diagnose mortality in LT3081 adjustable current source

[RustyShackleford]

I have built an adjustable current source based on the LT3081, as shown.    The load is a series-connected string of LED emitters connected via several hundred feet of wire in an outdoor environment (wet soil etc).

I am experiencing a number of failures of the LT3081 device and I can't understand why.   Specifically, it ceases to carry current.   The SET pin is at a reasonable voltage (e.g. 100mv).   With a diode load (on my test bench instead of outdoors) the IN pin is about a diode drop below the supply voltage (30v as supplied by the pfet "header"), 1ma of current flows, and the OUT pin is at 1mv.  With no load, IN and OUT are near ground.

At least some failures have occurred while I'm fiddling with the LED string, while the circuit is activated.  So I might be quickly transitioning between presenting a short-circuit load and an open-circuit load, and anything in between.   Maybe I'm hoping for too much, but it doesn't seem like this should be frying the device.   With a short-circuit load, the thing should regulate properly (and pull 100ma with the SET pin at 100mv).   With an open circuit, the device should pull IN close to ground vainly trying to pull 100ma to match the SET and OUT voltages.    If somehow the SET pin were to go too high (and call for too high a current), the current-limiting resistor (on the Ilim pin) should limit the current to well under 1 amp and failing that the device's internal current limit is something like 1.5amp.   The device is attached to a 6.4-degreeC/watt heatsink.

A penny for your thoughts !

[dom0]

Can't help beyond saying that I had some LT3080s die and didn't understand why, either.

[jhpadjustable]

I don't see several hundred feet of wire in the model. To start, add 400nH/ft on the LED+ and LED- wires. If you have data from the manufacturer on mutual capacitance of the cable, add that into the sim too and see if everything still looks alright. Or approximate it with 1nF-10nF across LED+ and LED- on the driver side of those inductors.

If you add a simple freewheel diode across the LED output on the board itself or permanently affixed to the near end of the cable, I think you will have solved the failures. Or maybe just don't play with the load wires while the power is on. 

[SiliconWizard]

connected via several hundred feet of wire

That starts smelling like the source of your problems. If you plug/unplug this while the circuit is powered, it is likely to experience some amount of kickback due to the parasitic inductance of the cable.

As jhpadjustable said, a flyback diode could be in order here.
I would also add some serious protection against ESD while I'm at it.

[RustyShackleford]

I don't see several hundred feet of wire in the model. To start, add 400nH/ft on the LED+ and LED- wires. If you have data from the manufacturer on mutual capacitance of the cable, add that into the sim too and see if everything still looks alright. Or approximate it with 1nF-10nF across LED+ and LED- on the driver side of those inductors.

Good point.  The cable is just 18awg landscape-lighting cable, which looks a lot like zip cord.   But yeah, and then make and break the connection instantaneously (in my sim).

If you add a simple freewheel diode across the LED output on the board itself or permanently affixed to the near end of the cable, I think you will have solved the failures. Or maybe just don't play with the load wires while the power is on. 

Yeah, except the absolute max ratings of the LT3081 allow IN to be +/-40v wrt. OUT.   That's a lot of flyback.

As prudent as it may be, it'd actually be kind of inconvenient to turn the power on and off every time I mess with the wires and LEDs.

[RustyShackleford]

I would also add some serious protection against ESD while I'm at it.

Yeah, I thought about that.   

[SiliconWizard]

Yeah, except the absolute max ratings of the LT3081 allow IN to be +/-40v wrt. OUT.   That's a lot of flyback.

Not really. Your power supply is 30V. Think about it again... that's only +10V above it.  You can consider its OUT pin to be almost at 0V.

Attached is a simplified model. You can see for yourself what can happen with and without the diode at the IN input of the regulator.
I modeled the cable with an inductor (with some ESR and capacitance too), and the plugging/unplugging with a switch.

In your circuit, the diode would have to be put between the LED+ and LED- points on your board (and not in the cable, even less so on the LEDs' side!)

[RustyShackleford]

[Attached is a simplified model. You can see for yourself what can happen with and without the diode at the IN input of the regulator.

Thanks for that.   The simulation slows WAY down at 6ms or so, but I get the point.   I'll add a fast diode, and sure, the board is where I'd want to put it.   (Probably doesn't even need to be fast, since I can't possibly be connecting/disconnecting with such rapidity).

Dunno if I'll test it right away, those 3081's are kinda pricey (this isn't a high-dollar operation :-) ) but the diode sure can't hurt.

As far as ESD, I guess I'll just use clamping diodes (to the rails) on LED+ and LED-.   Probably not gonna help with a lightning strike though (kinda an issue with hundreds of feet of barely buried wire).

[RustyShackleford]

If you add a simple freewheel diode across the LED output on the board itself or permanently affixed to the near end of the cable, I think you will have solved the failures.

You are correct, sir.   I added 1N4148 between LED+/LED- near the LT3081, and I have had no failures since (I see it's been about a year).   This is on 3 different instances of the circuit, and some non-zero number of hot connect/disconnects. Thanks all.

[David Hess]

Rcmp seems way too high which impacts the ability of Ccmp to dampen the load inductance.

[RustyShackleford]

Rcmp seems way too high which impacts the ability of Ccmp to dampen the load inductance.

I am going from the LT3081 data sheet, page 15: https://www.analog.com/media/en/technical-documentation/data-sheets/3081fc.pdf
... and using Rcmp=100 and Ccmp=1u (film).  It seems to be working well, but I should do more analysis of what values are best.

[David Hess]

I am going from the LT3081 data sheet, page 15: https://www.analog.com/media/en/technical-documentation/data-sheets/3081fc.pdf
... and using Rcmp=100 and Ccmp=1u (film).  It seems to be working well, but I should do more analysis of what values are best.

The series RC network is a suitable compromise in applications where the impedance of the current source must be maintained at higher frequencies.  In cases like yours where the AC output impedance is irrelevant, the normal input and output decoupling configurations with solid tantalum or aluminum electrolytic capacitors are more suitable, meaning that the series resistance would be a couple ohms at most.  If the series resistance is too low, like with ceramic capacitors, then inductance in long leads to the load or source can cause problems, which is hardly unique to the LT3081.

So the application example with only the capacitor and no added series resistor is more suited to your application, unless the capacitor is ceramic and then a low value series resistor, just a couple ohms, should be added.