EEVblog #929 - Designing A Better Multimeter

[Kleinstein]

Usual depletion mode FETs are high R_on and only for low current.  There are special ones (e.g. GaN based) for high currents but these are really expensive and still not that low in R_on. So depletion mode fets are not an option.

One might be able to use a depletion mode FET to control the gate, so that in off mode the fets will be on - as long as the battery is still alive. So not using the meter to conduct a high current with the batteries empty would be the users responsibility.

Depending on cooling the meter might still survive quite some current even without power - as the fet gets hot, the forward voltage will drop and thus reduce the power a little. I have not seen thermally operated switches that turn on at high temperature - this would be a option for this rare case.

[David Hess]

IXYS has high current depletion mode MOSFETs available but they are not cheap.  The ones made by Supertex (bought by Microchip) are lower current.

[splin]

Using the (parasitic) reverse diodes of the MOSFETs for protection purposes will affect the precision of the sensitive ranges, due to their leakage currents. These lower ranges resolve nA, maybe pA currents.

Especially, if you apply a reverse current, in the "negative" direction of the DMMs input jacks, these 50mV burden voltage already create relevant forward currents in these MOSFET diodes, and if you combine two of such MOSFETs in parallel, this would be the case for either direction.

More common precision current protection schemes use either the low leakage of the B-E diode of a bipolar transistor, like in the HP3458A, which includes power dissipation protection by software, or they use at least two diodes of a rectifier bridge in series, combined with an OpAmp to reduce these diode leakage currents to near zero, like in most of HPs 6 1/2 digits DMMs.



Frank

Does anyone have any information on those transistors? They are listed in the BOM as SQW5011A (or similar) 'TRANSISTOR-PNP SI TO-220AB PD=1.67W'. I can't find anything having tried various combos such as 5QWSOIIA, etc.

Presumably they were selected for very low leakage which I'm guessing is unusual in a power transistor.

Also, what might be the purpose of the purpose of the FET switch Q206? My guess is that it's used to reduce the leakage of those transistors when performing an auto-calibration, with K201 disconnecting the input. The leakage could originate from the PCB round the unguarded traces to the diodes and relay, which they perhaps wouldn't want to guard because, being normally grounded, could couple ground noise into the signal lines or perhaps excessive leakage into the guard trace.

Not very convinced though, and if so, why not simply have the protection transistors on the other side of the relay, or use the other relay contact to apply the protection when needed? The current input is earthed by K201 when it is disconnected so the transistors wouldn't dissipate any power when an overload is applied when K201 is off.

Presumably the protection transistors and high current shunts are protected from excessive dissipation by disconnecting K201 in the event of an overload, but how is it done quickly if the meter is in a low current range? I guess that when it sees an over-range current it would set the DC amplifier to x1 and make a quick measurement to detect an excessive voltage which could damage one or more current shunts or see the 6 to 8V [EDIT: wrong, 2V max - see follow on post] or so that would indicate a protection transistor + diode conducting. This wouldn't be particularly quick though if a 10NPLC, 200ms (the longest integration time) measurement had just started.

In the meantime the protection transistor could be carrying 3A (1A fast acting fuse) [EDIT: 3A for the 200ms before the fuse blows] dissipating 15 to 21W. [EDIT: 3 to 4W] - I don't suppose that's likely to be a problem thermally for a TO220 device but what about the leakage through that device subsequently, or if it took a hit with a much larger current, say 50A for 1ms? Reverse base-emitter breakdown can damage a transistor so perhaps these were selected because of their ability to handle this scenario. I suppose the meter is designed to remain operational, within spec, after any type of current overload provided the correct fuse is fitted - after all they didn't seem to see the need to add a H/W overload cct to reset the relay ASAP as they did for over voltage protection.

[Cerebus]

Also, what might be the purpose of the purpose of the FET switch Q206? My guess is that it's used to reduce the leakage of those transistors when performing an auto-calibration, with K201 disconnecting the input. The leakage could originate from the PCB round the unguarded traces to the diodes and relay, which they perhaps wouldn't want to guard because, being normally grounded, could couple ground noise into the signal lines or perhaps excessive leakage into the guard trace.

The signal 'boot' at the source of Q206 is a low impedance copy of whatever the input signal is (in this case the 'HI' end of the shunt voltage) and is used to bootstrap the protection 'diodes' Q207 and Q208 to reduce the leakage across them by making sure they have the same voltage at both sides. Q206 can be used to apply or remove the bootstrap voltage under control of the 'HBST' signal (you can guess that the BST in that signal name stands for 'bootstrap').

K201 can isolate the input and also lifts the ground from the protection 'bridge' at the same time. This looks like the way you'd want things during autocal and with the 'bridge' floating it would make sense to turn off the bootstrap drive to it to reduce leakage of the bootstrap current.

My guess is that K201 is quite delicate and K200 comparatively meaty and that a little dance is done between them so that K201 never has to make or break the input without K200 being closed while this goes on. I suspect that neither K200 or K201 are used as protection, there's a fuse upstream of them and that, the spark gap and the 'diode' 'bridge' are enough to do the protection task way more effectively than any relay ever will. I think the relays are purely for isolation to keep any outside influences (such as a lead left plugged in) from affecting calibration cycles.

[splin]

Also, what might be the purpose of the purpose of the FET switch Q206? My guess is that it's used to reduce the leakage of those transistors when performing an auto-calibration, with K201 disconnecting the input. The leakage could originate from the PCB round the unguarded traces to the diodes and relay, which they perhaps wouldn't want to guard because, being normally grounded, could couple ground noise into the signal lines or perhaps excessive leakage into the guard trace.

The signal 'boot' at the source of Q206 is a low impedance copy of whatever the input signal is (in this case the 'HI' end of the shunt voltage) and is used to bootstrap the protection 'diodes' Q207 and Q208 to reduce the leakage across them by making sure they have the same voltage at both sides. Q206 can be used to apply or remove the bootstrap voltage under control of the 'HBST' signal (you can guess that the BST in that signal name stands for 'bootstrap').

I made a mistake when I was trying to work out this circuit thinking that diodes CR201/would clamp the collector of Q207 to +/-.6V preventing them being bootstrapped when measuring current - overlooking that the maximum signal and bootstrap are +/-100mV in this mode. But the question remains as to why would you ever want to remove the bootstrap using q206? [Edit: could it be to allow the protection transistors to be checked as part of the self-test? Except that would require K201 to be operated and thus the input jack would have to be open circuit to allow a voltage to be applied via the acal signal]

K201 can isolate the input and also lifts the ground from the protection 'bridge' at the same time. This looks like the way you'd want things during autocal and with the 'bridge' floating it would make sense to turn off the bootstrap drive to it to reduce leakage of the bootstrap current.

My guess is that K201 is quite delicate and K200 comparatively meaty and that a little dance is done between them so that K201 never has to make or break the input without K200 being closed while this goes on. I suspect that neither K200 or K201 are used as protection, there's a fuse upstream of them and that, the spark gap and the 'diode' 'bridge' are enough to do the protection task way more effectively than any relay ever will. I think the relays are purely for isolation to keep any outside influences (such as a lead left plugged in) from affecting calibration cycles.

Good points. K201 is a Panasonic DS2E-SL2 with 2A/30V contact rating and K200 is an even smaller custom Panasonic reed relay, probably with even lower contact ratings. So those relays probably aren't used for protection.

Doh! More importantly I've just realized why most of my question was rubbish - for some odd reason I had it in mind that Q207/208 were NPN and was using the reverse base-emitter breakdown as an ultra-low leakage protection mechanism - except that they are quite obviously PNPs (as I actually pasted into my previous post)! 

Thus the protection will clamp the input to 2 diode drops - say 2V maximum rather than the 6V to 7V if they had been NPNs.

Anyway the type of transistor is still interesting; the bootstrap/signal offset is calibrated out with a DAC so I'm guessing the maximum base-emitter voltage on Q207/208 would be a few uVs - perhaps a little more given how far the guard travels on the board, especially if the board isn't particularly clean and the total leakage into the guard increases significantly. So how much 'leakage' current might a standard PNP power transistor have with a few microvolts of base voltage and 100mV collector-emitter, at 55C ambient (say 60C internally)? The transistors are Nat-semi parts. Could they be specially designed for this type of application?

Even with a 2V clamp, the 9ohm, 10mA shunt R211 would be dissipating nearly three times its 125mW rating so the software will have to detect an overload and disconnect K202. As before, I believe this could take slightly more than 200ms worst case, but that's not likely to be long enough to cause damage, and any shift in value due to the shunt's overload would be calibrated out at the next autocal. Similarly the 1ohm 100mA shunt, rated at 3W could be dissipating 4W with a 2V clamp but the fuse would blow within 150ms at 4A - so all is well after all. 

[Kleinstein]

The clamp voltage should be less than 2 V, more like 1.5 V. So the power ratings for the shunts should be OK, though just barely.

At a voltage in the low mV range, the forward current of a transistor or diode can be really low. Further up in reply #79 bktemp did a measurement for a big MOSFET (which also has a diode inside). There the current was below about 200 pA even at 50 mV.  So the leakage in the transistors should be of similar size, possibly even smaller. I don't think the transistors need to be that special.

For dave's proposed circuit, bootstrapping the protection diodes is not an option as, as there is just one diode drop. Even this can cause a power problem with the 11 A fuse and the small case. For possibly critical low current ranges (e.g. less than 50 µA) there would be the option to use an transimpedance amplifier and this way keep the voltage even lower than the currently planed 50 mV, and this way reduce leakage a little.

[Cerebus]

Anyway the type of transistor is still interesting; the bootstrap/signal offset is calibrated out with a DAC so I'm guessing the maximum base-emitter voltage on Q207/208 would be a few uVs - perhaps a little more given how far the guard travels on the board, especially if the board isn't particularly clean and the total leakage into the guard increases significantly. So how much 'leakage' current might a standard PNP power transistor have with a few microvolts of base voltage and 100mV collector-emitter, at 55C ambient (say 60C internally)? The transistors are Nat-semi parts. Could they be specially designed for this type of application?

You seem to be missing that Q207/Q208 have the base-collector leads shorted, i.e. they are 'diode wired' with the base-emitter junctions being used purely as low reverse leakage diodes with lousy reverse breakdown voltages. My go-to transistor for that usage is a selected 2N3904/MMBF3904 but under these circumstances it'd be too small for the currents under consideration. Note that the BOM calls out a Pd of under 2W yet a TO-220 case - that suggests a low-power power transistor. A non obsolete part that might fit the bill (depending on leakage) might be a BD139/140 or a ZTX851/951 (the latter 5A continuous/20A peak rated but in an e-line case or there's a SOT-223 part available).

The offset between the main signal and bootstrap signal is going to be on the close order of 100uV. The offset correction DAC is 8 bit and (from memory) has an effective adjustment range of +/-30mV giving one LSB = 234 uV.

[phunkz]

Hi
i'm again the guy who asked the stupid question why not to use 2 fuses in parallel.

Here I do have a  late 70's Hartmann & Braun Multavi3, 30 Amps, analogue Multimeter with an automatic fuse.
(I don't want to send it in for a teardown, i like it very much, therefore some additional information:
http://www.radiomuseum.org/r/hartmann_multavi_3.html :-)

But what about an automatic fuse? Regardless of price and size, because you are planning an absolut high-end-multimeter...
As far as i know they do have 2 paths using slow bimetal and fast magnetic switching, so they should met any specs in terms of speed or current...

Sorry, just another stupid question...

[Kleinstein]

The difficult part is getting the high DC voltage rating - normal circuit breakers are not good at this. So it would be a really special automatic fuse.

The circuit is still for a relatively small battery powered meter - so some compromises need to be made to make is small and power power.

For the proposed circuit the burden voltage for the 10 A range would be more than 100 mV, as this range would also have to got through the MOSFETs - so maybe another 20-50 mV. So the combined power dissipation might become an issue. Using a separate plug like before would be unsafe with just one fuse. One less plug would also free up some space.

[mqsaharan]

Hi everybody.
First of all, a big thanks to Dave for his efforts on this video blog. I have always found his blog very informative.
Secondly, after watching this video, I started searching for commercially available DMMs with low burden voltages and found that Gossen is using such technique to lower the burden voltages but they are prohibitively expensive for hobby work. So, I kept searching and found another company that is using the same idea, Rishabh. I do not know how good their DMMs are. I have attached the specifications for their cheaper meters. On paper at least, they have low burden voltages but cannot say how do they behave in practice.

Qasim.

[Kleinstein]

My guess is the Rishabh meter is ignoring the voltage drop at the fuse. The values look to low just for an HRC fuse alone.  The 66 mV burden value is otherwise suspicious for a 6600 count meter. So this is just the voltage drop on the shunt.
The low resolution also helps in having a low burden, as the needed voltage resolution is not that high. There is also a very limited number of ranges and at low current the mechanical switch can be use.

[Chupacabras]

It might be a stupid question, but:
Is it possible to omit fuse in design? Just use mosfets (possibly several of them), and switch them off immediately when higher current is detected?
Would it work or be dangerous?

[bktemp]

It might be a stupid question, but:
Is it possible to omit fuse in design? Just use mosfets (possibly several of them), and switch them off immediately when higher current is detected?
Would it work or be dangerous?

Try finding a 1kV mosfet (or even more, look up the voltage levels for CAT ratings) with less resistance than a fuse.
So no, it won't work, and yes it would be dangerous, because mosfets typically fail short circuit if you overload them (too high current or too high voltage).

[analogNewbie]

Since the body diode is always there, there is forward current through the protection MOSFET body diode, why does not it matter in the uA range? Because the burden voltage is only 50mV?

[David Hess]

Since the body diode is always there, there is forward current through the protection MOSFET body diode, why does not it matter in the uA range? Because the burden voltage is only 50mV?

If you are talking about in the HP3458A schematic, those are JFETs.

Some designs use lateral MOSFETs with a separate substrate connection so the body diode does not conduct.

[Cerebus]

Since the body diode is always there, there is forward current through the protection MOSFET body diode, why does not it matter in the uA range? Because the burden voltage is only 50mV?

If you are talking about in the HP3458A schematic, those are JFETs.

Some designs use lateral MOSFETs with a separate substrate connection so the body diode does not conduct.

No, I think the question is about Dave's proposal, where 2 back to back MOSFETs body diodes are used as replacements for the more common bridge rectifier as a protection element in parallel with the current shunt.

[bktemp]

Since the body diode is always there, there is forward current through the protection MOSFET body diode, why does not it matter in the uA range? Because the burden voltage is only 50mV?

I did some measurements here:
https://www.eevblog.com/forum/blog/eevblog-929-designing-a-better-multimeter/msg1042459/#msg1042459
At 50mV there is a small current flowing through the diode, but it is low enough even for uA range. At 100mV or 150mV the current can be a problem for uA range at elevated temperatures.

[kvresto]

Hi Everyone.
In the pic I've uploaded Dave has drawn up a circuit to illustrate his idea. I’m not too clear on how this circuit deals with common high mode voltages? I hope someone can clear it up for me.

I assume the DMM will be battery powered, so the mux and opamps will run off this voltage, but if I insert this in a system that runs say 20V, would I have to select components that have a high common mode voltage?

I see GND is marked, but this is not really GND, after all its in series with the system being measured.

EDIT: what if this cct was powered by the system under test?

[Kleinstein]

The DMM has no other external connection. So the circuit ground will just float with a external voltage. So the circuit does not have to deal a lot on that, except for shielding so that higher frequency external field will not disturb the readings like it does with some of the newer Aligent meters.

[drussell]

Secondly, after watching this video, I started searching for commercially available DMMs with low burden voltages and found that Gossen is using such technique to lower the burden voltages but they are prohibitively expensive for hobby work. So, I kept searching and found another company that is using the same idea, Rishabh. I do not know how good their DMMs are. I have attached the specifications for their cheaper meters. On paper at least, they have low burden voltages but cannot say how do they behave in practice.

You could always just buy one of Dave's uCurrent units...  They are specifically designed to be a better current measuring device than most multimeters.

P.S.: Welcome to the forum! 

[kvresto]

Absolutely sensational Dave!!    I don't actually work in the industry and rely on Blogs such as this one for information, and I always find it here. A couple of well placed dumb questions always seems to clear things up. Cant wait for what's to come. Thanking all those answering my questions.

[Yansi]

How does the Gossen meter (and possibly others) get over the fact, that when the meter is powered off and current applied accross the current terminals, the mosfet won't overheat, due to excessive current being pushed through their body diodes?

10A through the body diode of the protection mosfet, is something like 10W of heat. How do they get away with that? Haven't seen any heatsinking in there.

[f4eru]

3 possible solutions to that:

The easy solution : With a multimeter in off state you could still put voltage to the gate to hold the mosfet shorted.

The complex solution : a LV oscillator-step up ("joule thief" like) to get charge into the gate, from the diode voltage.

The expensive solution : use a depletion mode mosfet, which is saturated @ Vgs=0V

[Yansi]

I think the second two won't happen, as we know the Gossen uses IRL3302, not depletion mode mosfets. And joulethief-like converter is most likely just an imagination. Probably only the battery voltage on the gate in off state.

But does the user manual state then the meter should never get amps going through when the battery goes flat, otherwise it's toast?