Power Supply Design - Input Protection Scheme

[Dan]

ignore the above, it has been clarified that we need the engine to keep running with the fault occurring, so shutting down the 5v switched rail is not acceptable.

other ideas such as using the TL-431 to switch a FET or Fred's original plan.

these need to be tested! and will be tested by me in the next day or so!

[Dan]

OK, more ideas to test (PDF also attached)



Decided to use a transistor instead of a FET.

Please comment, argue, bitch, moan, advise, etc

[TonyS]

Perhaps rather than clamping inputs to one of the 5V rails, you could clamp to a separate V_CLAMP net. This would prevent the issues that come up if current is sourced into the 5V rail.

If your requirements are that you should be able to handle 1 input shorted to battery, then all you would probably need is a single TL-431 and a couple of resistors on the V_CLAMP net. If your requirements are that multiple inputs being shorted at the same time be tolerable, then you would need to add a suitable PNP transistor to the circuit (see High Current Shunt Regulator application in the data sheet).

Please note that the TL-431 is a fairly precise shunt regulator and not a zener. You should be able to "dial-in" whatever clamp voltage you desire with fairly good precision and stability.

Thanks,
Huff

[Dan]

then you would need to add a suitable PNP transistor to the circuit (

see above image Huff.

I don't like the separate net idea though

I shall test out some stuff and provide some results/feedback.

Thanks for posting! Glad people are reading/thinking/helping!

[TonyS]

Hi Dan,

I don't like the separate net idea though

What issues do you see with having a separate net?

I realize that running a separate trace around for a V_CLAMP net would present a challenge, but I really feel that it is a "clean" solution to the issue.
It would also prevent the (admittedly unlikely) issue of a shorted input attempting to power up the board when the 5V regulator is supposed to be off.

Thanks,
Huff

[Fred]

Re the separate net, that's not necessarily a bad idea, all but the temperature pins don't involve power in the first place, so it would be a special run for them anyway. My concern would be that the non-5V net could leak current through the diode and bias the readings. I guess that could be alleviated with a resistor to the 5V rail that held it high in absence of a load (absence SHOULD be guaranteed, right?). Then the 5V rail could get by with nothing except its normal load provided the resistor matched that. HOWEVER that normal load is zero in some cases anyway, which is why I saw the very obvious skew in the first place. Thanks for posting :-)

[Dan]

A separate NET would be bad for a few reasons. It unnecessarily takes up more PCB real estate. It would make the PCB more complex in terms of current capability, noise, heat. Adds extra complexity to the already complex routing process ( if 2 layers are to be used instead of 4). I simply don't like the idea of introducing a new NET with associated passives to overcome an issue that could be rectified in a more technical way.

Thanks for your feedback, I wish others would post about these things

[Fred]

I guess the way I'm looking at it is that there is actually no need to connect the diodes to the 5V rail. The idea is to clamp the end of the diode to max 5V, which is something that the rail doesn't naturally do anyway. We still need to clamp the rail somehow for the poly fuses to be worthwhile, but that would be hugely simplified as the precision thing is not relevant anymore at that point. I think it should stay an option, even if it's a difficult one to make reality.

[Dan]

We can agree to disagree then I still don't like the idea at all. But nevermind it for now. I am keen to test out the circuits to see how they perform.

[Fred]

We don't even need to do that, we're formulating a list of possibilities to solve a problem in the best possible way. Once that list is fully populated and proven, then items from it that make the grade can be selected for specific designs, such as yours ;-)