op-amp locks up on powerup

[JLavoie]

I am starting to suspect there are more than values different about the schematic I am looking at...

If R11 and R12 are the 1k input resistors (R12 and R13 on my schematic), you can drop them some, but will run into diminishing returns quickly, and dropping that value too low will load down the microphone output at some point. I wouldn't advise much below 750 ohms to maintain a bridging termination to 150 ohm source impedance.

The R6 I mentioned is in the wiper of the mic input level pot, and should not interact with the link input gain (at least on the schematic I am looking at.) If R6 is feeding an inverting input to the output opamp, only it's gain will be changed (increased) by dropping the resistance value.

JR

wow, I can't believe I missed that.. of course each feed to the mix bus has it's own gain ratio with the output driver.
so the only hit would be increased noise. I'll have to try it out and see..

thanks.
Jason

 

[ricardo]

wow, I can't believe I missed that.. of course each feed to the mix bus has it's own gain ratio with the output driver.
so the only hit would be increased noise. I'll have to try it out and see..

If John's suggestion is good for you, this might the simplest solution.  The only suggestion I'd add is change R25 from 10k log to 10k lin to better approximate a log pot with small R6.

For those still interested in latching OPAs on single supplies ...

TLC27x were what QUAD used after their 34 & 44 preamps to avoid latching problems with TL07x.  These have CMR including the -ve rail (which is 0V with single supply)  Not sure if there are low noise versions in that range.  Maybe TLVxxx or TLC07x but I've not tried them.

I've used MAX414 in a low noise, high gain, single supply preamp.  But only 10k on the +ve OPA i/p and I haven't checked for latching.  It's CMR doesn't include the -ve rail.

However, TL27x, TLVxxx, TLC07x & MAX41x are all recommended by their manufacturers for single supply.

When the i/p pins go out of the specified CMR, all bets are off.  Some OPAs will start taking significant i/p current & latch.  Low single supplies, Vref which ramps up slowly, make the input pins spend more time out of CMR and exacerbate this.

TI recommend taking loads to mid supply with their TLE2426 virtual ground generator when using OPAs not specified for single supply.  Duu.uuh

Should point out that gurus JR & PRR's original suggestions to check voltages and figure out where currents are going is still the first resort for any trouble shooting.

 

[JLavoie]

Breaking news..
I've breadboarded the circuit segment in question, and it still latches..
and because this seems to be the most asked question, here are all the voltages

Latched:
V+ 10.13
Vref 4.87
- input 0.990
+ input 0.970
output 0.835

unlatched:
V+ 10.35
Vref 5.13
- input 5.17
+ input 5.08
output 5.18

nothing new here, except that we're now working with a Dip IC and breadboard parts.
this rules out interactions with the rest of the circuit, PCB issues, solder issues.
only thing that's the same is the topology and values, and the power supply.

so, success! but now what??

Jason

 

[ricardo]

JR's solution is to keep R9 = R19 = 50k and reduce R6 to make up the gain.  (R25 to 10k lin for nicer law on the knob)  If this works, stop fixing thngs.

If you want to see if it is the OPA latch up, use your breadboard and try removing C15 & R37.  This isn't a permanent soln. cos you must decouple Vref.

Then try TLC272 & MAX412 (even if above dun wuk) in the same breadboard w/o C15/R37.
If OK, add C15/R37 & test.
If OK, try in full circuit.

The breadboard has Vcc 10V so takes one parameter further from possible trouble.

 

[tomelectro]

With R19 = 200K and the input shorted, the + pin of the op-amp will take a couple seconds to reach Vref.  Don't know if that matters.  You might try using smaller input capacitor C10 when using larger R19 and see what happens.

Have you looked around with a scope?  Might want to look at how the rails come up on power-up, look for oscillations, ground-bounce at the op-amp, etc...

 

[JohnRoberts]

Breaking news..
I've breadboarded the circuit segment in question, and it still latches..

nothing new here, except that we're now working with a Dip IC and breadboard parts.
this rules out interactions with the rest of the circuit, PCB issues, solder issues.
only thing that's the same is the topology and values, and the power supply.

so, success! but now what??

Jason

OK, time to identify exactly what is going on...

I'd start disconnecting parts until we get the simplest circuit that latches.

Minimum configuration will be one resistor from output to - input, and one resistor from V ref to + input.

If it doesn't latch with minimum configuration, look at what was removed that caused it to latch.

If it still latches when reduced to just 2x 200k resistors, drop value of the negative feedback R (my guess is this won't make any difference.

If we distill this down to just the resistor from V ref determine sensitivity to that (range of values that latch).

Other questions include how does power supply come up?  Does making PS rise slower/faster  help/hurt.

JR

PS: I still don't follow why this is latching up, but adding a small signal diode (in914, or in4148) from Vref to + input, might be adequate to conduct and pull it up out of being latched and would not conduct under normal operation. With the full parts complement the diode could connect to the junction of 1K and input cap.

 

[JLavoie]

Ok,
with just the chip, 2x 200k resistors, and supply bypass caps etc. it still locks up.

you're right in that the feedback resistor doesn't seem to have any effect.
also, PS rise time doesn't seem to matter. I tried bypassing the input diode and resistor so that V+ comes up a lot faster, and no change

the only thing that did seem to help was I tried pulling the wire supplying V+ to the chip, so that Vref is on well before V+ and it powers up fine (obviously not a good way to work, but it does give us a bit of insight)

after that, I went back to playing with Vref.
reducing the divider resistors made no difference.
I moved the 10uf bypass cap to V+ instead of Gnd and it works!
so then I went back to trying symmetrical bypass caps from Vref to V+ and Gnd and it seems happy..
* Note that someone suggested this early on, and I had tried it on the full units and it didn't help. so maybe with more of a load on Vref we would still have problems..

but at least we're getting somewhere

Jason

 

[JohnRoberts]

Ok,
with just the chip, 2x 200k resistors, and supply bypass caps etc. it still locks up.

you're right in that the feedback resistor doesn't seem to have any effect.

So smaller value on + input doesn't latch, 200k and larger does.

also, PS rise time doesn't seem to matter. I tried bypassing the input diode and resistor so that V+ comes up a lot faster, and no change

Not sure faster is better... but slower may not help with 200k.

Latch is probably occurring because + input is down at or close to ground with + supply already up. While the data sheet claims inputs can go -0.7V without damage, they don't claim proper CM operation below about 1V above - supply. This latching is apparently occurring in that no mans land within 1V of - supply rail.

The fact that lower value resistors do not latch, only means that the mechanism isn't strong enough to sink higher  current and hold them latched. 

the only thing that did seem to help was I tried pulling the wire supplying V+ to the chip, so that Vref is on well before V+ and it powers up fine (obviously not a good way to work, but it does give us a bit of insight)

after that, I went back to playing with Vref.
reducing the divider resistors made no difference.
I moved the 10uf bypass cap to V+ instead of Gnd and it works!
so then I went back to trying symmetrical bypass caps from Vref to V+ and Gnd and it seems happy..
* Note that someone suggested this early on, and I had tried it on the full units and it didn't help. so maybe with more of a load on Vref we would still have problems..

but at least we're getting somewhere

Jason

Maybe not... The input cap with 1k resistor, if the external input was driven from a low impedance with the cap discharged, could also hold the + input down for a period complicating a "trapeze act" solution (this happens before that happens so that other thing doesn't happen).

I repeat my non-linear (clamp diode) suggestion.. What happens with a simple diode in parallel with 200k with cathode end connected to opamp + input and anode end to Vref?  This would increase the current available to pull the opamp out of it's locked condition, and once it is no longer latched, there would be no DC across the diode. Max audio voltage at the + input will be too small for the diode to conduct (for gain of 200x and an output swing of roughly +/-5V, the audio input at output clipping will be +/- 25mV.) A silicon diode with 25mV across it will not conduct.

While this isn't an elegant finesse it should be more simple than gaming the PS rails coming up in sequence with VREF. Perhaps if the stage couldn't coexist with a clamp diode, sequencing would be worth pusuing.

JR

 

[JLavoie]

just tried it an the diode works like a charm..
I was hesitant to try it before because I wanted to make sure that we weren't just hiding from a gremlin, but now that I know that it's just a start-up condition, and that it has nothing to do with the circuit design (other than maybe power) I think I would sleep better at night with the diode in there than a bunch of resistor changes that might just take us out of the danger zone but problems might come back with thermal changes or power noise on startup.

I'll have to verify that it doesn't impact the audio, but this might be the way to go..

Jason

 

[tv]

Try a small (smallest you can find) red LED instead of the diode. It has cca. 1V larger voltage drop.

(It may add 10-50pF of capacitance when not conducting, if that matters. Try to measure it just to be sure).

 

[JohnRoberts]

It seems like this might be a useful thing to mention in their data sheet.... 

Another more linear patch, that should work, is using two resistors, instead of one diode. Connect additional resistors from Vref to the junction of input 1K resistors and the input electrolytic caps.  These 2 R's could be even lower than 20k and not really load down the microphone termination, but would be effectively in parallel with the 200k for DC so provide the extra current to pull it out of latch-down.

This will appeal more to the purists out there, while either way should work. The Rs need to be 1% to not degrade input CMRR.

Note: in theory the diode's small (low pF) capacitance will also imbalance the input CM, so R's may be the better fix all things considered..

JR

 

[JLavoie]

Try a small (smallest you can find) red LED instead of the diode. It has cca. 1V larger voltage drop.

(It may add 10-50pF of capacitance when not conducting, if that matters. Try to measure it just to be sure).

that's a great idea if I needed the headroom, but this particular opamp is for the mic input, so we'll be clipping the heck out of the circuit way before we approach +/- 0.7v input levels

Jason

 

[JLavoie]

It seems like this might be a useful thing to mention in their data sheet.... 

Another more linear patch, that should work, is using two resistors, instead of one diode. Connect additional resistors from Vref to the junction of input 1K resistors and the input electrolytic caps.  These 2 R's could be even lower than 20k and not really load down the microphone termination, but would be effectively in parallel with the 200k for DC so provide the extra current to pull it out of latch-down.

This will appeal more to the purists out there, while either way should work. The Rs need to be 1% to not degrade input CMRR.

Note: in theory the diode's small (low pF) capacitance will also imbalance the input CM, so R's may be the better fix all things considered..

JR

I don't get it.. 20k from Vref to the C10,R13 junction would be essentially like lowering R19 to 20k no?

Jason

 

[JohnRoberts]

I don't get it.. 20k from Vref to the C10,R13 junction would be essentially like lowering R19 to 20k no?

Jason

Yes and no... From the perspective of the latch mechanism it would look like the lower 20k path to Vref to help clear the latch-down, however from the perspective of the microphone, it looks like the input 1k is in parallel with 20k to ground, so no big deal... much less than your previous drop from 1k to 500 ohms.

The second 20k keeps the two inputs in symmetry for good CM balance.

JR

 

[ricardo]

I grovel at Guru JR's feet 

I've used a similar diode solution to prevent latching on LF Sallen & Key filters with TL07x.

Latch is probably occurring because + input is down at or close to ground with + supply already up. While the data sheet claims inputs can go -0.7V without damage, they don't claim proper CM operation below about 1V above - supply. This latching is apparently occurring in that no mans land within 1V of - supply rail.

Yes.  That's what happens with TL07x

Jason, if you ever get your hands on some MAX412s, you might like to try them on the original circuits w/o the diode or extra resistors.

I'd be very interested to know if these are immune to this latching "feature".

 

[JLavoie]

I don't get it.. 20k from Vref to the C10,R13 junction would be essentially like lowering R19 to 20k no?

Jason

Yes and no... From the perspective of the latch mechanism it would look like the lower 20k path to Vref to help clear the latch-down, however from the perspective of the microphone, it looks like the input 1k is in parallel with 20k to ground, so no big deal... much less than your previous drop from 1k to 500 ohms.

The second 20k keeps the two inputs in symmetry for good CM balance.

JR

now I see..
neat trick!

The diode is working well, and doesn't seem to have a negative impact on THD

with respect to maintaining CMR, the incumbent product that we are replacing has an unbalanced mic input  so we're already light years ahead in that department. even with the 5% resistors that I ordered by mistake  :-[

I'll definitely try the resistor trick if I can, although the diode is cheap and easy, and will fit well into the current layout.

thank you guys all soo much for your help! some really creative ideas to a problem that shouldn't exist

Jason

 

[JLavoie]

ok, this is like a good book that I just can't put down..
the resistors work well. I guess it's better to avoid the latching than to let it latch and then force it out..
the only trick is to be sure that it's setup with enough margin to ensure that we never get close to that condition.. but I guess if R19@50k was good, then the new resistors at 22k should supply more than enough current

I'll have to look into how to work it in, but it's crazy how we've gone from a total loss, to two really good solutions..

Jason

 

[JohnRoberts]

I will recast my vote for the two resistor solution as less compromised for audio performance, while the diode fix is pretty benign.

The LED as clamp diode is cute and you could reduce the capacitance by putting a small signal diode in series, but why add more cost and complexity when the simpler fix works.

JR

PS: Ricardo please stop calling me a guru.. This isn't my first rodeo so I just have more experience dealing with such things.

 

[tv]

The LED as clamp diode is cute and you could reduce the capacitance by putting a small signal diode in series, but why add more cost and complexity when the simpler fix works.

You can add a resistor in series with LED to reduce the effect of capacitance. 10K-some..


IOW, if you want to be taken serioulsy, use a stylish BLUE LED (the smallest you can find) with a 10K-something resistor in series.

On the front panel, add this text in small type:
"STABILISED BY ADVANCED LED TECHNOLOGY"

And don't forget to brush the traces with acetone..

 

[ricardo]

... we've gone from a total loss, to two really good solutions..

... and another OPA to add to the list of what NOT to use in single supply or LF Sallen & Key circuits which might be overloaded 

Ricardo please stop calling me a guru..

OK JR.  But I'm still grovelling at your feet  ;D