Dual rail voltage supervisor for op amp supply

[lk.dgironi]

Hello,

for a project I'm building I want to add a voltage supervisor for the +-15V rails of the op amp.
The main supply is 24V, then a DC to DC converter generate the +-15V. GND of the DC to DC is in common with the main +24V.
The +-15V feed a few op amps that output some voltage.
The purpose of this circuit is to power off the op amp section if one or both the +-15V fails.
My idea is this:
If 15V fails, shut down the -15V feed to the op amps.
If -15V fails, shut down the 15V feed to op amps.

Find my circuit below.
Notes:

• voltage divider resistors (the 5k one) are 0.1%.
• +-15intV came out from the DC to DC
• +-15V foes to the op amps (one reported as sample on the image)

Questions:
I've a few doubts especially on the negative sense (bottom left of the image) and positive feed (top right of the image) side.
Do you think it could work?

Thanks

[jwet]

I think you'll get into a race condition and it will kind of oscillate depending on who comes up first.  Breadboard it.  These simple little things can gets so complicated....

I would do the following.  Power it from +24 up stream of the DC's to DC's.  Use a dual HV comparator like an LM393.  Have plus and minus 15 pull one opposite ends of a divider.  If they're both just right, have this be say +5v.  Use the 393 in a windowed and kill both supplies on either failing.  Could just disable the DC to DC making +-15V from 24.  Make sure it latches or has lot of hysterisis.

This built in test stuff was very common in Mil Spec equipment that I worked on in the 80's.  Mil-23200 for Nuclear Instrumentation went to an a crazy level.  Overall, it think it created more trouble than it was worth. Mil-16400 required just pilots on a card edge for supplies- this was really much better.  If you wanted to get fancy, you could put zeners in series to make sure you had enough volts.  You might use an 11v zener for example for a 15v rail.

Good luck.

[lk.dgironi]

Thank you!
I'll check it and post updates here.

[magic]

Are you sure you can't design the circuit to "fail gracefully" if one supply is missing?
What sort of disaster are you trying to prevent, can the protection circuit react fast enough to prevent it?

[lk.dgironi]

Thank you.

CAT809 to my understanding should be 20uS delay, then optocoupler is in the order of 10uS.
The op amps drive a electro mechanical device, with a max freq of 1kHz, so 1ms. The circuit is fast enough.
It is enough to let the op amps connected to the +-15V line to shut down and output a 0V value.

[David Hess]

Tektronix often included something like what you describe in their test equipment.

The common way they used was to tie all of the power supply outputs together using a set of resistors to make a "balance" signal at ground potential.  The balance signal was tested using a window comparator so if any one supply failed or went significantly out of regulation, then after a delay the power supply was shut off and reset.

Another design they sometimes used had a pair of resistive dividers between the positive and negative supplies to produce about plus or minus 2 volts.  Each divider then drove either a PNP or NPN transistor so if one of the supplies was low, it would trigger the transistor which would then clamp nodes to ground through diodes stopping unintended circuit operation until the power supply voltages were correct.

[lk.dgironi]

Thank you @David Hess for sharing this information.

The "set of resistors to make a "balance" signal at ground potential" was something I was thinking too.
Now I've design the board like the circuit you find above.
I've simulate this on Spice and seems to work. Within the end of this week I'm going to make it assembly and then I'll update you here.

[lk.dgironi]

I've some problem due to the push-pull stage of the supervisor (series 809)
It does not works with 10k-5k resistor partitor, I've to go down to 200r-100r partitor to make it works.
Also the partitor resistor get warm due to the push-pull stage output VCC, that comes from the partitor.
Indeed I've shifted to an open-drain design.
I take the opportunity to simplify the output stage, just using one photocoupler for both the lines, and one supervisor (series 803, active-low, open-drain, threshold 4V).
The new schema is below, I've some doubt on the supervising part (the one on the left of the photocoupler), but it should work. 2k2 will makes the led consume 13mA almost, supervisor max current is 20mA, so it should work. Don't know if a pull-down is needed or not.
Side note: I'm not able to simulate this nor in hardware (I don't have any 803 supervisor here), nor in Spice (don't have a working one).
Side note: C22 is a filtering cap, should makes the job but again is something I'm not sure about.

Any suggestion?

[coppercone2]

yeah, order the part and test it. power supply protection problems is the last thing you want showing up on your PCB. that is like foundation cracking

its a good chip, my version of this had much more parts using simple transistor and required something else to make a decision about shutting it down.

[lk.dgironi]

Thanks!

Thinking I maybe going to have Drain to Source max voltage problem, but I can not find this stats in the datasheet

[lk.dgironi]

Just for completeness, find attached the working solution.
It uses TPS3847.
Complete discussion here: https://e2e.ti.com/support/power-management-group/power-management/f/power-management-forum/1299945/tlv803-voltage-supervisor-on-dual-rail-with-tlv803-and-opto-coupler/

[magic]

Okay, but this means that if the positive rail decreases to 10V or less while the negative is solid -15V, the good rail will be cut off but the chips will still be fed 10V positive.

With LM358, loss of negative rail only can cause very weird outputs to appear due to phase reversal...

I would simply turn off both switches when the total supply is less than 25V. Or even 28V, isn't that good enough?

[lk.dgironi]

If the positive rail decreases to 10V (15Vint is 10V), Q1 is off, the negative -15V is cut off.
If the negative rail is -15V (-15Vint is 15V), Q2 is on, the positive +15V is on, but it's 10V (cause 15Vint is 10V).
The op amp VCC+ still get 10V, but the VCC- is not connected. Isn't like this?

The problem is that "easily available" supervisor usually has max voltage up to 6V, the TPS3847 has a supply range up to 18V and I can find it.
Monitoring the full 30V (-15V to 15V) like was a thing, indeed to use this as a single supervisor for 28V monitoring I've design a circuit like the one attached, the problem is that at TI forum they suggest me not to use this cause the TPS3847 in their simulation the RESET pin gets more than 20V for an instant ("In your schematic A, there are instances when the RESET pin will see more than 20V, which is above spec for the device"). I also prefer something like "A" that cuts both -15V and 15V lines. Honestly I don't get the instant when RESET gets 20V, cause the voltage divider should act "instantly"... but I'm not so expert. What's your tought?

[lk.dgironi]

Note: in order to make the A circuit to work the voltage divider must be 1k and 680ohm, otherwise the Reset pin will be High but it will not be able to request current. But if 1k and 680 are used (also I've try 2k2 and 1k5 but that's too much) they get hot (70C in 1 min by thermal camera).

[magic]

This is because the chip takes power for the RESET pin from its voltage measurement pin, which is a little silly particularly as it comes in a package with unconnected pins left

The common as dirt TL431 has these functions separated (and handles 30V without drama). Connect a high resistance divider to REF, connect a current limiting resistor and optocoupler to C. In fact, the optocoupler could even be replaced with a PNP but you must include a fairly high base current limiting resistor.

TL431 has no bulit-in hysteresis, but that could be provided by adding a high value resistor from REF to N-MOS gate.

BTW, this simple protection scheme still doesn't work correctly if one rail has much higher than nominal voltage, not sure if you consider this a likely problem to ever occur.

[lk.dgironi]

Thanks. That's sounds interesting. Do you mean the "Figure 17. Shunt Regulator" configuration from TI datasheet.

(I'll be off for vacation next week, as soon as I'll be back I'll give it a try)

[magic]

Look for "voltage monitor" in "system examples".
The left TL431 and its resistors can be omitted because there is no high limit.

R4 should be near 1k to let TL431 sink at least 1mA current required for operation without turning on the LED.
R3 limits maximum current through the LED.

[lk.dgironi]

Thanks for the tip!
I think I'm going to check the TL431 to replace the TPS3847, seems simpler.
Find attached the LTSpice model simulation circuit for the TL431, 27V almost threshold.

Also, I'm using a 5V ref (REF195), which I'm thinking to replace with the attached version of TL431. The reference just goes in a -10V..10V to 0..10V op amp based converter (find attached schematic). Thinking about his the 5V there can be even not that accurate, cause later on this schematic I've a trimming pot that can balance this reference error. Do you think the attached TL431 5V reference work? Or is it better ot use an AMS1117-5.0?

[lk.dgironi]

Thanks for help. Circuit tested. Works at almost 27V threshold.

The only thing I've to note is the limit led resistor value.
Now it it's 4k7, otherwise it get's hot.
With 4k7 it get 6mA, considering the voltage diff of almost 27.5V between this resistor it's 6mA*27V = 162mW. I'm using a 1206SMD 250mW.