active low-pass filter from Valhalla 2701C DC standard

[fenugrec]

Hi,
I'm looking at the analog portion of the Valhalla 2701C "precision DC voltage / current standard" and am intrigued by this one low-pass filter that I don't recognize.

For a bit of context , this unit works by generating a PWM waveform with a period of 40ms, filtering aggressively then driving the voltage/current amplifier (chopper amp, relay-switched feedback networks for different ranges; nothing very special).



Their low-pass filter is very interesting - instead of feeding the signal *through* the opamp like for example a Sallen-Key (which would hurt offset and drift specs), it's acting as a kind of "shunt" filter [EDIT - perhaps more akin to a capacitance multiplier ?] The only DC effect I can imagine is whatever leakage through the caps C6 and C7, but even that would be minimal since all nodes of the opamp should be within a few mV of 0 V.

I'm wondering if this is a common type of filter, how is it called, and how is it designed ?

[tggzzz]

The Tek 4x5 DM44 contains a similar filter. To function correctly it needs a constant source impedance, unsurprisingly.

[Kleinstein]

This type of question came up before, but AFAIK no name for this type of filter was found.
If nothing other helps one can do the design / choice of values with the support of a simulation. If needed one could so the circuit analysis by pen and paper the old way.

This type of filter and the higher order variants are used quite a bit in other instruments too, e.g. Fluke 5700, Datron 4910, Fluke 8840, Datron 1281. The analog filter in the 3456 is similar, though not exactly the same.

A change in the source impedance only shifts the cross over frequency. So some DMMs like the Datron 1061 have such a filter also at the input, where the impedance can change.

[Wimberleytech]

Never seen this before.  Excluding the effects downstream of the right side of the 99.5k resistor, we have:
G1 = 1/95k
G = 1/45k
C = 1u

H(s)= -G1*G² / s³c³ + s²*2c²G + SCG² + G₁G²

...assuming I did make some mistake

<edit>
lol

...assuming I did NOT make some mistake

[RandallMcRee]

This type of low pass filter is generally known as a "zero offset" or "no dc offset" filter.

E.g.
https://ir.library.louisville.edu/cgi/viewcontent.cgi?article=2077&context=etd
https://www.edn.com/zero-offset-active-lowpass-filter-part-1/

I think with better opamps now available it has seen its day in the sun.

[Wimberleytech]

This type of low pass filter is generally known as a "zero offset" or "no dc offset" filter.

E.g.
https://ir.library.louisville.edu/cgi/viewcontent.cgi?article=2077&context=etd
https://www.edn.com/zero-offset-active-lowpass-filter-part-1/

I think with better opamps now available it has seen its day in the sun.

Very nice!

[fenugrec]

This type of low pass filter is generally known as a "zero offset" or "no dc offset" filter.
...

Ah cool, thanks for the references !
Those implementations seem to lack C8 in the feedback path; probably 1 extra pole there ? Didn't try that in simulations.

Regardless of modern opamp performances, it's still a neat circuit ! Probably still useful in instrumentation when that "extra opamp" would really add too much drift. Or just added for street cred.

[Kleinstein]

This type of low pass filter is generally known as a "zero offset" or "no dc offset" filter.

E.g.
https://ir.library.louisville.edu/cgi/viewcontent.cgi?article=2077&context=etd
https://www.edn.com/zero-offset-active-lowpass-filter-part-1/

I think with better opamps now available it has seen its day in the sun.

For just a 2nd/3rd order low pass the zero offset advantage is not that large. One would normally have a buffer following the filter. A 2nd/3rd order active filter could however be done with just 1 OP too. There still is a small advantage, as the buffer amplifier does not need to be fast, while the OP in a Sallen-Key filter should be resonable fast. This could be an issue with AZ OPs.

The more interesting versions are higher order ones, especially those with more OPs. One would still only have a single DC relevant OP.

Another surprising use of similar filters seem to be with an inductor at the input and more powerful amplifier for use in power applications up to filtering mains harmonics as a kind of capacitance multiplier.

The filter looks a little similar to the GIC type filters, though still different and simpler.

[Someone]

This type of low pass filter is generally known as a "zero offset" or "no dc offset" filter.

Ah cool, thanks for the references !
Those implementations seem to lack C8 in the feedback path; probably 1 extra pole there ? Didn't try that in simulations.

The circuit you posted has 4 poles, no zeros. With the following designators:

The transfer function polynomial is:
0: 1
1: c1*r1 + c2*r1 + c4*r1 + c1*r2
2: c1*c2*r1*r2 + c1*c4*r1*r2 + c2*c4*r1*r3 + c2*c4*r1*r4
3: c1*c2*c4*r1*r2*r3 + c1*c2*c4*r1*r2*r4 + c2*c3*c4*r1*r3*r4
4: c1*c2*c3*c4*r1*r2*r3*r4

[Someone]

This type of question came up before, but AFAIK no name for this type of filter was found.
If nothing other helps one can do the design / choice of values with the support of a simulation. If needed one could so the circuit analysis by pen and paper the old way.

There are many interesting filter circuits that either never had design equations, or they were never published widely and the methods have been lost/kept private.

For just a 2nd/3rd order low pass the zero offset advantage is not that large. One would normally have a buffer following the filter. A 2nd/3rd order active filter could however be done with just 1 OP too. There still is a small advantage, as the buffer amplifier does not need to be fast, while the OP in a Sallen-Key filter should be resonable fast. This could be an issue with AZ OPs.

More comparable would be a Sallen-Key 3rd order with a passive RC after it, they both degrade similarly with reducing open loop gain or gain bandwidth of the opamp. But this shunt type filter does have lower noise at the output for practical implementations.

[David Hess]

Tggzzz mentioned that it was used in the Tektronix DM44 for the 465 series of oscilloscopes and I think Tektronix also used it in the 2236 oscilloscope which has a DMM built in.  I spent some time studying it when I first ran across it.

An alternative way to look at that circuit is as an operational amplifier based capacitance multiplier as part of an RC lowpass filter; it allows the use of low leakage film or ceramic capacitors in low frequency low pass filters.  The T-network provides the input bias current to the operational amplifier with the shunt capacitance increasing the effective feedback resistance as frequency increases.

[SiliconWizard]

Couldn't that be considered a variant of a gyrator?

[SilverSolder]

Same concept in Fluke 8505A:

[Kleinstein]

The simple version with just 1 OP has not that many advantages over a simple active filter. One still has the leakage from the 2 capacitors. The bigger advantage comes with the higher order versions. The other advantage is that one can use different OPs for there needs: a fast JFET bases OP for the filter part and a slow AZ OP for the buffer.

The circuit is a little similar to a gyrator, in that a special impedance is simulated with the OP circuit. However this time not an inductor, but more like a extra slow reacting capacitor or "super" capacitor.

[SilverSolder]

The simple version with just 1 OP has not that many advantages over a simple active filter. One still has the leakage from the 2 capacitors.

[...]

Do film capacitors have leakage issues, generally?

[David Hess]

The simple version with just 1 OP has not that many advantages over a simple active filter. One still has the leakage from the 2 capacitors.

[...]

Do film capacitors have leakage issues, generally?

In high impedance and high temperature applications, film capacitor leakage can become a factor.  Mylar (polyester) capacitors have higher leakage than the other types for example.

Film capacitor leakage is also a factor in precision applications.

[SilverSolder]

The simple version with just 1 OP has not that many advantages over a simple active filter. One still has the leakage from the 2 capacitors.

[...]

Do film capacitors have leakage issues, generally?

In high impedance and high temperature applications, film capacitor leakage can become a factor.  Mylar (polyester) capacitors have higher leakage than the other types for example.

Film capacitor leakage is also a factor in precision applications.

What would you expect from a good film cap e.g. 1uF with 10V on it...   more than 1pA?

[Kleinstein]

1 pA is not much current. So I would not be surprised to see more than 1 pA of leakage at 10 V. Even a reed relay may show more leakage.
It may be possible to get capacitors that good, but I would take it for granted.

Measuring the leakage of a 1 µF capacitor at 10 V, there are 2 additional problems: one is separating dielectric absorption from real leakage and to keep the voltage really constant. 1 pA with 1µF would give a rate of change of 1 µV/second, so a really slow drifting voltage. It would also need a stable temperature. With a TC of some 100 ppm/K for the capacitance, a changing temperature would result in about 1 mV/K of temperature effect.

[SilverSolder]

1 pA is not much current. So I would not be surprised to see more than 1 pA of leakage at 10 V. Even a reed relay may show more leakage.
It may be possible to get capacitors that good, but I would take it for granted.

Measuring the leakage of a 1 µF capacitor at 10 V, there are 2 additional problems: one is separating dielectric absorption from real leakage and to keep the voltage really constant. 1 pA with 1µF would give a rate of change of 1 µV/second, so a really slow drifting voltage. It would also need a stable temperature. With a TC of some 100 ppm/K for the capacitance, a changing temperature would result in about 1 mV/K of temperature effect.

Maybe we could try soaking the capacitor at 10V for a really long time, then disconnect the supply and measure the capacitor voltage as the first point...  - then, let it sit for a long time to leak down on its own...   finally, we come back later and measure its voltage as the second point, and calculate the leakage current based on the difference between the two measurements?  (in a temperature stable environment)

[Kleinstein]

Maybe we could try soaking the capacitor at 10V for a really long time, then disconnect the supply and measure the capacitor voltage as the first point...  - then, let it sit for a long time to leak down on its own...   finally, we come back later and measure its voltage as the second point, and calculate the leakage current based on the difference between the two measurements?  (in a temperature stable environment)

This is a reasonable method to get the real capacitor leakage. The difficult (but not impossible) part here is the switch to connect the voltmeter. It should be low leakage, probably switching the voltmeter between the cap and a 10 V source.
With enough time (e.g. 1l hour) the temperature is no longer that critical: 1 pA would still be 3.6 mV/hour and this equivalent to some 3.5 K in temperature change. Chances are one would do a few more than just 2 readings from the capacitor. So more like soak to 10 V for over night, measure for a few seconds second, wait 1 minute, measure, wait 1 hour , measure, wait 1 hour, measure.

[SilverSolder]

Maybe we could try soaking the capacitor at 10V for a really long time, then disconnect the supply and measure the capacitor voltage as the first point...  - then, let it sit for a long time to leak down on its own...   finally, we come back later and measure its voltage as the second point, and calculate the leakage current based on the difference between the two measurements?  (in a temperature stable environment)

This is a reasonable method to get the real capacitor leakage. The difficult (but not impossible) part here is the switch to connect the voltmeter. It should be low leakage, probably switching the voltmeter between the cap and a 10 V source.
With enough time (e.g. 1l hour) the temperature is no longer that critical: 1 pA would still be 3.6 mV/hour and this equivalent to some 3.5 K in temperature change. Chances are one would do a few more than just 2 readings from the capacitor. So more like soak to 10 V for over night, measure for a few seconds second, wait 1 minute, measure, wait 1 hour , measure, wait 1 hour, measure.

I'll have to measure the leakage of some of the GPIB switches I have here...  but maybe it isn't necessary with a switch at all?

Since it is relatively few measurements, the "switch" could simply be connecting a grabber for the measurement, then disconnecting it again.  There is no leakage when there is no switch...   and the grabber could be connected to 10V when not used to measure the cap?

[David Hess]

In high impedance and high temperature applications, film capacitor leakage can become a factor.  Mylar (polyester) capacitors have higher leakage than the other types for example.

Film capacitor leakage is also a factor in precision applications.

What would you expect from a good film cap e.g. 1uF with 10V on it...   more than 1pA?

1 pA is not much current. So I would not be surprised to see more than 1 pA of leakage at 10 V. Even a reed relay may show more leakage.
It may be possible to get capacitors that good, but I would take it for granted.

When I was designing my high temperature sample and hold, I figured that I needed a leakage specification at high temperature of better than 50 picoamps, at about 4 volts, because other sources of leakage, input bias current and clamped low leakage diode leakage, could be that high.  With a qualified 0.47 microfarad polypropylene capacitor, I did significantly better than that at high temperature.  At room temperature, droop was much better than 1 microvolt/second so less than 0.5 picoamps of which most was from the capacitor.

"Qualified" means that I bought a bunch of polypropylene capacitors types from different manufacturers and tested them all.  I did the same for the clamped low leakage diode which was just a 2N3904 collector-base junction.

[SilverSolder]

So, I tested a couple of capacitors using the method above (soak at 10.00000V overnight,  disconnect the supply in the morning and let sit for an hour, take the first measurement, let capacitors sit on the bench with nothing connected for 9 hours or so, then take a second measurement).

C1 -  Large Mica capacitor, rated 0.25uF at 100V
Start Voltage at 08:39 was  9.71413V
Stop Voltage at 17:49 was 8.04941V
Delta:  9.16h,  1.66472V
Lost Q:  0.41618E-6 Coulomb
Avg Leakage current: 12.6pA

C2 - Polycarbonate capacitor, rated 5uF at 50V
Start Voltage at 08:40 was 9.95304V
Stop Voltage at 17:50 was 9.79666V
Delta:  9.16h,  0.15638V
Lost Q: 0.7819E-6 Coulomb
Avg Leakage current: 23.7pA


Do the numbers seem reasonable, or did I mess up the math...