Measuring 2kVpp @1MHz as precise as possible

[e61_phil]

Hi,

to check some electronic specifications I would like to measure 2kVpp with 1MHz as precise as possible (with a realistic amount of money). Some people are dreaming of 500ppm accuracy, but I think that is nearly impossible. The voltage is sinusoidal so one could measure RMS (~707V) (however, the specification is the ppVoltage).

One can't load the output (<= 3pF would be fine), but probably one can build a loadable source first and calibrate the actually used divider against a better one on that source.

Does anyone have an idea how to measure such a voltage to better than 1%?

Thx
Philipp

[schmitt trigger]

Browsing thru the web, I once found the schematic for a Textronix high voltage probe. If I remember correctly, it required trim caps to get the correct frequency response, but otherwise it wasn't terribly complicated.

The real killer requirement, IMHO, is a loading capacitance of no more than 3 pF.

[e61_phil]

We already build a divider to measure the voltage with a scope.

But how should one calibrate such a probe?

[schmitt trigger]

I am going to assume that the voltage probe's transfer function is completely linear across the whole voltage range.

But it requires that the high voltage source can be adjusted down
Therefore, using a scope and a scope probe rated for -let's say- 400 volts, adjust the output voltage to that safe value.

Put your probe on CH1 and the commercial probe on CH2, which you have already compensated.
Adjust the trimmers on the CH1 probe until both readings are identical.

[e61_phil]

No, the transfer function isn't perfect linear. And 400V isn't much more precise on a scope.

At the moment we do it quite similar as you described it. But that is far away from 1% uncertainty.

[e61_phil]

The actual divider is more or less only capacitive. This is neccessary to reduce loading problems.

But even if you build one with resistive elements. That will not be the same on DC and on 1MHz.

Stable frequency isn't a problem. Stable amplitude on arbitrary voltages is a bigger problem I think. But probably one could use a RF power amplifier with a transformer to create a loadable test source.

[Marco]

So you'll need to learn about compensated dividers first.

He doesn't need DC (RMS was good enough) so compensation is futile, just do it all AC. Personally I'd try to make a capacitive divider from PCB with a buffer on board, put it in a grounded metal project box to mostly remove effects from variable parasitic capacitances to ground, strip some shield from coax and glue it in as a high voltage feed through and fill it with oil. Maybe temperature control the oil as well.

The problem is calibration. I'm sure there are high voltage labs which will do it for a price.

How about constructing a variable frequency up to 2kv "square" wave generator with a halfbridge from say IXTA02N250HV? You can be relatively certain about the output waveform from principle and by varying the rail voltage and frequency and making some assumptions you can characterize the divider.

[David Hess]

We already build a divider to measure the voltage with a scope.

But how should one calibrate such a probe?

Oh dear, high voltage high frequency calibration is not for the faint of heart.

This does not directly apply but Jim Williams discussed high voltage probe calibration in appendix C of Linear Technology application note 65.

There are two ways I would consider:

1. A mechanical charge line type of pulse generator like the old Tektronix 109 can produce high fidelity high voltage pulses but pulse width depends on charge line length which will be a problem for response down to 1 MHz.  It also requires a mercury relay.

2. A high voltage version of the standard pulse generator can produce a very accurate and clean output.  This is sort of like the pulse generator above but in reverse where the transistor turns off to release the high voltage pull-up resistor which drives the probe input.  The problem is that it will not be fast enough because of excessive output capacitance of the high voltage transistor.

A high voltage pulse generator with push-pull class-b output driving switching diodes can do much better because of lower output capacitance and is what I would try.  The peak-to-peak output voltage is set by the DC biasing of the switching diodes.

I wonder what Tektronix used to calibrate probes like their P6013 and P6015.

[pigrew]

What about using a transformer to step down there voltage to something that can be measured at lower voltage?

Then again, measuring accurately at 1MHz isn't easy either... it perhaps it would call for a calibrated thermocouple power meter (though this reads RMS, not Vpp, and I think it's calibrated to what a 50 ohm source would deliver, not what was actually delivered)?

[T3sl4co1l]

Not hard. Use a small capacitance (~0.1pF?) and measure the current through it.  Calibrate against another measurement method at a smaller signal level.

Tim

[e61_phil]

I have a Behlke high voltage switch which can switch 24kV and 100A. Normaly I limited the current, therefore I never looked how fast it can be. But it should be very fast. So with that it is easy to build a 2kVpp squarewave. But I think that isn't so easy for calibration. One has to get rid of the harmonics, otherwise everything will depend on the frequency response again.


I also thought about a transformer, but at 1MHz one will have a lot of parasitic effects. I think that isn't easy.


The capacitor is more or less the way it works anyway. Something like the Fluke 792A (which I don't have) can measure 1MHz up to 20V only. The best equipment I have for such a job would be a HP 3458A or a Fluke 8508A. The HP could measure up to 100V RMS to 1.5%. The Fluke can only measure to 2% but up to 200V. Doesn't sound so bad. 

But how could one proof the linearity of the circuit between 100/200 V RMS and 700V RMS?
Everthing behind the capacitor can be proofen with a larger capacitor and smaller voltages. The coupling capacitor is the real uncertainty here.

[CopperCone]

how accurate are those 8.5 digit meters at 1MHz? I thought they were pretty much built for DC.

You might be interested in the 6.5 digit thermal RMS meter that HP made a while back. I don't know if it went up to 1MHz though.

Maybe you want to use a voltage divider connected to a thermal converter? 500ppm is 0.05 percent.
 
8.5 digit meter:


thermal converter:


Now, the tempco of the thermal converter is horrid, and the input impedance is 50ohm, but you have something better then 100ppm accuracy at 1MHz if the graph is to be trusted, with a 40V input max, while If I am reading correctly the 8.5 digit meter is only accurate to 1.5% at 1MHz, so 1500ppm?

maybe check spectral content first using a SA to see what kinda spread you got on the frequencies/crest then if its tonal you can trust the RMS converter as not averaging a buncha garbage.. oscilloscope wont provide enough resolution for you requirement, and probably wont detect the distortion that will effect your measurement right? But the thermal meter will just take everything into account, so you would need to filter it, but idk... and the thermal imposes its own time constant on the measurement, you would need to figure out the ramifications of this. I think you would need a 1MHz distortion meter to trust anything, the spectrum analyzer would just basically be useful to tell you that something is really wrong.. I think.. if you wanted 500ppm,  I did not think about this that much though

but then you have to play thermal offset games at 2000ppm/degree C, so you need to calculate dissipation and prob calibrate out for temperature.............

and no idea on the HV side of shit

I consider an analog scope accurate to like 4%, digital maybe 2%? Are they better then that?

https://www.keysight.com/main/editorial.jspx?cc=US&lc=eng&ckey=806222&nid=-32453.753266&id=806222

Not really, its 1.5 percent best bet, but its a scope, so I would multiply that times 1.5 since its not designed by voltnuts, plus it tells you nothing about the spread that might result from distortion along the frequency axis, its probably fairly linear if they can boast 1.5 though, but that's still pretty much killing your 500ppm, but your not getting anything more using a 8.5 digit meter vs a scope, unless your measuring the DC output of a thermal converter.. infact its less useful because you don't have any frequency data, can't see glitches and it requires alot of averaging,

8.5 digit meter offers a bit a piece of mind though in terms of how stable its gonna be compared to some lab scope that can be used for anything really

sounds like a pain in the ass measurement

[Doctorandus_P]

If the specifiction is P-P then the "sine wave" part becomes pretty much irrelevant.
I have a vague memory of an old application note where a Multi kV voltage reference was built by putting hundreds of shunt reglulated voltage references in series.

Maybe you can do something with putting peak detectors at both ends of such a multi kV reference stack. (Better: Use the bottom of the P-P value as a reference for the 2kV stack of shunt references).
Or measure only the parts around the peaks with optically isolated oscilloscopes.
Some USB scopes also come with WiFi, but I do not know if that is acceptable in your environment.

[ycui@eml.cc]

Hi,

to check some electronic specifications I would like to measure 2kVpp with 1MHz as precise as possible (with a realistic amount of money). Some people are dreaming of 500ppm accuracy, but I think that is nearly impossible. The voltage is sinusoidal so one could measure RMS (~707V) (however, the specification is the ppVoltage).

One can't load the output (<= 3pF would be fine), but probably one can build a loadable source first and calibrate the actually used divider against a better one on that source.

Does anyone have an idea how to measure such a voltage to better than 1%?

Thx
Philipp

500ppm is well within reach of existing technology.

Measuring 2KVpp at 1MHz is a common technical problem in quadruple mass spectrometer. This problem has been resolved decades ago.

Although this is a solved issue, There are many technical details goes into it. A few point:

1. A oven is likely required
2. The RF amplitude(Vpp) is determined through active rectifier
3. A high voltage capacitor divider is needed to divide the 2KVpp sine wave done to 5V range for low voltage circuits
4. The capacitor divider will need to be calibrated (this is difficult and mass spectrometer don't need it, mass spectrometer need to the detector to be stable and precise, but not necessarily accurate.)
5. In exotic case, vacuum capacitor is used for dividing.
6. Custom vacuum capacitor were also developed to minimize thermal drift. The mechanical structure of the vacuum capacitor need to be compensated with different materials.
7. The analog signal processing chain will need high speed amplifier going into even GHz bandwidth, although you only measuring 1MHz signal

If you are still interested in this project, please contact me.

[chuckb]

To put some numbers to a 100:1 divider.

A Jennings 1.5pf 15kV Vacuum RF capacitor can be purchased here (surplus?) for $25. The second cap for the divider could be a 150pf 100V NP0 capacitor or another vacuum capacitor. Reference https://www.tedss.com/2020001359

What is the voltage dependency? I would say it's in the ppm range based on the GenRad information provided below. Using a 15kV capacitor at 1kV will also help minimize capacitance change with voltage. Also, how can you get better than a vacuum to minimize undesired dielectric effects?

GENRAD Precision 1404 capacitors (Hermetically sealed in dry nitrogen)  -
"Voltage and Frequency
Additional sources of capacitance change are variations in voltage and in frequency. Changes of capacitance no greater than a few ppm can be expected in standard air capacitors with voltages in the usual measurement range below, say, 100 volts. In silvered-mica capacitors, such as the TYPE 1409, changes from 10 to 200 ppm may occur for voltage changes from 1 to 100 volts. Such changes usually result from small isolated sections of the silver film that connect to the main body of film and add capacitance as the applied voltage increases. The magnitude of the change varies widely from capacitor to capacitor, depending upon the quality of the silver film. When foil electrodes are used instead of silvered mica, the changes with voltage decrease to the order of a few ppm."

Just using a GENRAD Standard capacitor will not work in your application because the 1pf models have 30pf stray cap to case.

Good Luck!

[e61_phil]

500ppm is well within reach of existing technology.

Measuring 2KVpp at 1MHz is a common technical problem in quadruple mass spectrometer. This problem has been resolved decades ago.

Although this is a solved issue, There are many technical details goes into it. A few point:

1. A oven is likely required
2. The RF amplitude(Vpp) is determined through active rectifier
3. A high voltage capacitor divider is needed to divide the 2KVpp sine wave done to 5V range for low voltage circuits
4. The capacitor divider will need to be calibrated (this is difficult and mass spectrometer don't need it, mass spectrometer need to the detector to be stable and precise, but not necessarily accurate.)
5. In exotic case, vacuum capacitor is used for dividing.
6. Custom vacuum capacitor were also developed to minimize thermal drift. The mechanical structure of the vacuum capacitor need to be compensated with different materials.
7. The analog signal processing chain will need high speed amplifier going into even GHz bandwidth, although you only measuring 1MHz signal

If you are still interested in this project, please contact me.

Hi,

Thanks for your reply! The system is almost exactly build up as you described. The voltage is adjusted with ions and it is stable.

My question points to your point 4. That is the difficult part. It isn't needed for the spectrometer, but it would be very nice for unit testing.

Best regards
Philipp

[1audio]

You could use an electrostatic volt meter. Sensitive research made them with .5% accuracy and they can be found. It's RMS sensing and good to DC and several MHz at the high end. There are several on eBay.

Sent from my LG-H830 using Tapatalk

[Gyro]

Unfortunately, Electrostatic voltmeters work on the basis of physical attraction between fixed and moving capacitor plates (vanes). They're great for 'lossless' measurement and low frequencies but Capacitance becomes the dominant loading factor at higher frequencies. Sadly it wouldn't meet the OP's <=3pF requirement.

Very high voltage models can have low capacitance but their lower voltage resolution is limited due to their square-law scale.

[1audio]

I overlooked the 3 pF requirement. The specs match a P6015A probe + a good digital scope. Checking calibration would be involved but not that onerious. A known 1 MHz 10V source would be adequate with a few steps.

[e61_phil]

I overlooked the 3 pF requirement. The specs match a P6015A probe + a good digital scope. Checking calibration would be involved but not that onerious. A known 1 MHz 10V source would be adequate with a few steps.

Do you really think so? My Keysight 4000 Scope isn't really specified for the Y-axis. Only 2.5% DC (which is already to much) and 3dB up to 350MHz.

I also doubt that the ratio on 10V is the same as at 2kVpp.

We have several 6015(A) probes. But they haven't such close tolerances. And if they have such tolerances: Who could calibrate them?

[1audio]

You would need to do some careful manual calibrations- first confirm the 1 MHz reference amplitude. Then check the attenuator on the scope ideally with an accurate 60 dB pad. Then check the P6015 on the 10 mV range and finally you should be there. I surprised the scopes calibration is so loose. However for a single point calibration you should be able to get there. with some patience with an 8 bit scope, certainly with a 12 bit scope. It seems we used to do that with an analog scope and a differential comparator amp, 7A13. 

[e61_phil]

Calibrating the scope might be easily possible. But the probe is the real problem.