Measuring INPUT Power Accurately and with no Oscilloscope

[poynt99]

Using a "switching circuit" that exhibits self-oscillation as an example, let's examine the possibility of making an accurate INPUT power measurement using only two readily available digital multi-meters set on DC Volts, and a 0.25 Ohm CSR.


The schematic multi-fet-schema04.png illustrates the 3 measurements taken for all scope shots.

The purple trace is instantaneous power p(t) taken directly from the battery as shown. In the last scope shot, the average of this trace is shown to indicate the real power supplied by the battery.

The red trace is the voltage across the CSR, including the parasitic inductance associated with it. The trace value is divided by the CSR value 0.25 to indicate the instantaneous current i(t).

The green trace is the voltage measured at the point shown. Some experimenters erroneously call this "the battery voltage", and as such it has been labeled "BAT" in the scope shots. This is the instantaneous voltage v(t) representing what is supposed to be the true battery voltage. We will see however that using this two-DMM measurement method does not appear to be detrimentally-affected by the dislocated battery voltage measurement.

The scope shots multi-fet-wave07.png through multi-fet-wave09.png illustrate the wave forms we are dealing with here; quite dynamic, high frequency components, burst-mode oscillation, trace values above and below zero. This is not necessarily a situation that is easy to nail down in terms of an INPUT power measurement. However, by examining the values shown, you will see that the averaging method is "un-phased" by the oscillatory nature of the traces, and is able to extract the correct values for both the average battery voltage and average CSR current.

Simply multiplying the average battery voltage (60V) by the average CSR current (2.8A) produces a result that is the same as the actual real power being supplied by the battery (168W), as determined by the Watt probe placed directly on the battery.

Providing this works in all cases and is not erroneous, this is good news for experimenters. It would appear that stray inductance both in the battery feed lines and that associated with the CSR resistor have no affect on the true values required to obtain the INPUT power measurement, if this measurement method is utilized.

1) Now an accurate INPUT power measurement is readily obtainable by anyone, and there is no reason not to perform this measurement.

2) In cases where excessive heat is claimed as the output, the true COP can now be calculated by comparing this INPUT power measurement with the power required to produce the equivalent output heat using a simple DC source as a control.

3) 3) This method can be applied similarly to any circuit using an input DC voltage source.

.99

[MileHigh]

Hey Poynt,

A question about the use of the multimeters.  Are you saying that you just have to connect the multimeters across the the measurement points and the multimeters alone can do the averaging?

I remember how you showed that the DC averaging built into all multimeters is very accurate.  If I recall there is an RC-type circuit inside the multimeter that gives you the infinite-resolution averaging function.

On the other hand, Humburger showed a nice elegant solution where he put an RC low-pass filter across the inductive shunt resistor.  Certainly that will work but would it be fair to say that it can be considered redundant?

Well I can't resist commenting on the irony about how a whole lot of fuss could have been avoided by using some humble $20 multimeters instead of a DSO.

On that topic I would hazard a guess that the 'team' never double-checked if the apparent very large currents as seen across the shunt resistor were also confirmed to exist across the heating element.  For example, if you believe that there are 30 amps going through the shunt resistor, then you should double-check that by looking at the voltage across the 11-ohm heating element.  There should be (30 x 11) = 330 volts across it.  I doubt that the 'team' ever did this.  If my assumption is true, it's just another baffling mystery associated with the 'experts.'

MileHigh

[poynt99]

Hey Poynt,

A question about the use of the multimeters.  Are you saying that you just have to connect the multimeters across the the measurement points and the multimeters alone can do the averaging?

MH, yes. The only consideration here would be to try and keep the meter leads away from being in parallel with any current-carrying wiring, so as not to induce a voltage. Some Farady shielding around the main part of the apparatus would be helpful in mitigating interference as well.

I remember how you showed that the DC averaging built into all multimeters is very accurate.  If I recall there is an RC-type circuit inside the multimeter that gives you the infinite-resolution averaging function.

Most meters use an "integrating A/D" conversion method, so the averaging is essentially built in. It is accurate, and out to at least a few MHz.

On the other hand, Humburger showed a nice elegant solution where he put an RC low-pass filter across the inductive shunt resistor.  Certainly that will work but would it be fair to say that it can be considered redundant?

It is somewhat redundant, but it can only help. It certainly will not do any harm. In particular, it may prevent damage to the meter by taking care of most of the high voltage transients.

I would like to try this method but with the CSR replaced with a DMM used as the CSR directly. If this works, then not only have we conquered these INPUT power measurements for good, but no additional CSR's nor RC filtering would be required.

.99

[MileHigh]

Poynt:

Most meters use an "integrating A/D" conversion method, so the averaging is essentially built in. It is accurate, and out to at least a few MHz.

I just realized that there is a potential monkey wrench associated with the averaging bandwidth of the multimeter and the waveform you are trying to measure that may skew the results a bit.

It seems like we have seen high-frequency waveforms with sharp peaks/spikes above ground and smoother rounded curves below ground.  So depending on the bandwidth of the waveform being averaged and the bandwidth of the multimeter averaging function, there could be problems.

When you run the simulation I assume the bandwidth of the averaging is one-half the step frequency.  I also assume that your step frequency is very very high.

So, it might be a worthwhile exercise to see what happens if you put low-pass filters between the the circuit nodes you are looking at and the measurement points.

Note that there is still a back-up plan, using Humbugger's discrete low-pass filter across the current sensing resistor.  The assumption being that the bandwidth of the averaging for the discrete low-pas filter is potentially much higher than the bandwidth of the multimeter dual-slope integrating A/D conversion function.

The more I think about it the more it makes sense.  You are eliminating the long leads of the multimeter like this.

I suppose that you could even put a choke on the end of your positive scope lead so that you prevent the scope lead from sucking up any AC energy that you want to go exclusively into the averaging filter.  Just thinking out aloud.

MileHigh

[humbugger]

Pretty easy to test.  Just use a filter and no filter and see if there is dead-on agreement.  I put my faith in the RC because the bandwidth is pretty much infinite, at least until series L of the cap starts approaching say 1% or more of the R.

A good test would be to try a few MHz PWM-type signal with a very lopsided duty cycle 98/2 or something and a range of DC offsets.  then check with just the DMM and then with the DMM plus RC filter.  See if the numbers agree.

Hum

[poynt99]

You bet guys,

That is another added benefit of using a front-end RC filter.

The sim has no bandwidth per se. It uses a running average.

.99

[ion]

There are few blanket statements about integrating A/D converters that can be made regarding noise rejection.
A lot depends on the type of integrating converter used. There are single slope, dual slope, triple slope etc with or without autozero functions.

I have designed all of these types as part of my livelihood from discrete components (op amps, cmos switches and microcomputer control and timing). They served as the low level thermocouple data acquisition portion of temperature measurement and control devices.

While integrating A/D converters tend to be relatively noise immune compared to other types, there are some caveats, especially with the cheaper single slope types.

For dual slope types Normal Mode Rejection is enhanced by setting the integrating time period to an exact multiple of the measured frequency.

I have compared cheap meters vs additional RC filtering and found little difference in the result, which can be far more accurate than trying to set up and measure extremely noisy signals with a DSO.

I have in the past posted these results for the JT, but as with forums it all gets buried and forgotten.

[poynt99]

Thanks for that timely post Vince.

I'll be trying with and without the RC filter, as a double-check.

.99

[humbugger]

Poynt should clearly preface this thread, I think, with the statement that this method only works in the case where the power factor on the input is known to be 1.00

This will always be the case in a constant voltage DC-powered circuit or where it is known for sure that PF=1.00 (by other means of testing) in a circuit powered by a source which is totally, primarily or even significantly AC.

In other words, the averaging of individual voltage and current measurements and then multuplying the averages to find input power is only acceptable if we know for sure that the power factor is 1.00.  In any other situation, the method will only report "apparent" power and cannot make any distinction between real power and reactive (reflected) power.  Take the case of hooking a capacitor across the AC line, for instance.  You'd measure an average voltage of 120VAC and some average current depending on the capacitor value.  You'd mulltiply them and get a positive number.  But the real power is close to none.  The averages would show the reactive power, a.k.a. reflected power.  The only real power would be whatever is dissipated as heat in the wiring and the capacitor's internal resistance, a much smaller number than would be shown by multiplying the two averages.

That common knowledge was the basis which, sadly, was used by many here to convince each other and a certain Miss Ainslie that all the multiplying had to take place before any averaging...a misconception that she now clings to tenaciously despite the fact that hers is truly a DC-operated circuit where the actual input DC voltage is a constant with quite negligible AC ripple.  She now is convinced and is trying very hard to convince others that Poynt keeps changing the rules regarding this; earlier being a strong proponent of "multiply before averaging" and now arguing the opposite.  

The reality is that none of us "old pros" had the foresight to realize that Rosemary was, of course, going to hook up her battery stack using 30-odd feet of wire and no bypass or local energy storage caps anywhere and then measure the "battery voltage" at the wrong end of those long wires so that the scope would be fed a gigantic AC waveform as the voltage argument of the multiply.  Nor did we much think about the inductance of the shunt causing that signal (the current argument of the multiply) to be shifted in time, amplitude and waveform as well, by almost 90 degrees and fourfold amplitude.  

In defense of the poor foresight of everyone advising her at the time, one of the big reasons we all underestimated the importance of these things at the time is that none of us knew that Rosemary had changed the ballgame from using a 2kHz primarily hard-switching circuit to using a full-blown continuous linear feedback mode oscillation at 1.5 MHz, where the importance of these inductance effects becomes overwhelming.

Essentially, Rosemary has ended up feeding mainly di/dt (shunt) and -di/dt (battery) signals into her scope on both channels instead of sending true input voltage and input current signals to be multiplied.  She has the scope multiplying the rate of current change times -1 times the rate of current change as provided by the voltages taken across shunt and battery wiring inductances that are overwhelmingly dominant over the shunt and load resistances and hundreds of times larger than any battery internal resistance.  

And some of us on this forum and on prior forum discussions elsewhere inadvertently reinforced the idea in her mind of the correctness of that method because we agreed in the general case that the real-time multiplying had to be done before the averaging.  This was correct advice ONLY if both the voltage and current were time-varying quantities (and they are not)  and (as we assumed wrongly) if proper precautions were taken to assure that the actual battery voltage and a correct analog of the actual current were the terms being multiplied.

In retrospect, we all knew that batteries were being used and that therefore, the voltage argument of the multiply was going to be essentially a constant with possibly a very small AC ripple, so the whole argument over whether to multiply first or average first becomes moot when one term of the multiply is a positive constant.  We should have focused our collective "educate Rosemary" efforts on getting her to understand that her battery voltage is not 158VAC at 1.5MHz.  20/20 hindsight.  Oh, well...Rosemary has proven to be uneducable anyway.  

Now that that is off my chest, the real reason I'm posting this time is to show you something I have just tonight realized for the first time in all of this:

IT IS EASY AS PIE TO COMPENSATE FOR THE INDUCTANCE OF A SHUNT IN REAL TIME AND OVER AN ENORMOUS BANDWIDTH!

So, in the event we ever actually needed to do real-time "multiplying before averaging" using a scope-sampling math function (say in the case where the power supply is a 1MHz square wave by chance or a DC source with high impedance and lots of ripple or even 60Hz sinewave mains power with heavy harmonics) we can fairly easily get our inductive current shunt to give us true-to-life real-time waveforms that are in-phase and amplitude correct and waveshape-identical no matter how complex the waveform.

How?  Simply use the RC filter technique I've already shown many times but with one difference:  Instead of pounding to a damn-near ripple-free DC average by heavily filtering (RC time constant hundreds or thousands of times longer than the waveform period), we simply set up our R and C such that the product of R*C is equal to the quotient of L/R (shunt inductance over shunt resistance).  Now the inductive shunt's time constant is exactly compensated by the filter's time constant.  The result is exactly the same as if we were somehow able to eliminate perfectly the shunt's inductance.  And it's broadband.  All the signal differentiation done by the unwanted inductance is exactly undone by the reciprocal integration of the RC filter.

Below, I show a +/- 1A current trace of arbitrary shape with lots of rich harmonics and sudden slope changes and asymmetries to represent a signal with energy over a wide bandwidth.  You can see how the scope is hooked up.  Both traces are exactly identical and superimposed perfectly at all points.  So the effect is just as if we had an ideal pure resistance for our shunt.  No phase shifts, no distortions.  Ready for real-time sampling and multiplying.

Humbugger

[poynt99]

Nice trick Hum.

If all goes well, there won't be any need to utilize it, but it is always an option. As you mentioned, being that 99% of the projects being built and tested by FE enthusiasts use DC supplies as the source, the averaging method is the way to go IMHO. The simpler the better.

.99

[humbugger]

Nice trick Hum.

If all goes well, there won't be any need to utilize it, but it is always an option. As you mentioned, being that 99% of the projects being built and tested by FE enthusiasts use DC supplies as the source, the averaging method is the way to go IMHO. The simpler the better.

.99

Agreed.   

[humbugger]

As far as input power measurement goes, for constant-voltage DC powered devices, the averaged-current method as described here by Poynt is bound to be very accurate, limited only by any bandwidth limitations on the DMM's averaging ability (as has been discussed and as is being checked out).  If those limitations become a problem, adding a simple external RC lowpass filter will take care of it.

It is, however, still very worthwhile to have a way to visualize current flow on your scope and to be able to trust that the scope trace is accurate and properly time-aligned with any other traces you are observing simultaneously.  For those of us that do not have a $4,000 wideband AC/DC current probe, the method I described above is a very good and totally inexpensive way to accomplish that.

If your circuit is a slow-pulsing or low frequency affair and there are no sharp current transitions involved or frequencies above a few kHz, then just using a low-inductance shunt is probably fine for all purposes.  But if your circuit involves sudden changes in current flow, current spikes that you'd like to accurately measure and visualize or any other high frequency AC or pulsed DC currents, you will get a distorted picture of the timing, amplitude and waveshape of your instantaneous currents unless you take measures to compensate your shunt against its parasitic inductance.

Humbugger

[GibbsHelmholtz]

How?  Simply use the RC filter technique I've already shown many times but with one difference:  Instead of pounding to a damn-near ripple-free DC average by heavily filtering (RC time constant hundreds or thousands of times longer than the waveform period), we simply set up our R and C such that the product of R*C is equal to the quotient of L/R (shunt inductance over shunt resistance).  Now the inductive shunt's time constant is exactly compensated by the filter's time constant.  The result is exactly the same as if we were somehow able to eliminate perfectly the shunt's inductance.  And it's broadband.  All the signal differentiation done by the unwanted inductance is exactly undone by the reciprocal integration of the RC filter.

Brilliant.  You are something Humbugger.  So RC = L/R ==> C=L/R^2 .  That's some nanoH over Ohm.

This will always be the case in a constant voltage DC-powered circuit or where it is known for sure that PF=1.00 (by other means of testing) in a circuit powered by a source which is totally, primarily or even significantly AC.

In other words, the averaging of individual voltage and current measurements and then multuplying the averages to find input power is only acceptable if we know for sure that the power factor is 1.00.  In any other situation, the method will only report "apparent" power and cannot make any distinction between real power and reactive (reflected) power.  Take the case of hooking a capacitor across the AC line, for instance.  You'd measure an average voltage of 120VAC and some average current depending on the capacitor value.  You'd mulltiply them and get a positive number.  But the real power is close to none.  The averages would show the reactive power, a.k.a. reflected power.  The only real power would be whatever is dissipated as heat in the wiring and the capacitor's internal resistance, a much smaller number than would be shown by multiplying the two averages.

Good, now we all know that the circuit PF is 1.00.  Why she even bother use a switch.  Just connect the battery straight to the load.  And all this fuss over out of phase thing when we know the PF is 1. 

[humbugger]

So RC = L/R ==> C=L/R^2 .  

Gibbs...the two Rs are not the same here so R^2 is not appropriate.

[GibbsHelmholtz]

Gibbs...the two Rs are not the same here so R^2 is not appropriate.

You are a courageous man.

I take it that you think RC and L/R separately.   In that case, R and C can varies independent of L and R so that the product or quotient remains the same.  The L is compensated by the C leaving the exact R value at anytime, not the L is compensated by some RC.

[humbugger]

You are a courageous man.

I take it that you think RC and L/R separately.   In that case, R and C can varies independent of L and R so that the product or quotient remains the same.  The L is compensated by the C leaving the exact R value at anytime, not the L is compensated by some RC.

To perform the cancellation of the effects of Lshunt, when Lshunt is in series with Rshunt, we must use any set of values for Rfilter and Cfilter that satisfy the equation Lshunt/Rshunt = Cfilter*Rfilter.  In practice, Rfilter should be >> Rshunt and << than the input impedance of the measuring device.

The equation cannot be reduced or rearranged such that Rshunt * Rfilter is expressed as R^2 except in the single case where Rshunt = Rfilter.

I cannot quite understand, Gibbs, whether you are arguing with me or exactly what. 

Humbugger

[GibbsHelmholtz]

To perform the cancellation of the effects of Lshunt, when Lshunt is in series with Rshunt, we must use any set of values for Rfilter and Cfilter that satisfy the equation Lshunt/Rshunt = Cfilter*Rfilter.  In practice, Rfilter should be >> Rshunt and << than the input impedance of the measuring device.

The equation cannot be reduced or rearranged such that Rshunt * Rfilter is expressed as R^2 except in the single case where Rshunt = Rfilter.

I cannot quite understand, Gibbs, whether you are arguing with me or exactly what. 

Humbugger

What you saying make sense, but I found that only one RC that could do the cancellation.  Well, this is my argument.

I posted a pix below showing the case of an ideal resistor and a case with L cancellation. Initially if you apply a voltage source of say 1V to the 1 Ohm ideal resistor, you would expect 1V if measure across the resistor, vice versa for the current source case.  Agree?  Now if you apply the same to the cancellation case. Initially, L will act as infinite impedance( or an open leg).  This would allow all current to pass through the bottom leg.  If R on bottom is not identical to top, I don't know how it going to give you 1V measurement. 

[humbugger]

I think maybe I see where the confusion comes from, Gibbs.  You seem to be thinking about cancelling the effect the inductance has on the power path of the circuit itself.  That was never my claim or intention.

The method I describe is effective to eliminate the inductively-induced errors from the measurement of current by compensating the measurement path, not to actually cancel the inductance out of the power path.  The latter is much harder to do and the approach to that entirely different problem that you show would not work anyway.

The problem we are trying to solve (and have succeeded in solving) is to extract an accurate picture of the true current flow in the power path by exactly cancelling the di/dt term due to the differentiator action of the inductance by adding an equal capacitive integrator term to the measurement path...so we can see the voltage as it really is across just the resistive part of the shunt.

Hum

[GibbsHelmholtz]

I think maybe I see where the confusion comes from, Gibbs.  You seem to be thinking about cancelling the effect the inductance has on the power path of the circuit itself.  That was never my claim or intention.

The method I describe is effective to eliminate the inductively-induced errors from the measurement of current by compensating the measurement path, not to actually cancel the inductance out of the power path.  The latter is much harder to do and the approach to that entirely different problem that you show would not work anyway.

The problem we are trying to solve (and have succeeded in solving) is to extract an accurate picture of the true current flow in the power path by exactly cancelling the di/dt term due to the differentiator action of the inductance by adding an equal capacitive integrator term to the measurement path...so we can see the voltage as it really is across just the resistive part of the shunt.

Hum

I think I see your point of view.  You can eliminate the inductance, but only in average form and I'm trying to do an instantaneous form.  Well, I hope that I already convince you that the only correct method for calculating power is multiply first then take an average.  If you multiply the average, then you can only get apparent power.  So if we can't eliminate the inductance instantaneously, then we still have error in power measurement (just a little bit).  I have high hope for my method though.  Think about it, what's the different between my RC and your RC?  not the product.  Then there must be a dynamic different then.  What is this dynamic difference looks like is the problem I'm tackling now.  It's interesting because if that is the case, we have a method to efficiently tune high Q circuit. 

[PhysicsProf]

  I'm still traveling, very limited time and computer connection, but this is an important thread IMO and I have a question.

Working on the "reverse-JT" circuit described on my bench, I saw the attached waveforms for Pin (red) based on Vin (2V DC applied by a power supply, yellow waveform) and Iin (through a 1 ohm resistor, blue waveform).  
Note that the transistor gates the flow so their is only a trickle and that has an "AC component" -- the impedance of the circuit (etc) is such that there is essentially no "DC current" flowing; the voltage across the 1 ohm input CSR is shown in blue.  

The MEAN power IN is very small, 2.3 mW (red waveform).

Under these conditions, will your method (.99) work?  - will it give the same result as the Tek 3032 for the MEAN power in?

[PhysicsProf]

PS -- you can tell the periodicity from the waveforms; the frequency is approx 1.4 MHz.

Would like to know whether .99's method would work to give an accurate measure of input power under these circumstances (near zero, oscillating, etc.)
The PS is a DC power supply set at 2.0 volts, but the waveforms are much the same when I used two AA batteries (about 2.8V).

[humbugger]

Gibbs, are you taking lessons from Rosemary Ainslie?  Your replies keep getting further and further off track.

WRONG!  The method I describe is a real-time method and does not require averaging.  It is intended for use specifically where it is desired to observe the current in a circuit and the inductance of the shunt distorts the current waveform.  Using the method I described, the true current waveform is re-established in real time and without phase shift (time delay) or other distortions.

Humbugger

[poynt99]

 I'm still traveling, very limited time and computer connection, but this is an important thread IMO and I have a question.

Working on the "reverse-JT" circuit described on my bench, I saw the attached waveforms for Pin (red) based on Vin (2V DC applied by a power supply, yellow waveform) and Iin (through a 1 ohm resistor, blue waveform).  
Note that the transistor gates the flow so their is only a trickle and that has an "AC component" -- the impedance of the circuit (etc) is such that there is essentially no "DC current" flowing; the voltage across the 1 ohm input CSR is shown in blue.  

The MEAN power IN is very small, 2.3 mW (red waveform).

Under these conditions, will your method (.99) work?  - will it give the same result as the Tek 3032 for the MEAN power in?

Professor,

In theory the method works for all cases where a DC source is used to supply power to the circuit. The wave forms become immaterial with this method, as they are heavily averaged.

Be patient, as I plan on doing some testing this weekend.

.99

[GibbsHelmholtz]

Gibbs, are you taking lessons from Rosemary Ainslie?  Your replies keep getting further and further off track.

WRONG!  The method I describe is a real-time method and does not require averaging.  It is intended for use specifically where it is desired to observe the current in a circuit and the inductance of the shunt distorts the current waveform.  Using the method I described, the true current waveform is re-established in real time and without phase shift (time delay) or other distortions.

Humbugger

Alright, then why your RC works and my RC don't work?  Or are they both work?

[PhysicsProf]

Professor,

In theory the method works for all cases where a DC source is used to supply power to the circuit. The wave forms become immaterial with this method, as they are heavily averaged.

Be patient, as I plan on doing some testing this weekend.

.99

I look forward to your test results, .99 -- thanks.