DIY Precision AC-RMS to DC Transfer Standard

[Qmavam]

Defining my problem.
 I just bought two HP400E's AC voltmeters.
I drove then with my metered HP651A, so all measurements are relative to the HP651A accuracy.
I measured each range from 1mV to 3 V.
Unit A reads 9% to 19% Low, Unit B reads 6% to 13% High.

This make me serious about building an RMS to DC converter.
 I'm not a designer, but I can build from a schematic, with pretty good knowledge of concerns about parasitics.
I think I could do a pretty decent job building the resistor/diode in a insulated package.
 My concern is, I have a need to measure 4mV between 500kHz and 2MHz.
What is needed to measure that low? It doesn't create me heat energy.
If I used a 50 ohm resistor, 4mV would only create 0.32microwatts at the diode.
Is that enough to heat the diode, to the point where I could get an accurate reading.
 I may need a several ranges to get to 3 volts.
 But to start, I could build just a 0 to 10 mV unit, to do the 4mv measurement I working towards.
 Any help appreciated.
                                    Mikek
                     

[Qmavam]

Hey guys, would it make any sense to use an LM35 and lap the cover down to a point where you are very close to the internal diode and
then mount your resistor. It has 10mV/C* output so starting with a gain of 20 and an inversion.
I've been looking for a mask or internal layout of the chip. no luck.
 Would it be better to lap the to front or back of a TO-92 case.
I thought the 8 pin case might be easier to work with but it also would have more thermal mass.
                                             Mikek

[amspire]

Ok, I read 5 and 6. He used a LM358 as an amplifier.
I see no mention of how much current he used as bias in the diode.
 Anyone have an idea, my concern is self heating of the diode,
 Am I just being over critical or is that a concern?

It doesn't matter how much heat the diode generates as the amount of heat is exactly the same in both the AC and DC phases of the measurement. The heat completely cancels out. You could use 1mA or 0.1mA - whatever you find works best.

Do I have the theory correct,  in that I can measure a dc voltage with good accuracy, I apply 1Vdc across the
100 ohm resistor and read the diode drop voltage.  Then I put AC across the 100 ohm resistor,
and adjust the amplitude until it reads the same as the 1Vdc did. Now I have 1Vac.

Yes.

I don't understand the 100*C target, is that a maximum temp?

In theory, the higher the better. Trouble is that with higher temperatures, unwanted things can happen. The SMD transistor package may only be rated at 125 deg C. Bonding glues could start to change. Also, if the resistor gets too hot, the long term resistance could start changing. This is all about getting very high stability - at least for a few minutes. If you can run it at 200Deg C and it is stable, then it would be easier to do the AC/DC transfer.

Also this "we can easily get 100MHz+ opamps that should have a flat enough frequency response to get DC to 1kHz accuracies of 0.01% easily."
 Why 100MHz opamp, aren't we amplifying a DC voltage from the C-B junction?
                                    Thanks, Mikek
 PS. I want to calibrate an AC meter up to 10 MHz, or at least 2MHz. (HP400E) I bought two and they don't agree.
I was happy when I only had one! :-)

The thermal transfer depends on heat, and unless you have both AC and DC sources that are stable with a fairly high current load, then you need to buffer the input. It is pretty common to see the load of a direct thermal AC/DC transfer to cause the output of a signal generator to drift too much for an accurate transfer.

The thing is, you do not want the AC performance of the buffer causing an error. If you want to measure AC up to 1KHz, then a 1MHz op-amp will have far too much error.  The output could be down by 0.1% at 1KHz. The faster the op-amp, the less likelihood of errors due to the op-amp. If you want to do a 0.01% transfer, then you need errors in the amplifier of much less then 0.01%. You need well over a 10MHz op-amp speed and that is just for 1KHz.

Richard

[Qmavam]

The thing is, you do not want the AC performance of the buffer causing an error. If you want to measure AC up to 1KHz, then a 1MHz op-amp will have far too much error.  The output could be down by 0.1% at 1KHz. The faster the op-amp, the less likelihood of errors due to the op-amp. If you want to do a 0.01% transfer, then you need errors in the amplifier of much less then 0.01%. You need well over a 10MHz op-amp speed and that is just for 1KHz.

Richard

I'm sorry, I still don't understand the 100MHz. As I see it we are only amplifying a DC signal and fairly slow changing DC signal.
Where does the AC come from?
 Ask another way, are we concerned about the AC characteristics of an opamp, if we are only amplifying DC signals?
                                      Thanks, Mikek
 Oh maybe the answer was in the previous paragraph.
"
The thermal transfer depends on heat, and unless you have both AC and DC sources that are stable with a fairly high current load, then you need to buffer the input. It is pretty common to see the load of a direct thermal AC/DC transfer to cause the output of a signal generator to drift too much for an accurate transfer."

So, are you say we need to buffer the input RF voltage that we are measuring?
 If that is the case, I don't thank that is a concern for me, all I want to do is drive/heat a 50 ohm resistor, just the load my sig/gen asks for.

                                    Mikek

[amspire]

The thing is, you do not want the AC performance of the buffer causing an error. If you want to measure AC up to 1KHz, then a 1MHz op-amp will have far too much error.  The output could be down by 0.1% at 1KHz. The faster the op-amp, the less likelihood of errors due to the op-amp. If you want to do a 0.01% transfer, then you need errors in the amplifier of much less then 0.01%. You need well over a 10MHz op-amp speed and that is just for 1KHz.

Richard

I'm sorry, I still don't understand the 100MHz. As I see it we are only amplifying a DC signal and fairly slow changing DC signal.
Where does the AC come from?
 Ask another way, are we concerned about the AC characteristics of an opamp, if we are only amplifying DC signals?
                                      Thanks, Mikek
 Oh maybe the answer was in the previous paragraph.

I mentioned 1KHz. For that, you want about 100MHz as I explained. What AC frequency are you wanting to measure? If it is 50/60Hz, then you can use a slower 5MHz opamp.

Anyway, I am not sure what the problem is - fast op-amps are not expensive.
https://www.aliexpress.com/item/32844836421.html?

If you put a buffer amp in, both the AC and DC have to go through the same buffer amp to attempt to cancel out the buffer amp errors.  Do you understand that the output error on a 1KHz signal through a 1MHz opamp has an error of at least 0.1% in amplitude? The 1Mhz opamp only has a gain of 1000 at 1KHz. The AD812 chip I linked to has a gain of 1,000,000 at 1KHz.

"
The thermal transfer depends on heat, and unless you have both AC and DC sources that are stable with a fairly high current load, then you need to buffer the input. It is pretty common to see the load of a direct thermal AC/DC transfer to cause the output of a signal generator to drift too much for an accurate transfer."

So, are you say we need to buffer the input RF voltage that we are measuring?
 If that is the case, I don't thank that is a concern for me, all I want to do is drive/heat a 50 ohm resistor, just the load my sig/gen asks for.

                                    Mikek

I am not sure who is mentioning RF voltages - the thermal transfer method can work great with RF, but there are many more potential errors.

[Qmavam]

Your sentence makes me think we are approaching the problem from two different solutions.
 I'm looking a using the thermal method with this circuit. I do note that is a high bandwidth device,
But I don't understand why when you are only amplifying DC. But, I'm sure there is a reason.
The problem is there are no LT1088 available, so another thermal input device is needed.

I'm open to another method, do you have a circuit.

                                    Mikek

[2N3055]

 
LT1088 has transistor pair thermally isolated from each other, each with its own thermally coupled resistor. That is exactly what Gyro (and others) made from discrete components.

On one resistor you connect your unknown voltage (AC or DC, or any combination). Electronics senses change in Vbe of transistor and starts changing voltage on other resistor until its temperature sensor transistor has exactly same Vbe. At that moment, second resistor will have voltage on it that will be exact DC equivalent of RMS voltage on first resistor.

You can connect measured voltage to input resistor directly, or you might need (most likely ) some kind of front end buffer/amplifier/attenuator in front of it.
For low frequency bandwidth it might be enough to calibrate the input with DC.

[chuckb]

The LT1088 is available on Ebay.

If you have an oscilloscope, does it have a 5V square output for calibration of the probes? That output can be 1% accurate on some scopes. I would put a small RC (1 kohm and 0.01 uf) to filter the edges of the square wave so the energy is inside the bandwidth of your voltmeter. If your meter errors are the same with the HP651A and with the filtered scope output, then use the HP651A to calibrate the meters.

If the Voltmeter error is constant on all ranges and frequencies they may just need the meter movement gain adjusted.

[Qmavam]

The LT1088 is available on Ebay.

If you have an oscilloscope, does it have a 5V square output for calibration of the probes? That output can be 1% accurate on some scopes. I would put a small RC (1 kohm and 0.01 uf) to filter the edges of the square wave so the energy is inside the bandwidth of your voltmeter. If your meter errors are the same with the HP651A and with the filtered scope output, then use the HP651A to calibrate the meters.

If the Voltmeter error is constant on all ranges and frequencies they may just need the meter movement gain adjusted.

I'll look at that, I have Tektronix 2465 that may have a good calibrator.
                       Thanks, Mikek
Edit:
 I tried that, I have not seen a calibrator like the 2465 has, it changes frequency with the horizontal time base.
 Anyway it is designed to be able to drive 50 ohms at 1/2 voltage or 1M ohm at full voltage. (0.2vpp or 0.4vPP) New info to me.
So I set it up without you filter because that was easy. With a 50 ohm termination the meter read 0.119V RMS
with supposedly 0.2Vpp signal in. I thought, well that messed up, but, this is a squarewave and the HP400E is an average responding, RMS calibrated meter, Making me think, the average of a squarewave is 50% so 50% of 0.2V is 0.1V.
Can I get a confirmation?
 So it reads 19% high. The other unit reads 0.109 or 9% high.
 I put the filter in and (I removed the 50 ohm termination) the reading dropped low, so I decreased the cap by a factor of 100, this just makes me think I can get any value by picking the right cap. So I dropped the filter idea. FWIW, the scope reads 4.3% high as far as my eye can tell.
                                    Not sure I learned anything,  Mikek
 I have sent out for calibration quotes.

[chuckb]

I thought, well that messed up, but, this is a squarewave and the HP400E is an average responding, RMS calibrated meter, Making me think, the average of a squarewave is 50% so 50% of 0.2V is 0.1V.
Can I get a confirmation?

I don't know what the HP400E will read with the square wave input. I glanced at the Tek Manual, It said the probe calibration output was to be used with 1ms / division.

One path is -
Calibrate the scope with x1 probes (you will be working with small signals), 1 Meg input Z.
Use the scope to calibrate the HP651 signal generator. 1 Meg input Z
Use the generator to calibrate your meters.

[IconicPCB]

a miniature dual filament incandescent bulb could be used as a basic building block... two such bulbs could form a half of a self balancing wheatstone bridge.

[2N3055]

Let's go back to original problem: how to check whether HP400E is reading correctly.

You need relatively stable signal source. That means signal generator. Chinese FY6900 or similar is plenty good for that. If you have any other signal generator that use that.
You need a RMS reading multimeter. Any handheld (even cheap ones) will be precise enough to verify 400E reading.
Use sinewave, 400E is not thermal (true RMS) instrument. It says RMS, but it is average reading with RMS calibrated scale.
Get 400E manual on Internet. Read it. There is a performance verification procedure.

Despite being cool and all, 400E is not really any better at measuring things than, say, UNI-T 60E.

[Qmavam]

I did a search for UNI-T 60E specifications, and didn't find it.
I found the  61 and it's spec's only say 10kHz bandwidth.
 Does the UNI-T 60E have a 10 Megahertz bandwidth?
                                        Mikek

[2N3055]

I did a search for UNI-T 60E specifications, and didn't find it.
I found the  61 and it's spec's only say 10kHz bandwidth.
 Does the UNI-T 60E have a 10 Megahertz bandwidth?
                                        Mikek

Sorry, my bad, I missed 500KHz to 2 MHz in your first post.
In that case what ChuckB said..

[Qmavam]

I sent the HP400E in for calibration today, I will end up spending more to have the unit calibrated that I paid for it.
 But at least I will have something I can trust, for a while.
                                       Mikek

[babysitter]

You made me thinking of using a laser cutter for decapping, with a high speed setting and low power, tough.
Black epoxi could be simple to remove. shiny metal and silicone reflects the beam so not much energy will stick.... SiO2 isolation layers might burn up, at least it was simple to engrave glas. Hm, ok, nevermind.

[MegaVolt]

Re-reading the topic came up with an idea. We need a vacuum, a heater, a temperature meter. This can all be found in thermostated crystal oscillators. Inside the vacuum flask there is a quartz crystal with a sprayed heater. Measuring the frequency deviation is not difficult.

Similar vacuum resonators were used in precision generators. And I think now it's not very difficult to find them

[Kleinstein]

The heater for a OCXO is quite slow to react to the crystal. In addition the crystals used in precision gear tend to be low TC and not really good for a temperature measurement. In addition the heater may not be a low parasitics resistor to high frequencies. A vaccuum is not absolute needed, it is just a way to get good isolation.

[mawyatt]

An effective method to create an accurate AC waveform from a known DC value is to create as close to an ideal squarewave as possible with the AC amplitude Vpp of the Vdc value and a fast rise/fall time relative to the period with a 50% duty cycle. This waveform has an average DC value and RMS value of Vdc/2.

A simple CMOS Flip-Flop (74AC74) creates a very accurate 50% duty cycle at low frequencies and swings the unloaded output to VDD and VSS. Make VSS ground and VDD a reference voltage, say 5.000 Volts, then the AC output will be 2.500Vrms, and the average DC value 2.500Vdc. You can buffer the FF output for a lower output impedance with a CMOS driver or discrete CMOS buffer. This works really well as we've done this quite a few years ago as an in-house AC reference.

Of course the waveform isn't perfect with finite rise and fall times, however, as long as the waveform period >> rise/fall time the error is small. If our analysis is correct the error is ~ -6*Trise/Tperiod, with equal rise and fall times assuming linear slopes, so a 10ns rise/fall in a 100Hz squarewave is just 6ppm low.

Another source of error is the meter bandwidth in reading the squarewave harmonics and the ideal harmonics fall off as 1/N. Since a squarewave contains waveform energy in the harmonics, this energy is attenuated by the meter bandwidth. For example assuming a meter has an simple low pass characteristic then the effective noise-bandwidth NBW is pi/2 times the BW, then the error will be the squarewave frequency F times /(2*NBW), or F/(2*NBW). So a meter with a 1MHz BW and a 100Hz squarewave will read 100Hz/(pi*1MHz), or 32ppm low.

Edit: Math error corrections!!

Anyway, the squarewave is easy to generate accurately and can serve as a low frequency AC reference is many cases where the ultimate precision isn't required.

Best,

[Conrad Hoffman]

Curious if a different way has merit. We can build a low THD sine generator, 0.001% or better. It should be possible to control the gain of the oscillator by comparing the waveform peak to a DC reference voltage. I don't see why this couldn't be done to an extremely fine degree. Not sure how fast it could be done, but it eliminates the harmonic issues. If you know the peak and it's a near perfect sine, you know the RMS.

[1audio]

You just described the Fluke 510. https://xdevs.com/doc/Fluke/510A/510A_AA_imeng0000.pdf  Set the reference voltage to 14.1214V and you get 10V out.

[Kleinstein]

Curious if a different way has merit. We can build a low THD sine generator, 0.001% or better. It should be possible to control the gain of the oscillator by comparing the waveform peak to a DC reference voltage. I don't see why this couldn't be done to an extremely fine degree. Not sure how fast it could be done, but it eliminates the harmonic issues. If you know the peak and it's a near perfect sine, you know the RMS.

Some (could be many) of the calibrators produce there AC reference that way.
Measuring the peak voltage is still not easy.

[Conrad Hoffman]

The Fluke circuit seems excessively complicated, but that's coming from somebody who's never done it. Need to think about how to do it with modern parts/methods. It just seems like this method is more accessible to the average volt-nut than thermal. I have a 540 transfer standard and have never trusted it. Also takes too many voltages to run it. I think they also came with a correction sheet of some sort, that I don't have. Anybody have an example of that?

[mawyatt]

Would think a good high resolution DAC could produce a low distortion, low frequency sine wave that has accurate amplitude if derived from a good reference.

One could also extend this to sampling the peak either by an ADC or analog technique and compare to a fixed reference for feedback like Fluke did. Since the sinewave is created by the DAC the peak should be known and thus the sample point. If an ADC is utilized to resolve the peak it could be "offset" and expanded to a small region around the expected peak amplitude. Control loop speed isn't an issue and a SD ADC could be employed with high preamp gain around the expected peak voltage.

Best,

[dietert1]

A modern part with extremely small distortions would be a 24 Bit Audio ADC. One could run the left channel on the sine and the right channel on the chopped DC reference. Those ADCs are pretty cheap and sample up to several hundred KHz.

Regards, Dieter