-controllable via GPIB. The protocol is not SCPI based but is very simple. It has a loopback distortion (generator output to analyzer input) of 0.0020%, so it is reasonably accurate for measuring from about 0.0100% and up. It can also act as a very sensitive true-RMS AC millivoltmeter with 750 kHz bandwidth. The 8903A is older and slightly less desirable due to not as good differential input stage; the 8903C is identical to 8903B but only has the analyzer portion, not the low distortion source (which limits usefulness). The 8903B should be available from $400 $700? and up. The market for these appears much stronger today than just a few years ago, as is the case for many instruments.

Panasonic makes some really nice instruments that can do an even better job than the HP: the VP-7722A through VP-7726A and "B" versions too. Expect to pay $1k to $2k depending on model.

Many people have great results with a good quality soundcard and the right software. Rightmark audio analyzer is one such software, and an example of a good soundcard is the M-Audio 192. That card has a very low THD+N, 192 kHz sampling, and differential inputs and outputs which can help reject ground loop related noise. The software approach has the substantial benefit of giving you a complete picture of the harmonics and noise, not just the sum total level of them. One downside is the limited <90 kHz bandwidth. You just can't see all the higher order harmonics of higher frequency input signals with that bandwidth.

[German_EE]

It's maybe difficult to measure using the following method but I know how to view it:

1) Connect a 1 KHz sinewave source to your amplifier via a splitter

2) Connect channel A of a scope to the 1 KHz source and channel B to the amplifier output

3) Adjust the channel B vertical gain so that both waveforms are the same height

4) Using your scope 'math' function subtract channel A from channel B, what's displayed is your distortion.

[oldway]

Transistor amplifiers generally have a high rate of feedback and thus have a very low distortion.

As the ear perceives only high distortion (over 2%), it is irrelevant to accurately measure very low distortion.

0.01 or 0.03%, whatever, it makes no difference to the hearing.

I use an old HP331A that gives me satisfaction.

As you should not often measure the distortion, it is not necessary to have an automatic and expansive test equipment.

[Performa01]

It's maybe difficult to measure using the following method but I know how to view it:

For this approach to work well, the two scope channels would need to have the exact same gain and frequency/phase response, which is very unlikely, given the high tolerances in scope frontends.

With analog scopes, one could indeed adjust the gain for a near perfect null (but not the phase). On a DSO there is only a fine adjust in discrete steps (if at all). Resolution is e.g. 2%, which means that a max. error of 1% will remain and cannot be trimmed out – at least not on the scope. Consequently, distortions <1% get swamped due to the imbalance.

This approach might still be beneficial for watching crossover and transient intermodulation distortions, but not so much for the 2nd harmonics that are prevalent in tube designs.

[deadlylover]

If you're very comfortable waiting for a good deal and you're an expert online shopper  , some of the old analysers from Japan may be an option for you. More search terms cant hurt anyway.

I think the two stand out models are the Panasonic VP-7722A and the ShibaSoku 725 series (725B or 725C). The Panasonic is the more traditional analyser with a oscillator built in, whereas the ShibaSoku is a distortion analyser only, you'll need a separate oscillator (which ShibaSoku also makes).

The Panny VP-7722A performs much better than its sisters, even the later models like the VP-7723A, it is equal in residual distortion and noise to the flagship VP-7725A, which is basically a VP-7722A with an LCD display. It typically performs at 0.00023% (-112dB) THD+N, 1kHz, 30kHz BW. For THD only it will go down to 0.00004%(-128dB) or so at 1kHz, and the display limit for each harmonic is -140dB.

The VP-7722A is a fantastic analyser, it is as good as you'll realistically need for any tube amp, it can even display each harmonic up to the fifth at the touch of a button, no need to boot up a PC and fiddle with a sound card just to check the 2nd or 3rd when fiddling with biasing, too easy, it can even do IMD.

Next up is the ShibaSoku, I have to admit I've never even heard of them until recently and didn't give their analysers much thought because because their residual THD+N spec was something like -105dB, respectable for a 25 year old instrument but not remarkable in any way these days, even a sound card can reach that no problem.

But here's the secret to the ShibaSoku 725B/C, they absolutely DESTROY their rated spec. They typically do 0.00014% (-117dB) THD+N, 1kHz, 30kHz BW. That's AP 2700 series territory which is still 3.5k+ used if you're lucky!
Perform a few op-amp swaps and minor mods, and the 725 will go all the way down to -120dB THD+N without breaking a sweat, which is more than good enough for any DIY audio adventure. For reference the flagship Audio Precision APx555 goes down to around -123dB, but it starts at 28 grand...

I think I paid around $400 for the Panny 7722A, and $300 for the ShibaSoku 725B, they are big beasts though so shipping for each of them almost cost as much for the instrument itself. Unbeatable value IMO for the hobbyist. Unless you can get the older HP analysers for real cheap, these units are the only ones that hold up today.

I'll still echo the suggestions on getting a sound card to start with, just throw together a buffer so you don't accidentally blow it up with the full fury of a power amp . You will also be able to use it to have a looksie at the monitor outs of an audio analyser if you ever get one, it's only good for looking at low level harmonics though and not at anything very close to the fundamental due to the notch.

[ebastler]

I'll still echo the suggestions on getting a sound card to start with [...].  You will also be able to use it to have a looksie at the monitor outs of an audio analyser if you ever get one, it's only good for looking at low level harmonics though and not at anything very close to the fundamental due to the notch.

Sorry, you lost me there -- could you elaborate a bit, please? I have played with soundcards for audio analysis but have no real expertise in that field.

- Can't the soundcard (plus suitable software) be used as an audio analyser? Why would I use it merely to look at the monitor output of, I assume, a dedicated hardware analyser?
- What's the thing about the harmonics and the notch filter? Where is the notch filter?

Thanks for explaining!
Juergen

[deadlylover]

Sorry, you lost me there -- could you elaborate a bit, please? I have played with soundcards for audio analysis but have no real expertise in that field.

- Can't the soundcard (plus suitable software) be used as an audio analyser? Why would I use it merely to look at the monitor output of, I assume, a dedicated hardware analyser?
- What's the thing about the harmonics and the notch filter? Where is the notch filter?

Yep you can certainly use a sound card as an audio analyser, I wasn't trying to suggest otherwise. I was just trying to say that it could be used to get additional capabilities from a dedicated hardware audio analyser using the monitor output.

The notch filter is from the dedicated audio analyser. Super simplified basics of an old audio analyser;

1. Input level control and attenuation
2. Notch the fundamental and buffer the output (monitor output is after this stage)
3. Measure everything else, by using a wide bandwidth AC voltmeter or by digitising and then performing analysis

The reason for the notch is because even today, ADC's don't have enough dynamic range to dig real deep when looking at very low level harmonics (-140dB or lower from the fundamental). You can still get excellent performance though, I think all the entry level audio analysers don't use a notch based system, they can get up to -112dB or so THD+N, and maybe -120dB ish for individual harmonics.

Once we start getting into state of the art territory you have to notch the fundamental, it isn't that simple because we need an auto tuned notch circuit... which is quite expensive because of all the damn $$$ relays you need. 

The Audio Precision flagships use a notch in high performance mode, with lots of software processing goodness so you don't really notice the notch in normal use, it displays the fundamental as normal. Since it samples the signal before and after the notch filter and combines them, you kinda get the best of both worlds, excellent detail close to the fundamental for looking at phase noise and a devastatingly low noise floor due to the notch.

Here is the monitor output of my fatboy Shibasoku AM51B, the fundamental is at 1kHz, ignore the shown levels because its quite fiddly to get the correct levels to show on ARTA. The 400Hz high pass filter is engaged hence the dropoff at low frequencies.

You can see the little dip from the notch filter, hence we won't be able to use the monitor output to look at anything close to the fundamental. The second harmonic is at around -139dB from the fundamental, and it is very clearly visible from the floor. Just to put that in perspective, the noise floors of excellent sound cards and entry level audio analysers is at that point!

[SteveK]

As mention by someone else,  Audio Precision is the gold standard.  I use a Audio Precision Portable One Plus at work can easily read down to 0.001% distortion.  I don't believe it is made anymore but we have purchase several on eBay.

Before Audio Precision, the Sound Technology 1700 series was the standard.  I have a ST1700B at home. Not quite as good as Audio Precision, but readings below 0.01% are easily obtainable.  Back in the 70's, every Pacific Stereo store had one.  The story goes that when Pacific Stereo went out of business, it flooded the market with used Sound Technology equipment and caused Sound Technology to go belly up too.  I see a number of listing on eBay for cheap.

If you buy a Sound Technology, you may have to repair it.  The switches are typically noisy and there are LDRs that are used for auto nulling and stabilizing the generator output.  Those LDRs drifts with age and requires a realignment occasionally.

Way back, I used the HP 333/334 series.  They are not near as good as Audio Precision or Sound Technology.  If I remember correctly, they don't do auto nulling or auto leveling so you have to tweak with the controls to null out the fundamental.

Good luck.

Steve

[Loboscope]

I want to take the liberty to repeat here a post that I wrote recently:

I could recommend the analysis-software "HpW-Works" (http://www.hpw-works.com/).
It is used normally with a soundcard and for reliable measuring it is obvious that a really good one is necessary. Therefore I would recommend the interfaces from RME (http://www.rme-audio.de/), especially the "PCI-Card HDSP Hammerfall 9632" or the new "Babyface Pro". These interfaces have not only dgital IN/OUT (SPDIF / ADAT) but extremely high performance analog IN-/OUT-connections.

The software can connect to the interfaces up to 32 bit and max. samplerate. The software limit of samplerate will be 1 GHz, but there is no soundcard available capable of more than round about 200 KHz. So this will be the real limitation in frequency range, normally ca. 100 KHz.
Because the software is able to manage multiclient ASIO, it can handle separate interfaces for Input/Output. This means, you can measure a DUT, for example a Digital Mixer-Desk by sending the stimulus signal via PC-Interface to the analog-In of the Desk and use the digital interface-connection of the desk, for example USB to receive the signal from the Desk. PC-interface and Desk are connected via its own ASIO-driver parallel to the software and so there will be no additional AD/DA-converter in the signal chain.

HpW-Works has its own signal source witch can produce absolutely clean test signals who are a must for doing any reliable testing. The test signal can be adjusted in 0,5 / 0,01 dB steps or by entering a free value.

In my opinion HpW-Works is fantastic analyser tool what produces solid and professional results - at a reasonable price.
Looking at the main user list (http://www.hpw-works.com/index.php/2015-08-09-08-41-49/users) you will find quite a lot of the "who-is-who" in the professional audio world.

Greetings, Jürgen

Supplement:
The digital I/O skills might not be from interest for building analog amplifiers, but the measuring results are indeed of a "official" quality. It would be no problem to print out the THD / THD+N / IM results and give them together with the amplifiers to the customers.
For me it would never have been an option to buy an old-school heavyweight HP or Advantest. Above all, these machines are still rather expensive although they are aged.
New instruments for example from Audio Precision are state of the art but really, really upscale.

[cvanc]

Lots of good advice here, just want to highlight one other possibility for a 'vintage' analyzer - Amber.

Amber was a Canadian company that made very respected audio analyzers back in the day.  You don't hear much about them but their model 3500 and 3501 were quite nice, and meant for the travel/portable market, so had a fairly small form factor (handy for the home hobbyist).

[Tim F]

The Keithley 2015THD has pretty much laughable distortion specs in the audio band compared to even a cheap computer sound card. There are numerous free software suites which will do signal generation and THD measurement, for example ARTA http://www.artalabs.hr/download.htm, HOLMImpulse http://www.holmacoustics.com/holmimpulse.php

Here is a modest $150 USB soundcard/interface (E-MU 0204 USB), doing a lazy 0.00026%THD and 0.003%THD+N in loopback.


Even a $40 soundcard will get you well below 0.01%THD. You'll need an attenuator to provide a 10-100x voltage reduction so you can connect the output of the amplifier to the line level of the sound card. The maximum voltage input on a soundcard is usually only a few volts RMS. If you can, i'd recommend building the divider into your dummy load so you get a nice low impedance output which will help keep injected noise down - i.e. if you have an 8ohm dummy load resistor and put a 0.1ohm resistor in series to ground, tapping the signal off the 0.1ohm resistor you get an 81x voltage reduction and ~0.1ohm output impedance. Just make sure that the 0.1ohm resistor never goes open circuit or your soundcard will have a bad day!
Also be mindful that inputs to a computer sound card are not usually isolated from mains earth in desktops and potentially also laptops if using a mains charger. Floating a laptop on batteries comes with all the usual warnings that floating an oscilloscope does.

[vk6zgo]

1.0% THD+noise is fairly lousy for a tube audio amplifier.
The 55kW High level modulated MF Broadcast transmitters I worked on did better than that.

The Hall monitors at the same sites were specified at 0.25% THD+noise & normally passed that spec easily.

Back in the day,we used the same "N&D" sets that we used for 600 Ohm sources.
We just switched them to "Hi Z",& connected them across a dummy load.

The easiest ones to use were the AWA 'auto-null" units,which still pop up on eBay from time to time.

[mikerj]

The Analog Discovery 2 does THD measurements as part of its broad repertoire.