2
\$\begingroup\$

Using an opamp (in+ is fixed 4V ref, in- is a square wave 5/1V), I have an NPN-BJT in the feedback network, and this BJT acts like a latch; once the output goes "high" the BJT pulls in- to ground, effectively latching the opamp to high output until power cycle.

Works perfect in a simulator.

But, in Real Life(tm) there is a quirk that the simulator misses; immediately after power-up, before the power rails and all inputs are stabilized, in+ is higher than in- (which takes some time), so the latch-up happens immediately, which is quite bad.

I've pulled my hair trying to slow down in+ a bit, but realized it's simply not possible. What I need to do is grounding the opamp-output for about 0.5-1us (i.e. disallowing the latch) until all is stabilized.

Is there anything I can do to achieve this? Board space is limited, so if I need an IC + discretes it should be about an SO8 in size.

(I was thinking about a monostable 555 but that short a pulse is a bad idea I think)

Schematics: enter image description here

Explanation

The opamp (OPA2132) is connected as a comparator. In+ is a fixed reference to which in- is compared to. The whole circuit is a voltage monitor, so the output of the opamp is connected to a PMOS-gate (which is controlling the power line to the application).

The RED area: a TL431 voltage regulator sitting on a 5V Vcc-line. R1/R2 sets the working point to 5.5V, meaning that if Vcc rises to above 5.5V, the output (pin 3) is pulled down to 1V. If Vcc is below 5.5V, pin 3 is essentially following Vcc.

Now, disregard the BLUE area for a second and consider RED area + opamp; this works just as expected. When the voltage is tripped (> 5.5V), pin 3 goes low, meaning that in+ is higher than in-, and opamp goes "high".

So far, so good.

Now, the BLUE area is a latch feedback. When voltage is not tripped, in- is higher than in+, opamp is "low" and the BJT is cut-off, i.e. no latch. But when the opposite is true, when the voltage is tripped; in+ is higher than in-, opamp goes "high", BJT gets saturated and pulls down in- to ground. That is, the BJT permanently latches the opamp to "high" state.

This worked perfect in the simulator, but as I wrote above, it does not work in Real Life because - I think - the voltage rails hasn't been stabilized so in+ is initially always higher than in-, making the BJT latch immediately.

If I in this premature latch disconnect the BJT-emitter from ground and then re-connect it, the latch is gone and the whole circuit works as expected (latching only when the voltage is tripped).

For this voltage monitor to work as I want, I need the latch. That is; once the voltage is tripped, the only way to release it should be to power cycle the whole unit.

I yet haven't measured for how long it takes until the power rails and all inputs are stabilized, but is there any way to make the latch work as I want? Like, pulling the opamp output down to ground, once, for time X during the startup? Or delaying in+ for a while so in- manages to reach its initial level?

Or anything at all?

(If I sound too desperate it's because I am; I've been pulling my hair for days)

\$\endgroup\$
  • 1
    \$\begingroup\$ I was going to suggest almost exactly what Andy did, but as he has I sha'n't \$\endgroup\$ – Russell McMahon Apr 20 '17 at 7:23
4
\$\begingroup\$

I've pulled my hair trying to slow down in+ a bit, but realized it's simply not possible.

Of course it is possible - it's the same circuit that keeps a microprocessor in a reset state when power is applied.

So, try putting an RC network in the reference voltage feed to the non-inverting input. When power is applied, the non-inverting input will remain low (due to the capacitor needing to be charged) and during the initial period of power settling, it should work to solve your problem.

R is in series feeding the op-amp input and the C is from the input to 0 volts. Also, to discharge the capacitor when power is removed, a normally reverse biased diode across R will be helpful.

\$\endgroup\$
  • \$\begingroup\$ Like this? i.imgur.com/DQru6nU.png \$\endgroup\$ – bos Apr 20 '17 at 7:35
  • \$\begingroup\$ @bos If your voltage reference source is a potential divider (as shown), put the diode cathode all the way to the top supply. \$\endgroup\$ – Andy aka Apr 20 '17 at 10:58
  • \$\begingroup\$ Connect the cathode to the Vcc-pin, instead of the divider junction? How does that work? I mean, how will the capacitor discharge through that way? \$\endgroup\$ – bos Apr 20 '17 at 11:29
  • \$\begingroup\$ It discharges through the power rails (unless of course your power rails draw so little current when unpowered that it becomes problematic). \$\endgroup\$ – Andy aka Apr 20 '17 at 12:13
  • \$\begingroup\$ Thank you. This solution solved my problem, which is great, and while testing it out I found another problem which I did not anticipate, which is also good, because I'd rather find the problems now before I send the board for manufacturing. \$\endgroup\$ – bos Apr 20 '17 at 13:01
1
\$\begingroup\$

ADDITION TO ANDY'S ANSWER

The low pass filter will indeed keep the reference under your threshold during power up, however it also adds a small issue to power glitches in that the capacitor may still be charged when the power returns.

As such it is prudent to add a fast discharge path to the top of the capacitor to handle that event.

schematic

simulate this circuit – Schematic created using CircuitLab

\$\endgroup\$
  • \$\begingroup\$ It's not an addition to my answer!!!! \$\endgroup\$ – Andy aka Apr 20 '17 at 18:19
  • \$\begingroup\$ @Andyaka yes it is, the diode should go to the rail, not just across the resistor. \$\endgroup\$ – Trevor_G Apr 20 '17 at 18:34
  • \$\begingroup\$ Read the comments under my answer you know, that place where you advertised your answer lol. \$\endgroup\$ – Andy aka Apr 20 '17 at 18:43
  • \$\begingroup\$ @Andyaka OH riiight... good man. I must admit I added my comment there as an afterthought and did not read the others. Oh well at least the schematic is an add LOL.. \$\endgroup\$ – Trevor_G Apr 20 '17 at 18:46

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.