shutting a tube off with a relay

[Gainzilla]

Thoughts on this?

HV_Mosfet_Switch_v1.0.png

[R.G.]

Thoughts on this?

Yes.
The MOSFETs are connected incorrectly. Both gates should be together, as shown, and both sources should connect in the middle. The two drains are the external contacts.
The 12V zener should be from the two sources in the middle to the two connected gates.

The idea is that the two body diodes meet in the middle as well, so that a higher-than-B+ voltage is not connected to B+ by a conducting body diode. The VOM1271 datasheet is a little anemic on that point. Also, the "B+" labeled point would be to one end of the OT, not to B+, and the "TubeHV" would go to the plate. The circuit sits in series with the tube plate. Turning the LED off means that the tube can no longer conduct at all.

Notice that this introduces a new class of problems where the LED is not properly driven on for normal operation; but any scheme that turns off a tube does that.

I really ought to go get out the simulator and mess with this a big. My in-brain simulator is nagging at me about something.
Edit: Never mind. I was forgetting that this was to be used to stop an already-damaged tube from conducting. It works for the intended case.

[Gainzilla]

Aaaah, ok, I think I see what you mean.

Thoughts on this version?

HV_Mosfet_Switch_v1.1.png

[Gainzilla]

Also, do you think the IRFBG20 need a heat-sink? They're rated at 1000v. Most amps I build are below 475v.

[trobbins]

One issue with using the 'disconnect the cathode' method is that the cathode may float to a high voltage level. That is not a concern if the valve is already damaged, but some faults may not be due to the valve itself, and so a 'good' valve could then be damaged by excessive heater to cathode voltage. That situation was managed for fusing or PTC type cathode protection by adding a parallel zener or resistor to the fuse or PTC, so as to keep Vhk within datasheet limits.

[nuke]

Also, do you think the IRFBG20 need a heat-sink? They're rated at 1000v. Most amps I build are below 475v.

What’s the Rds of the mosfet?

If it is low enough, the mosfet won’t generate much heat.

Figure watts as I^2 * Rds.

[R.G.]

Also, do you think the IRFBG20 need a heat-sink? They're rated at 1000v. Most amps I build are below 475v.

The voltage is not the big thing, the current and Rdson are. When the MOSFETs are turned off to stop tube current entirely they may have high voltage across then, but nearly zero current, so they don't dissipate much if any heat. The heating all comes when they are turned on to let the tube operate normally. In that case, they heat up by Ims-squared times Rds. Rdson is specified for these as 11 ohms max.

The rms current in the tube is trickier to figure out without just measuring lots of overdriven amps. Guitar amps have a history of their output stages being abused erratically. the 6L6 data sheet says that they can be pumped to about 250ma with Vgk = 0. Bounding the problem by assuming the tube is driven to quasi-saturation at 250ma with a square wave. The RMS value of that 0->250ma wave is 125ma. Some AB2 amps could be driven harder, so maybe as much as (... pulling a number out of the air...) 175ma.

So the plate current squared might be 0.175*0.175 = 0.030625. Multiplying by 11 ohms gets a power dissipation of 0.337 W. Not too bad. A TO220 can get rid of 2W with no heat sink tied to the tab if it's in free air.

But what if the tube has lost its bias voltage? The typical plate curves for a JJ 6L6 say it could go to 350ma constantly. That's 1.25W. Still not too bad. I would use a smallish PCB heat sink like this: https://www.mouser.com/datasheet/3/6118 ... b20-np.pdf

The really bad case is when a tube shorts, though. In that case, the tube no longer has anything to do with the current. The current is limited only by the power supply, which is still trying to apply 450 to 500V across the 22 ohms of the MOSFETs. The power supply voltage sags because of the resistance of the PT primary and secondary. Effectively, the PT winding resistances appear in series with the MOSFET Rds, and this is driven from about 500V. The rectifiers still rectify and the first filter cap still filters, so you're getting a DC level with a massive ripple voltage on it. This problem needs some careful simulation or several days of math, or a wild a$$ guess.

It will be less than (500*500)/(22+5)= 9.260 kW. Yikes!!! The time your circuits spend deciding to turn off a tube that has gone mad will be what determines if the MOSFET survives. The thermal data on your MOSFET indicates that it might survive if you can turn it off in a millisecond or less. The LED-PV gate driver needs 54uS of that, so a digital circuit would have to sense the failure and decide to turn off the MOSFETs in under about 900uS. This scenario is one that fairly screams for an analog solution, running at raw device speed instead of a digital one where the controller has to sense or get notified of the failure and then make the decision and turn off the FETs.
You can see why NASA and military stuff is so expensive.

[nuke]

This problem needs some careful simulation or several days of math, or a wild a$$ guess.

Or a handy, smoke-preventing, fast-acting fuse.

[Gainzilla]

I’m working on drawing up a fast-trip addition to the above circuit, but probably won’t have time to finish it until tomorrow.

[Gainzilla]

Here is the full design. I worked out the details with ChatGPT, so definitely look for errors (large or small), but I think it's pretty solid?

Normal operation (tube conducting):
• MCU drives the VOM1271 LED.
• VOM1271’s PV side makes ~8–9 V across Gate–Source, turning on the back-to-back MOSFET pair.
• The MOSFETs act like a near-short (a few hundred mΩ effective). Tube current flows normally from OT primary → MOSFETs → tube plate.

Overcurrent detection (fast analog path):
• Each tube’s 1 Ω cathode resistor makes a small voltage proportional to current (1 mV/mA).
• Comparator (+IN) sees this voltage; –IN sits at ~0.21 V Vref.
• If cathode current exceeds threshold (e.g., 210 mA for EL84), +IN > –IN.
• Comparator output flips HIGH → drives PC817 LED → turns on PC817 transistor.
• PC817 transistor shorts MOSFET Gate→Source, instantly yanking the MOSFETs OFF.
• Current stops in microseconds, before the MOSFETs’ SOA is exceeded.

Hysteresis:
• 1 MΩ from comparator OUT → +IN adds ~5–10 mV margin so noise doesn’t chatter the cutoff.

Protection elements:
• 12 V zener G–S: keeps MOSFET gates safe from surges.
• TVS (~600 V standoff): clamps drain–drain spikes.
• Snubber cap (10 nF, 1.6–2 kV): tames ringing so the TVS doesn’t eat everything.
• Dual MOSFETs back-to-back: cancels out body diodes so it blocks current either way.

Fallback:
• If MCU never reacts, the analog comparator + PC817 path is independent — it alone will shut off the tube on a fault.
• MCU can still log the event or retry later, but safety is handled in hardware.

⸻

So in short:
• MCU + VOM1271 = normal ON.
• Comparator + PC817 = emergency OFF.
• MOSFET pair + TVS/snubber + zener = the switch cell itself.

PowerTube_Switch_v1.2.png

[trobbins]

The prospective fault current has to exceed 210mA +/- your sensing tolerances.

If you disconnect the anode then (depending on circuit config) the screen may get stressed - not a concern if the tube has had a non-recoverable fault, but a concern if the tube is not the cause of a glitch.

[R.G.]

Looking better!

One issue you would probably run into is that the comparator won't stay latched. When it turns off the plate current, the cathode current drops and the comparator will un-latch if I'm reading the schemo correctly. The MOSFETs turn back on and the current goes up...

I solved this issue with the current sensors setting a CMOS latch. The latch stayed set until something re-set it, so its output was what turned off the MOSFETs (in my design, in the cathodes). This approach would hold things steady until the laggardly human could do something. I used the latched condition to drive indicator LEDs, etc. to give the humans a hint about what happened.

If I were to be revisiting this design in a world of microcontrollers and programming, I would be tempted to turn all the tubes off once I had captured the failing item(s) by the uC. This would cure the issues of collateral damage to good tubes that trobbins brings up. I can rationalize this to myself by thinking that once the failure condition has been captured, the rest of the amp is unlikely to run in any useful kind of way with one output tube turned off. If I remember correctly, I think I shut down all the tubes' cathode MOSFETs after latching. I'd have to go dig out the design to be sure. This approach might save you some MOSFETs and money, as you could penny-pinch by only shutting down the main B+.

Something that was not possible back in the early 2000s was using a Hall-effect current sensor chip to sense screen current. These exist now, and might offer a way to monitor isolated screen current to catch screen runaway separate from plate runaway.

[trobbins]

ACS712 based pcb seems small and with enough resolution.

[R.G.]

Yes, that's the one I was thinking of.
It's cautionary, though. The datasheet withstand voltage is about 500, and the screens could well be riding right at that DC level. Probably OK mostly, but I hate to se datasheet maximums with no margin. It's safety rated at up to 5000V for its UL/CSA/ETC ratings, but I could not reconcile that with the 500V rating between the sensed current pins and the logic side.

Edit: oops. I was thinking of another ACS chip that has an analog output. The ACS712 has a working voltage spec of 420V peak AC or AC + DC. So the 712 is a bit less.

[Gainzilla]

Great discussion! This was beginning to feel a bit over engineered/costly. I do kind of feel like must putting the amp into standby by switching off the B+ is a very simple solution. I guess the previous rules necessitating the fast analog path still apply, though, right? So that would mean the MCU could detect a fault, turn on a single opto which would trigger a dual-mosfet setup to switch off the B+. If the analog path gets tripped, it would also trigger the mosfets. In both cases I guess the MCU would need to remain on and hold the mosfets in place until it detects a happy tube situation. Am I missing anything? That seems pretty simple if so.