An Experimental AC Voltage Calibrator

[enut11]

Experimental AC Voltage Calibrator
I debated whether this project fits in the ‘Projects, Designs and Technical Stuff’ thread or the ‘Metrology’ one. Since I love precision electronics, here goes.

After spending a brief session with a Fluke 5200A AC Voltage Calibrator, I realised what a handy instrument it is. The 5200A offers a wide range of accurate high resolution voltage at frequency settings. Biggest drawbacks for me: 1) Bulk: The 5200 is a huge and heavy box that needs lots of bench space, 2) Cost: around $2000 USD or more for a good one.

So, whenever I come across an idea for a DIY AC Voltage Calibrator, it gets my attention. Recently, a friend suggested an interim solution. Not a ‘proper’ calibrator but a stable, variable amplitude/frequency source of AC volts. Such an  instrument, together with an accurate DMM, could be used to ‘transfer’ an AC voltage/frequency combo to another instrument.

The concept
Take a stable low distortion audio frequency generator and feed the signal into a HiFi audio amplifier. This alone will give you up to 20vRMS in the range of 20Hz-20KHz and possibly well beyond. For higher output voltages, connecting a 100v/8ohm audio line transformer, back-to-front, to the amplifier will give you up to 200vAC.

A simple concept, but does it work? How stable would the output be? These are the questions I needed to answer for myself. More to come…
enut11

[Jester]

I made something  like this quite a while ago before I had an ac calibrator. I used HP sig generator as source driving audio amp and conventional transformer running at about half its voltage rating. It’s a lot better than using MAINS voltage and capturing and correlating two sets of measurements, at least the measurements are relatively stable.

The bulky size, cost and cost of the conventional ac calibrators was also a deterrent for me. I prefer to have all the instruments that I use regularly within reach at my bench. Bottom row 2nd from right.

Clarke Hess (smaller less expensive ac calibrator) show up on eBay once in a while, I picked one up in non working order for next to nothing and fixed it, pretty decent little calibrator if super accuracy is not required. I tried to get a schematic from Clarke Hess, they were uncooperative. Troubleshooting a bad one without the schematic is a bit tricky. Mine just had a leaky diode.

[mzzj]

I did take a part in DMM intercomparison (8½ digit Fluke 8508A) with shoestring budget equipment some years ago.
Laptop soundcard, audio amplifier and bunch of transformers to get up to 700Vac 
Results looked good with approx 0,03% uncertainty up to amplifier max voltage.
transformer coupling increased the uncertainty to 0.06% at 700 Vac (3458A as reference is not that great here either)

Laptop sound card was much better choice than either of tested (digital) function generators.

[mawyatt]

Interesting idea. How stable is the audio amp gain and how flat over frequency?

A very simple and useful low voltage AC calibrator we used some time ago is based on a CD4060 divider chip and a 4.096MHz crystal. The VDD voltage for the CD4060 is created with a stable 5 volt reference, since the CD4060 is CMOS, the selected frequency divided output will swing from ground to ~VDD without any load (typical DMM isn't much of a load). The output voltage is an accurate 5.000 Vpp 50% duty cycle squarewave with an AC rms and DC average value of VDD/2 or 2.500 Volts and stable.

Of course the squarewave has harmonics and the frequency range must be taken into consideration, but works quite well on the various DMMs we have, KS34465A, HP34401A, AG34401A, SDM3065X, and DMM6500 just to verify low voltage AC performance.

Edit: Just plugged in the reference and let it run for a few minutes, it's AC reading on DMM6500 is 2.499264V RMS with 30uv SD. Also reads 2.500049V DC with 10uv SD.

Edit: Found the calibrator thread.
https://www.eevblog.com/forum/projects/diy-dvm-simple-dvm-cal-device/msg3157650/#msg3157650

Best,

[bdunham7]

The concept
Take a stable low distortion audio frequency generator and feed the signal into a HiFi audio amplifier. This alone will give you up to 20vRMS in the range of 20Hz-20KHz and possibly well beyond. For higher output voltages, connecting a 100v/8ohm audio line transformer, back-to-front, to the amplifier will give you up to 200vAC.

A simple concept, but does it work? How stable would the output be? These are the questions I needed to answer for myself. More to come…

Having done both (but not the transformer part) I can tell you that it works fine as long as you have a known good audio amplifier, an ordinary AWG and a good reference meter.  The accuracy limits of this setup will likely be determined by the reference meter.  A transformer will likely work as long as you stay well within its saturation limits.  I use one or two (bridged) Marantz monoblock amplifiers and I can do better than 80V_rms @ 50kHz.

The 5200A has significant limitations and having had/fixed/used them I don't miss having one.  The frequency controls have fine resolution, but poor accuracy--you need to use an external AWG to control them if you want precise frequency.  They only go to 120V_rms on their own, and that only up to 80kHz or so.  The frequency range goes to 1.2MHz, but that is only at a low voltage.  You can get those low-cost add-on amplifiers for your AWG that will do nearly as well if you need to go beyond what your audio amp can do.

[mawyatt]

You can get those low-cost add-on amplifiers for your AWG that will do nearly as well if you need to go beyond what your audio amp can do.

We designed and couple as AWG Buffer amps as supplements to the Juntek DPA-2698 and DPA-1698, one we built so far is based upon the HV OPA462 Op-Amp discussed here:

https://www.eevblog.com/forum/testgear/hv-buffer-amp-for-awg/msg3623953/#msg3623953

As a quick test, just plugged in the 2.5V RMS AC calibrator mentioned above and connected it to the OPA462 based AWG ~10X Buffer Amp.

The DMM6500 reads ~25.00667V, KS34465A 25.0085V, HP34401A 25.0558V, and AG34401A 25.0419V.
Then connected up the AWG at:
50Hz with 50.0550, 50.0554, 50.0245, and 49.9998 respectively
100Hz with 50.0460, 50.0437, 50.0215 and 49.9920
1KHz   with 50.0475, 50.0480, 50.0345, and 50.0030
10KHz with  50.0775, 50.0740, 50.0620, and 50.0330
50KHz with 49.9135, 49.9048, 49.8860, 49.8350

Haven't tried driving a transformer tho for higher voltage, that would be interesting to see how well these relative readings hold up across the various DMMs at higher voltages.

Edit: here's the AWG 10X Amp.

Best,

[enut11]

Well, some interesting ideas and suggestions already. That is why this is a great Forum for electronics enthusiasts. For now I will continue with my story and address any issues later.

I had a surplus custom instrument on hand with all the parts inside for a mono audio amplifier. It was based on the LM1875 20W audio amp chip and, with the included power transformer (36v CT), was able to give around 15vRMS output when fed with a 1KHz sinewave at the input. This was a good start. Using this setup I could generate AC voltages from 10Hz to 50KHz.

The next step was to find a suitable audio transformer for higher output voltages. Well, this was a learning experience in itself. High quality audio output transformers can cost many thousands of dollars! No doubt well worth it to the golden-ears brigade. For my purpose, I wanted to keep the cost down to well below a 5200A. So, for the initial proof-of-concept, I settled on a Jaycar Electronics (Australia)  MM1900 for less than $11.

I might add that audio output transformers are better suited for this role because of the extended frequency response due to the special steel laminations.

[enut11]

The signal generator
The output of open loop systems is very dependent on the quality of individual interconnected components and this project was no exception.

I started out by analysing the amplitude stability of my signal generators.
1) Leader LAG126S audio frequency
2) Feeltech model FY2300
3) Wavetek Model 80
4) Chinese kit - 'Low Distortion Audio Range Oscillator'

The best turned out to be the last one. The design is based on a simple state variable design with FET output control. Using a regulated +/-15v power supply this oscillator produces a clean low noise signal. Even so, the plot below shows there is a 60min warmup period before it stabilises to the point that it can be relied on for making good measurements.

Stated distortion is 0.006%.
I built a 1KHz + 10KHz kit and may even look at a variable frequency version.

[mawyatt]

Is it necessary to have a low distortion sine-wave for checking AC meter performance? As mentioned we used a square-wave since producing and accurate squarewave is straightforward using CMOS logic and flip-flops.

Seems that as long as the distortion products don't extend beyond the measuring instrument meter bandwidth, that they get accurately accounted for in the RMS computations, or in the analog RMS converter chip, which ever the meter utilizes.

Might expect some differences with meters that use different RMS methods, like computational or analog, but not so much between "like" technique instruments, again as long as the BW isn't pushed.

Curious as to what others think about this?

Best,

[enut11]

The amplifier
The next item to check was the amp itself. All AB class amplifiers dissipate several watts, even doing nothing, in order to produce a low distortion output. So, even though I was only going to amplify volts at very low currents, the standing amp bias current necessitated a large heat-sink and even a cooling fan.

National specify the following for LM1875 amp:
The LM1875 is a monolithic power amplifier offering up to 30 Watts Output Power very low distortion and high quality performance for AVO Typically 90 dB consumer audio applications. Low Distortion: 0.015%, 1 kHz, 20 W The LM1875 delivers 20 watts into a 4Ω or 8Ω load. Wide Power Bandwidth: 70 kHz on ±25V supplies. Using an 8Ω load and ±30V
Protection for AC and DC Short Circuits to supplies, over 30 watts of power may be delivered.

My noise measurements with a shorted input show about 30uV PP.

[enut11]

Is it necessary to have a low distortion sine-wave for checking AC meter performance? As mentioned we used a square-wave since producing and accurate squarewave is straightforward using CMOS logic and flip-flops.

Seems that as long as the distortion products don't extend beyond the measuring instrument meter bandwidth, that they get accurately accounted for in the RMS computations, or in the analog RMS converter chip, which ever the meter utilizes.

Might expect some differences with meters that use different RMS methods, like computational or analog, but not so much between "like" technique instruments, again as long as the BW isn't pushed.

Curious as to what others think about this?

Best,

My view is that square waves are only good for low frequency tests (<1KHz). Clean sine waves are needed for the higher frequencies .
enut11

[bdunham7]

Is it necessary to have a low distortion sine-wave for checking AC meter performance? As mentioned we used a square-wave since producing and accurate squarewave is straightforward using CMOS logic and flip-flops.

Seems that as long as the distortion products don't extend beyond the measuring instrument meter bandwidth, that they get accurately accounted for in the RMS computations, or in the analog RMS converter chip, which ever the meter utilizes.

For a basic check perhaps not, but for calibration to within the meters specs, you probably need a low-THD sine as specified.

One thing to think about is that in addition to any error that might occur simply because of dissimilar RMS conversion of non-sinusoids, the meters have a frequency-dependent accuracy spec.  Higher frequencies result in wider tolerances.  Take your square wave calibrator and look at how much energy there is in those upper harmonics and which tolerance bin they fall into.  Say you have a meter with a specified accuracy of +/- 0.02% @ 1kHz, but +/- 3% @ 50kHz.  What percentage of the signal is in that 49th or 51st harmonic?  How about the percentage of the signal that is in all the harmonics above 50kHz?  And how does the 3% error multiplied by that  percentage compare with the 0.02% error at the fundamental? 

Now if you are talking about two sine stimuli, one with 1% THD and the other with 0.1% THD, both primarily in the 2nd and 3rd harmonics, then perhaps the difference will be less.  This would not be difficult to test.  Perhaps the next time I'm working on a meter or a calibrator, I'll try it.

[alm]

Is it necessary to have a low distortion sine-wave for checking AC meter performance? As mentioned we used a square-wave since producing and accurate squarewave is straightforward using CMOS logic and flip-flops.

Seems that as long as the distortion products don't extend beyond the measuring instrument meter bandwidth, that they get accurately accounted for in the RMS computations, or in the analog RMS converter chip, which ever the meter utilizes.

Might expect some differences with meters that use different RMS methods, like computational or analog, but not so much between "like" technique instruments, again as long as the BW isn't pushed.

I don't think DMMs are generally specified for square(ish) wave signals. I would definitely want to use a bandwidth-limited signal limited well below the bandwidth of all meters involved. Then I imagine would kind of work, but I'd expect the uncertainty of your comparison to be higher than with sine waves due to the different frequencies involved and the finite attenuation of any filter.

Performance verification generally calls for frequencies across the bandwidth. For example for the HPAK 34401A, they use frequencies ranging from 20 Hz to 300 kHz (there's a reason why the Fluke 5200A has such a wide bandwidth). How would you verify the high end. A 300 kHz square wave? A 100 kHz square wave limited to 300 kHz (won't be very square)?

[enut11]

Line Output Transformer
The next step was to introduce the line output transformer to generate voltages  beyond the capability of the basic audio amplifier - up to 200vAC RMS.
This was my first encounter with audio transformers so was unsure what to expect. The transformer used for these tests is shown below. It is a Jaycar Electronics MM1900 100v/8ohm line speaker transformer with 'primary' tappings for 0.5W to 5W. The latter proved useful in setting the transformer ‘gain’.

The transformer 8ohm 'secondary' was connected to the ‘speaker’ output of the amp and the high impedance ‘primary’ connected to a DMM. For the MM1900, the gain was found to be x35 on the 2W tap so 200vAC out needed less than 6vAC from the audio amp which in this case had a 20:1 gain. This means that the signal generator only needed to supply less than 300mV to the amp.

I decided to test at a reduced output of 100vAC and the results appear below. Again, for about 60min, the output drops steadily, mainly due to the signal generator. After that, the 100vAC 1KHz signal proved to be surprisingly stable with relatively low noise. The second graph below shows the estimated noise in the flat part of the curve of approx 1400uV PP which equates to 14 PPM for a 100v signal. Not too bad for such a simple setup.

Frequency response of the $11 audio transformer was somewhat less than the amp, 50Hz-15KHz, but still useful.

[enut11]

As added insurance, I decided to regulate the LM1875 amp power supplies using LM317/337 chips. The standing amp bias was around 50mA. The unregulated supplies were +/-25v so I set the regulators to +/-20v. This had the effect of reducing the amp output to about 13v before clipping. Still plenty enough to drive the audio transformer.

What's next? More output of course!
My friend suggested a better audio amp and a coupled pair of output transformers. Australia’s Silicon Chip Mag (Dec 2021) have come up with a brilliant ‘Hummingbird’ design based on a very compact PCB (64mmx88mm). Specs include <0.008% distortion and 60 watts RMS into 8 ohms (22Vrms). Noise is down at the -118dB level. The amp has a gain of x18 and a frequency response of 1Hz to 150KHz!
The existing low distortion audio generator (<0.003% THD) should be ok but would probably benefit from being in a bigger, properly shielded metal box.

The output transformer: Altronics (Australia) have a line of quality audio output transformers with a  better frequency response. I have ordered two units (catalog M1120) with a view to coupling them in such a way as to share the output voltage – up to 300vAC should be possible without straining the transformer primaries. The M1120 are 20 watt units with a stated response of 30Hz-20KHz. They also have 4, 8 and 16ohm output taps which will come in handy when selecting the optimum ‘gain’.
enut11

[bdunham7]

As far as audio transformers go, units made for single-ended tube outputs may work well because they are air-gapped and designed not to saturate.  That and they may be just fine with 300V+ outputs.  The PA ones are a little lower caliber.

[David Hess]

Interesting idea. How stable is the audio amp gain and how flat over frequency?

It might be stable, but a typical amplifier will not be flat even to 0.1% over a significant frequency range.  Better accuracy will require leveling.

A very simple and useful low voltage AC calibrator we used some time ago is based on a CD4060 divider chip and a 4.096MHz crystal. The VDD voltage for the CD4060 is created with a stable 5 volt reference, since the CD4060 is CMOS, the selected frequency divided output will swing from ground to ~VDD without any load (typical DMM isn't much of a load). The output voltage is an accurate 5.000 Vpp 50% duty cycle squarewave with an AC rms and DC average value of VDD/2 or 2.500 Volts and stable.

That is how I have done it in the past to at least 0.02%.  I am not sure how much better because I do not have anything I trust that is better than that.

Of course the squarewave has harmonics and the frequency range must be taken into consideration, but works quite well on the various DMMs we have, KS34465A, HP34401A, AG34401A, SDM3065X, and DMM6500 just to verify low voltage AC performance.

I was very pleased when my best meters return consistent results with average AC, RMS AC, and RMS AC+DC measurement of my DC calibrated precision square wave.

[mawyatt]

Is it necessary to have a low distortion sine-wave for checking AC meter performance? As mentioned we used a square-wave since producing and accurate squarewave is straightforward using CMOS logic and flip-flops.

Seems that as long as the distortion products don't extend beyond the measuring instrument meter bandwidth, that they get accurately accounted for in the RMS computations, or in the analog RMS converter chip, which ever the meter utilizes.

For a basic check perhaps not, but for calibration to within the meters specs, you probably need a low-THD sine as specified.

One thing to think about is that in addition to any error that might occur simply because of dissimilar RMS conversion of non-sinusoids, the meters have a frequency-dependent accuracy spec.  Higher frequencies result in wider tolerances.  Take your square wave calibrator and look at how much energy there is in those upper harmonics and which tolerance bin they fall into.  Say you have a meter with a specified accuracy of +/- 0.02% @ 1kHz, but +/- 3% @ 50kHz.  What percentage of the signal is in that 49th or 51st harmonic?  How about the percentage of the signal that is in all the harmonics above 50kHz?  And how does the 3% error multiplied by that  percentage compare with the 0.02% error at the fundamental? 

Now if you are talking about two sine stimuli, one with 1% THD and the other with 0.1% THD, both primarily in the 2nd and 3rd harmonics, then perhaps the difference will be less.  This would not be difficult to test.  Perhaps the next time I'm working on a meter or a calibrator, I'll try it.

Agree, one could calculate the effects harmonics have based on the meter characteristics vs frequency and as enut11 mentioned because the square-wave has such high odd order harmonics (squarewave vary as 1/n, triangle is 1/n^2 I recall) this limits the squarewave approach to a fairly low frequency.

Just did a quick set of measurements using one of those cheap XAR chip generators that produce horrible looking sine (and triangle) waves at 1KHz. The 2 year old KS34465A and 2 week old DMM6500 agree very well and use the same computational RMS method I believe. The two 33401As agree with each other, and with the DMM6500 within ~50ppm, but use an analog RMS conversion chip like the Siglent SDM3065X which differed by the highest at ~800ppm. Then changed to a ratty looking triangle wave, with almost same results.

Did same tests with better sine and triangle waveform AWG with good results, but actually the results from the KS34465A, DMM6500 and two 34401As were even more aligned with the "ratty" waveforms from the cheap XAR generator

The Siglent SDM3065X didn't agree well with any of the other meters, regardless of the waveforms used, but does use an analog RMS conversion chip, however so does the two 34401As, so go figure

Best,

[trobbins]

There are a few USB soundcards that have adequate audio bandwidth out to 80-90kHz before amplitude droop becomes noticeable.  The advantage of that type of oscillator source is that software can be used to null out harmonics from the 'output'.  As such, a very low distortion output can be sourced from the soundcard, but also extend further out to include any following buffer amp, as well as any step-up audio transformer.  Nulling harmonics in that manner can achieve low distortion levels that are intrinsic in just the ADC of the soundcard, and of course if the frequency is not too high in that bandwidth (eg. circa 20kHz fundamental for first few harmonics to be nulled), and the same applies for distortion introduced by a step-up transformer at the low-frequency end.  Certainly a cheap and easy way to start testing without the need for a low distortion oscillator.

Using a step-up audio transformer may benefit from some judicious loading of the high voltage output winding to maximise high frequency response flatness.  The output transformer likely needs to be 'hi-fi' in design to both push out amplitude variations from the first resonance, and to constrain the 'Q' of that resonance, given that resonance could well be below 50kHz depending on the transformer (a good transformer can push that resonance out to circa 100kHz).  The soundcard technique with suitable software (like REW) can even test the impedance of the transformer to confirm where the first resonance occurs (and hence the likely limit to frequency response for high voltage calibration efforts).

[mawyatt]

Is it necessary to have a low distortion sine-wave for checking AC meter performance? As mentioned we used a square-wave since producing and accurate squarewave is straightforward using CMOS logic and flip-flops.

Seems that as long as the distortion products don't extend beyond the measuring instrument meter bandwidth, that they get accurately accounted for in the RMS computations, or in the analog RMS converter chip, which ever the meter utilizes.

Might expect some differences with meters that use different RMS methods, like computational or analog, but not so much between "like" technique instruments, again as long as the BW isn't pushed.

I don't think DMMs are generally specified for square(ish) wave signals. I would definitely want to use a bandwidth-limited signal limited well below the bandwidth of all meters involved. Then I imagine would kind of work, but I'd expect the uncertainty of your comparison to be higher than with sine waves due to the different frequencies involved and the finite attenuation of any filter.

Performance verification generally calls for frequencies across the bandwidth. For example for the HPAK 34401A, they use frequencies ranging from 20 Hz to 300 kHz (there's a reason why the Fluke 5200A has such a wide bandwidth). How would you verify the high end. A 300 kHz square wave? A 100 kHz square wave limited to 300 kHz (won't be very square)?

This technique is simply a method to allow an accurately known voltage reference to the translated to an accurate RMS waveform within limits. Since a perfect squarewave has an RMS value (by definition) of the peak value. Same goes for a DC value, RMS is (by definition) equal to the DC level, and the average of unipolar squarewave is 1/2 the the peak value. A low frequency CMOS output from a Flip-Flop is almost a perfect squarewave with 50% duty cycle and swings unloaded from zero the VDD.

So in the AC cal reference mentioned we had an output that was accurate in both RMS and average DC level, and directly related to the 5.000V DC reference utilized for VDD for just a few $, not a bad cost/performance ratio IMO. Would really like a cal standard like the 5200A, but well out of our budget range, so we did the best we could.

Many years ago we developed a precision DC voltage divider that won EDNs Best Idea of the Year (patent 5030848) based upon a similar concept except this patented DC voltage divider had almost no ripple and achieve sub-ppm levels of voltage division accuracy (1/2) without any precision components! The general idea was to use a CMOS FF Q and Qbar output to toggle two resistors connected together, say R1 and R2, and the junction between them is Vout. When Q is high then the output is VDD*(R2/(R1 + R2)), when Q is low the output is VDD*(R1/(R1 + R2)), the average is VDD/2 independent of either R1 or R2. A mismatch in R1 and R2 just causes a small ripple which is filtered with a single shunt C and the output. Of course the CMOS FF output NMOS and PMOS fets are not equal, but can be made insignificant with just a simple buffer. We also did this with a discrete NMOS and PMOS inverter in our AC calibrator.

Anyway, CMOS digital logic can be utilized in many unusual useful ways, these are just a few.

Best,

[enut11]

There are a few USB soundcards that have adequate audio bandwidth out to 80-90kHz before amplitude droop becomes noticeable.  The advantage of that type of oscillator source is that software can be used to null out harmonics from the 'output'.  As such, a very low distortion output can be sourced from the soundcard, but also extend further out to include any following buffer amp, as well as any step-up audio transformer.  Nulling harmonics in that manner can achieve low distortion levels that are intrinsic in just the ADC of the soundcard, and of course if the frequency is not too high in that bandwidth (eg. circa 20kHz fundamental for first few harmonics to be nulled), and the same applies for distortion introduced by a step-up transformer at the low-frequency end.  Certainly a cheap and easy way to start testing without the need for a low distortion oscillator.

Using a step-up audio transformer may benefit from some judicious loading of the high voltage output winding to maximise high frequency response flatness.  The output transformer likely needs to be 'hi-fi' in design to both push out amplitude variations from the first resonance, and to constrain the 'Q' of that resonance, given that resonance could well be below 50kHz depending on the transformer (a good transformer can push that resonance out to circa 100kHz).  The soundcard technique with suitable software (like REW) can even test the impedance of the transformer to confirm where the first resonance occurs (and hence the likely limit to frequency response for high voltage calibration efforts).

Hi @trobbins.
I measured the audio transformer output impedance at around 1,700 ohms at 1KHz using a resistive load. Can you suggest what sort of transformer loading I could experiment with to maximise high frequency response?
Also, at the moment, the amp output is loaded only by what it sees at the 8ohm audio transformer winding.

[trobbins]

Perhaps start with an 8k resistive load that reflects 8 ohm back to the amplifier (eg. for the M1120, the 1.25W tap presents 8kohm for an 8 ohm speaker load, with a 31.6:1 turns ratio).  You could sweep through the frequency range to see what voltage response you get, but at low voltage (as an 8k load would consume some power for higher voltages).  You may still get a reasonably flat response for higher resistance loads.

[ledtester]

TheHWCave did a couple of videos on building a power amplifier for a function generator based on a OPA541 module available from aliexpress...

Building an OPA541-based power amplifier for function generators
https://youtu.be/925X0RDUy9w

#103: Some changes to my OPA541-based power amplifier for function generators
https://youtu.be/ERJ9Om-eBqo

It was used to test the accuracy of a PZEM-004T AC power meter:

How accurate is the Peacefair PZEM-004T AC Comms Module?
https://youtu.be/j0_y8dPfpKc

The setup in that case was: FY6600 -- Power Amp -- 5V/240V transformer

[alm]

This technique is simply a method to allow an accurately known voltage reference to the translated to an accurate RMS waveform within limits. Since a perfect squarewave has an RMS value (by definition) of the peak value. Same goes for a DC value, RMS is (by definition) equal to the DC level, and the average of unipolar squarewave is 1/2 the the peak value. A low frequency CMOS output from a Flip-Flop is almost a perfect squarewave with 50% duty cycle and swings unloaded from zero the VDD.

The Tektronix PG506 calibration generator (for scope calibration) used the same technique of providing both the DC and chopped DC level, although for oscilloscopes, so the accuracy was only 0.25%. The problem for DMMs is the probably imperfect response of the DMM to this almost perfect squarewave.

[mzzj]

There are a few USB soundcards that have adequate audio bandwidth out to 80-90kHz before amplitude droop becomes noticeable.  The advantage of that type of oscillator source is that software can be used to null out harmonics from the 'output'.  As such, a very low distortion output can be sourced from the soundcard, but also extend further out to include any following buffer amp, as well as any step-up audio transformer.  Nulling harmonics in that manner can achieve low distortion levels that are intrinsic in just the ADC of the soundcard, and of course if the frequency is not too high in that bandwidth (eg. circa 20kHz fundamental for first few harmonics to be nulled), and the same applies for distortion introduced by a step-up transformer at the low-frequency end.  Certainly a cheap and easy way to start testing without the need for a low distortion oscillator.

Using a step-up audio transformer may benefit from some judicious loading of the high voltage output winding to maximise high frequency response flatness.  The output transformer likely needs to be 'hi-fi' in design to both push out amplitude variations from the first resonance, and to constrain the 'Q' of that resonance, given that resonance could well be below 50kHz depending on the transformer (a good transformer can push that resonance out to circa 100kHz).  The soundcard technique with suitable software (like REW) can even test the impedance of the transformer to confirm where the first resonance occurs (and hence the likely limit to frequency response for high voltage calibration efforts).

I have been thinking of similar harmonic cancellation. Is there a ready plug-and-play software already available for that?

Having said that I think it is huge overkill for DMM calibration. Even best calibrators like Fluke 5730A specify distortion at something like 0.03%
Any decent audio gear is easily in -90dB level or 0.003%.

---------
ESL (electrostatic loudspeaker) folks might have some clever ideas on how to get some voltage swing without transformer coupling the output.
Full-bridge output with cascoded 1kV mosfets would give you easily  over 1kVrms output.