Gas discharge tube with varistor in protection circuit. Why?

[Cerebus]

I keep seeing input protection circuits for high impedance inputs that look something like this:


The whole thing is typically followed by the usual series current limiting resistor(s) and some diode clamps reverse biased by the rails or bootstraped by a guard voltage. Nothing special there.

The GDT is obviously necessary to keep the input port's impedance as high as possible and leakage currents as low as possible. What I can't see is the advantage of having both GDT and varistor. There's no mystery about what the capacitor and normal resistor are doing - with the varistor in circuit they are necessary to guarantee that the GDT will strike regardless of the current state of the varistor.

What I don't see is what the varistor is bringing to the party that the GDT doesn't already provide. The only advantage that I can see is that the combination would limit the short circuit current under some circumstances - that is, the combination would clamp at V_{clamp-varistor} + V_clamp-gdt rather than the (usually rather low ~15V) V_clamp-gdt on its own.

Is that it, or is there some other advantageous feature that I'm missing?

[Gyro]

The only advantage that I can see is that the combination would limit the short circuit current under some circumstances - that is, the combination would clamp at Vclamp-varistor + Vclamp-gdt rather than the (usually rather low ~15V) Vclamp-gdt on its own.

Have you just answered it yourself?  The ability to clear the overvoltage transient clamping without disconnecting the 1.2kV supply. The GDT is already in low voltage arc discharge from the transient. Without the VDR going high impedance again after the transient, it won't be able to terminate this discharge.

Edit: the cap and 1M resistor probably ensure that the GDT is able to fire in the first place.

Edit 1: It's not clear to me how, with the Cap and 1M in circuit, the GDT is reliably going to get back to it's non-conducting state though.

[Cerebus]

The only advantage that I can see is that the combination would limit the short circuit current under some circumstances - that is, the combination would clamp at Vclamp-varistor + Vclamp-gdt rather than the (usually rather low ~15V) Vclamp-gdt on its own.

Have you just answered it yourself?  The ability to clear the overvoltage transient clamping without disconnecting the 1.2kV supply. The GDT is already in low voltage arc discharge from the transient. Without the VDR going high impedance again after the transient, it won't be able to terminate this discharge.

As I said: "Is that it, or is there some other advantageous feature that I'm missing?"

Edit: the cap and 1M resistor probably ensure that the GDT is able to fire in the first place.

I already said that in the original post.

Edit 1: It's not clear to me how, with the Cap and 1M in circuit, the GDT is reliably going to get back to it's non-conducting state though.

GDTs have a minimum holding current, just pick a high enough resistance that this isn't supplied at non-fault voltages.

Edited to remove a greengrocer's apostrophe. How could I have? [Fx: hangs head in shame]

[Gyro]

As I said: "Is that it, or is there some other advantageous feature that I'm missing?"

The GDT is already in low voltage arc discharge from the transient. Without the VDR going high impedance again after the transient, it won't be able to terminate this discharge.

I don't think you mentioned the VDR being there to allow the GDT to stop conducting in your OP. That's its primary (Edit: only?) purpose.

[Cerebus]

As I said: "Is that it, or is there some other advantageous feature that I'm missing?"

The GDT is already in low voltage arc discharge from the transient. Without the VDR going high impedance again after the transient, it won't be able to terminate this discharge.

I don't think you mentioned the VDR being there to allow the GDT to stop conducting in your OP. That's its primary purpose.

Sorry, read a little too quickly and carelessly there.

[Gyro]

Something I suffer from far too often  .

Actually, thinking about it, it's the capacitor that helps with the termination of the discharge. It's only the resistor that is needed to allow the GDT to fire.... Once the transient passes, the VDR goes high resistance. At that point, the cap charges up to 1.2kV minus the GDT arc voltage. There isn't enough current to maintain the arc discharge and the cap remains charged for long enough to ensure that there isn't enough voltage across the GDT to re-establish a 'glow' or Townsend discharge (rather than minimum holding current).

[Cerebus]

The cap is going to help on the striking too. At the 10ⁿV/us slew rates that are typical of transients it's going to have something pretty close to zero impedance and keep the bottom of the GDT close to ground while it gets going. The type of parts that I've seen specified in this role are often specified in terms of 100V/us and 1kV/us slew rates for impulse breakdown.

I think the normal resistor's role is largely as a bleeder for the cap towards the end of a whole cycle.

[Gyro]

Yes, you're right. I was forgetting about transient rise times.

It was a little simplistic to try to isolate the R and C functions, they both play a role in both triggering and extinction.

[Cerebus]

So, tentative conclusion:

(Transient overvoltage only, a constant DC gross overvoltage would proceed somewhat differently.)

• GDT maintains low leakage during non-fault conditions
• GDT and capacitor together work to strike the GDT and start passing any transient to the varistor
• Varistor does its stuff and clamps the bottom of the GDT to some voltage that is slightly higher than the nominal working voltage of the circuit so that, once the transient passes, working DC bias on the circuit can't hold the GDT in spark (or glow) discharge.
• Incidentally, capacitor gets charged to the clamping voltage.
• Overvoltage condition goes away
• Varistor turns off, impedance to earth now too high to maintain GDT discharge
• Resistor bleeds charge off the capacitor once the GDT has stopped being a low impedance path
• Now reset, ready for next time.

The thing I wasn't getting before was that the varistor helps with turning the protection circuit back off in the presence of DC bias (a normal input) when a transient overvoltage condition has tripped the protection circuit. I'd failed to consider the effect of normal DC bias during a fault.

[Gyro]

Yes, that looks right to me (other than having 'the GDT not being able to hold / maintain discharge' in two separate bullets).

Incidentally, it's a shame that GDT packages don't have better leakage characteristics. I've found them to be quite leaky even at low voltages. The glass ones tend to be a little better than the ceramic envelope ones. Many of them have a radiation source in them too, causing ionisation leakage. I don't know if there are any special low leakage instrumention grade ones avaliable.

P.S. The Voltage rating of the capacitor in the schematic is too low - It's in danger of becoming a VDR in its own right. 

EDIT: I would simplify it as:

• GDT maintains low leakage during non-fault conditions
• GDT and capacitor together work to strike the GDT and start passing any transient to the varistor
• Varistor does its stuff and clamps the bottom of the GDT to some voltage that is slightly higher than the nominal working voltage of the circuit
• Overvoltage condition goes away
• Varistor turns off
• Capacitor charges through the GDT to near input voltage
• GDT voltage is now too low to maintain spark or glow discharge
• Resistor bleeds charge off the capacitor once the GDT has stopped being a low impedance path
• Now reset, ready for next time.

[SeanB]

Capacitor likely will survive a while, seeing as it rarely gets any voltage on it, and will be clamped to the VDR voltage peak in most cases. Capacitor will also likely be a ceramic device, and those are tested to survive 2 times rated voltage short term.

This circuit seems to give the biggest advantage of the GDT, low capacitance and high current capability, but also addressing the major drawback of hold current and voltage clamping the circuit after firing.

[Gyro]

Yes, it will probably survive, I suppose if the VDR gets that far above its clamp voltage, things are in melt-down anyway,

Good point on the low capacitance, I forgot that.

[Cerebus]

P.S. The Voltage rating of the capacitor in the schematic is too low - It's in danger of becoming a VDR in its own right. 

They're just 'typical' values representing the kind of circuit I kept seeing. Except, those specific values (and very specifically the voltage rating for the cap) are the same as on the input of an HP 34401A, except for the 1M resistor which is 2 x 1M in series.

[Cerebus]

On the subject of GDT leakage, the specs vary quite a bit between manufacturers and their ranges. In the kind of territory we've been talking about here you typically see a figure of either >1Gohm or >10Gohm on the datasheet. Neither is a great figure for adding protection to a precision instrument where it's not even slightly hard to arrange for the input stage to present 1Tohm to the input.

There are plenty of other ways to devise input protection that can provide a lot less leakage, but it's pretty hard to do that and come up with momentary fault handling for multi-kV, low-impedance fault sources like GDTs do, handling brief fault currents in the kA region. A 1000V CAT II rating calls for handling a 6000V transient with a source impedance of 12 ohms, if you're going to clamp that then you're dealing with a 500A fault current. Once you get into CAT III and above territory that source impedance falls to 2 ohms and the fault current rises to 3 to 4 kA.

[Gyro]

I must admit, this sort of input protection on a DMM worries me. Sure, I know that precision bench meters don't often have a Cat rating, but this sort of 'brute force' input clamping isn't exactly going to do the operator any favours in an accidental / gross stupidity situation. I know at least one older Solartron meter that just has a 1.2kV GDT soldered straight across the input terminals.

As you say there are other ways - I keep coming back to Datron and their series string of big carbon composition resistors, mounted on PTFE standoffs, adding up to 88k at 12W dissipation, before clamping with zener+JFET combinations. This was used on the 1040/1050/1060 (/1070/1080???) series. This gives 1kV continuous protection even on the 10mV range [Edit: and very low leakage]. Only for DC of course (is this the hang-up?).

I realize that, at some voltage, something is going to flash over to Guard and/or Chassis in a low impedance way, but it does seem a more 'public spirited' form of input protection.

[Cerebus]

When I've seen this particular style of input protection it's almost always followed by a resistor string and some diode clamping. As you say, the resistor string and semiconductor clamp is usually good for around 1 kV protection in and of itself.

If you want to increase that 1kV it all starts to break down (no pun intended)  when the requirement for your semiconductor clamps to dump current is in opposition with your desire for low leakage and low noise. For example, if you opt to clamp with a 2N4117 JFET wired as a diode you'll get good low leakage, but then you hit an absolute max. gate current of (dives for data sheet) 50 mA. Similar figures apply for the old trick of using a small signal transistor junction as a low leakage diode. Increase the value of your resistor string to get around that 50 mA limit and you start adding Johnson noise and raising the source impedance of your measurement which may in itself increase your effective input noise.

There's a neat trick using NMOS FETs as current limiters (App note here) in place of a resistor string that looks too good to be true. I'm sure that there's some drawback I'm missing and I'm waiting for the other shoe to drop. One thing that comes to mind is that you'd need to use a string of these to achieve higher voltage protection and there's the issue of balancing the voltage down the 'divider string' of MOSFETs in the face of a whacking great transient so that they don't pop one after the other. Sure, a capacitive divider is the obvious choice but will one with sufficiently low impedance to be effective also have sufficiently low impedance to bypass the protection. Some inductance in front of the whole lot would slow edges enough that you could work around this, but then you're in danger of inadvertently building a filter or resonant structure if you don't properly account for the (non-linear) capacitance of the FETs themselves. As you can tell, I haven't sat down and worked out the details for that one yet.

[Gyro]

Indeed, that's about the limit before you start cooking JFETs. I suspect even the 88k is starting to introduce noise at high resoslutions on the later instruments (I must check the implementation of the Datron 107x and 108x out of interest, the 1271 too).

Thanks for the App note it looks interesting, yes trying to cascade them could be interesting - and certainly fun testing  The non linear capacitances of the FET's would indeed make things interesting.

[T3sl4co1l]

Such circuits are typically required for clamping line-to-ground transient voltages.  The GDT keeps ground leakage low, even if the MOV goes leaky.

Tim

[Cerebus]

Such circuits are typically required for clamping line-to-ground transient voltages.  The GDT keeps ground leakage low, even if the MOV goes leaky.

Tim

For once Tim, we're way ahead of you. 

[splin]

There's a neat trick using NMOS FETs as current limiters (App note here) in place of a resistor string that looks too good to be true.

Bourns offer what appears to be an integrated version of this - though I can't see a proper description on their website of how they work. Max voltage is 850V and minimum triggering current is 50mA (which is rather high for this application): http://www.bourns.com/products/circuit-protection/tbu-high-speed-protectors.

[Kleinstein]

For current limiting there are depletion mode MOSFETs. With an isolated auxiliary supply (e.g. from an PV optocoupler) one can use normal hight voltage MOSFETs too. Keithley used this in some of there meters (e.g. 2000, 2001) - it can do switching too.  Using just a series resistor could introduce quite some resistor noise for a low noise input.

Another option for current limiting can be a PTC type resistor. It reduces the current and thus power to the resistor after some time. Just a resistor would need to be high power too.

There are low leakage diodes for a higher current too: The BAV199 is good for 160 mA and 500 mA peak.