LDO problems when ground level set by op-amp

[Fulcrum]

Hi everyone.
My problem deals with the following circuit:


Let's go through it. It consists of three main elements: An op-amp in inverting unity-gain operation (X1), an adjustable LDO regulator generating 15V (U1), and an adjustable LDO regulator generating 3.3V (U2). Capacitors C1, C3 and C5 are for stability of the LDO's, and the resistors R5 and R2 sets the voltage level of U1. The purpose of the circuit is to generate a base voltage (called faux_gnd on the schematic) and the two voltages 15V and 3.3V referenced to this faux_gnd. The idea is that I can change faux_gnd as I please with the op-amp, and the LDO output voltages will follow. The change in faux_gnd will be slow compared to the response time of the LDO's, of course.

The problem is that I am seeing two behaviors I did not expect.  In both cases, faux_gnd is set to 0V. One is understandable (I think!) and the other not so much. Let's start with the latter:
The 3.3V regulator does not generate 3.3V, but instead 5V! However, it works perfectly if I connect faux_gnd to GND, i.e. turning off the op-amp and connecting the output to GND via a 0 Ohm resistor. The 15V LDO generates 15V regardless. I cannot explain this behavior of the 3.3V regulator. It has a low-impedance path to GND via the op-amp, so it should be a happy camper, but obviously it's not. The op-amp is powered by +-5V, and so is able to source and sink. Any ideas for why this is happening?

The second unexpected behavior is that when I turn off the 20V line to the 15V regulator, faux_gnd oscillates strongly with a peak-to-peak value of almost 1V and a frequency of 7.2MHz. I realize now that it's probably unstable due to driving all those LDO capacitors of several µF. However, the oscillation stops when 20V is applied again. Also, here's a question: Can I connect the LDO capacitors to GND instead of faux_gnd? Or will that render the LDOs non-functioning? If I could, then I would not have the problem of driving them with the op-amp, and the oscillation would hopefully stop.

Thank you for taking the time to read this, and I hope someone can point out the mistakes I've made.

[wraper]

 
130MHz opamp with multiple capacitors on the output.
BTW opamps don't tolerate capacitive load in general.

[wraper]

Load is powered from real GND or faux-gnd?

[janoc]

Uhh, why not use an adjustable LDO instead of this?! (hint - that LDO *is* adjustable!).

Do you enjoy scratching your left ear with your right foot? 

[Fulcrum]

130MHz opamp with multiple capacitors on the output.
BTW opamps don't tolerate capacitive load in general.

Thank you for your reply. Those facepalms are entirely justified... To be honest, I don't know what I was thinking either. Of course it will oscillate. Now I'm curious as to why it stops when I apply 20V. It's not simply diminished, it entirely vanishes. As for your other reply, which load are you talking about? The LDOs are not connected to any other components.

Uhh, why not use an adjustable LDO instead of this?! (hint - that LDO *is* adjustable!).

Do you enjoy scratching your left ear with your right foot? 

I assume you mean why I don't just replaced the op-amp with adjustable LDO? That's because I need faux-gnd to both source and sink. How would you have implemented this circuit?

[wraper]

Thank you for your reply. Those facepalms are entirely justified... To be honest, I don't know what I was thinking either. Of course it will oscillate. Now I'm curious as to why it stops when I apply 20V. It's not simply diminished, it entirely vanishes. As for your other reply, which load are you talking about? The LDOs are not connected to any other components.

If you are going to connect the load between the Vout and real GND, GND pins of Vregs and decoupling capacitors certainly need to be connected to the real GND and only adjustment pins need to be used for voltage regulation. If the load will be connected to faux_gnd, opamp does not seem to be appropriate component to directly drive it.

[Fulcrum]

Thank you for your reply. Those facepalms are entirely justified... To be honest, I don't know what I was thinking either. Of course it will oscillate. Now I'm curious as to why it stops when I apply 20V. It's not simply diminished, it entirely vanishes. As for your other reply, which load are you talking about? The LDOs are not connected to any other components.

If you are going to connect the load between the Vout and real GND, GND pins of Vregs and decoupling capacitors certainly need to be connected to the real GND and only adjustment pins need to be used for voltage regulation. If the load will be connected to faux_gnd, opamp does not seem to be appropriate component to directly drive it.

The loads will be connected to faux_gnd, yes. I have a section of my circuit that needs to have a private ground level (faux_gnd). This ground level will be adjusted within a range of 0V to -5V by a microcontroller. How would you go about implementing it? I am thinking of simply replacing the op-amp with a different model which is specified as able to drive unlimited capacitive loads.

[wraper]

This ground level will be adjusted within a range of 0V to -5V by a microcontroller.

If you do this, max input voltage of TPS71501 will be exceeded.
As you need it to go below zero, you can use negative adjustible vreg. For example, something as trivial as LM337. All you need is driving the adjustment pin.

[Fulcrum]

This ground level will be adjusted within a range of 0V to -5V by a microcontroller.

If you do this, max input voltage of TPS71501 will be exceeded.
As you need it to go below zero, you can use negative adjustible vreg. For example, something as trivial as LM337. All you need is driving the adjustment pin.

Well, shoot. That's easily remedied by lowering the input voltage a bit. I can set it to 17V, no problem.
The reason I did not use a Vreg to generate faux_gnd to begin with is that I need faux_gnd to both be able to source and sink current. A negative LDO will only sink, yes?
Either way, as the board has already been made, I'd like to find a solution where I can just replace the opamp with something that'll work. This seems to be what I need: LM8261
It can drive unlimited capacitive loads and has the same pinout and package as the lmh6642. I just wish I could understand why the 3.3V LDO generates 5V... If it will do that even when I replace the opamp then I'm still stuck.

[michaeliv]

Can you apply power + ground from a power supply directly to the 3v3 regulator, leaving the rest of the board powered and check what the output is ?
I'm thinking there might be something else at play here.

[Fulcrum]

Can you apply power + ground from a power supply directly to the 3v3 regulator, leaving the rest of the board powered and check what the output is ?
I'm thinking there might be something else at play here.

Did you mean unpowered? I have already tested the 3.3V regulator, it works as expected when its ground pin is connected to GND, as I wrote in my first post (connect faux_gnd to GND and leave the opamp unpowered).

[michaeliv]

Yes, I meant unpowered.
Very strange indeed then, and there are no loads connected to 3.3v and no oscillations in the input 15V  ?
If no loads, if connecting 100k between 3v3 and faux_gnd, will it still output 5v ?

[Zero999]

I think you shoul take a step back:

What are your requirements?

Voltage, current?

You say you need it to be able to sink current so why not use a push-pull servo amplifier buffering a voltage reference? If you need to go all the way down to 0V, you'll need a negative voltage supply.

[Fulcrum]

Yes, I meant unpowered.
Very strange indeed then, and there are no loads connected to 3.3v and no oscillations in the input 15V  ?
If no loads, if connecting 100k between 3v3 and faux_gnd, will it still output 5v ?

You just made me realize something important. Oh I feel like an idiot....  The output is connected to something! I anticipated this problem when I was designing the circuit, but it went to the back of my mind when it was time to assemble it. I just have to populate a resistor I forgot about and that will take care of the 3.3V showing 5V. I will test this first thing in the morning and report back.

I think you shoul take a step back:

What are your requirements?

Voltage, current?

You say you need it to be able to sink current so why not use a push-pull servo amplifier buffering a voltage reference? If you need to go all the way down to 0V, you'll need a negative voltage supply.

The current requirements are low, only +-5mA. I have no heard of a push-pull servo amplifier but I will check it out. Faux_gnd need to go from 0V to -4.5V. In my current circuit it's done by inverting the reference voltage applied to the opamp. The reference voltage is supplied by a DAQ controlled by a microcontroller. However I would like to find a different solution, it's absolutely overkill. The DAQ I use was the cheapest I could find with 10bit resolution, but still it's quite expensive. I don't need hundreds of thousands of samples per second, which seems to be the norm with DAQs. Just have to set the reference voltage every minute or so.

[janoc]

I assume you mean why I don't just replaced the op-amp with adjustable LDO? That's because I need faux-gnd to both source and sink. How would you have implemented this circuit?

You have later said you need it for some part later than needs a separate ground. So no need for a source (power supply?) to be referenced to it, or? Why not to simply create another voltage rail and simply shift the voltages of the rest accordingly so that they are correct relative to this artificial ground?

However, I am not quite sure what and why are you trying to achieve here. There are standard designs for power supplies with multiple power rails, both positive and negative, even MCU controlled. I am pretty sure one of those will work for whatever you are trying to do.

[Fulcrum]

I assume you mean why I don't just replaced the op-amp with adjustable LDO? That's because I need faux-gnd to both source and sink. How would you have implemented this circuit?

You have later said you need it for some part later than needs a separate ground. So no need for a source (power supply?) to be referenced to it, or? Why not to simply create another voltage rail and simply shift the voltages of the rest accordingly so that they are correct relative to this artificial ground?

However, I am not quite sure what and why are you trying to achieve here. There are standard designs for power supplies with multiple power rails, both positive and negative, even MCU controlled. I am pretty sure one of those will work for whatever you are trying to do.

Well, in that case I would have to control three voltages. It seemed easier to me to only control the ground voltage and have the voltage regulators follow it by themselves. It also removes the possibility of the voltage rails not being set correctly. I could for example set faux_gnd to -5V and forget to update the voltage rails accordingly and so fry everything that suddenly would be subject to 5 volts extra.

The whole point of this circuit is that I have photon detector (an MPPC from Hamamatsu) that requires a voltage of about 60V to work. However, this voltage is dependent on the temperature of the MPPC, on the order of ~50mV per degree centigrade. So my plan was to apply a constant 60V to it, and fine tune it by shifting the ground level as I please depending on the temperature. The addition of various components used to read out and amplify the signal from the MPPC introduced the need for voltage rails referenced to the faux_gnd and made the whole thing a bit more difficult.

It seems to be working as expected though, now that I soldered on that resistor I mentioned in my previous post. Faux_gnd shows no oscillations when the voltage regulators are active, and I can set the value of faux_gnd with no problem, voltage rails following it. I would like to get rid of the oscillations when the regulators are not active, so I'll eventually replace the opamp with one that can drive unlimited capacitive loads.

[Zero999]

The current requirements are low, only +-5mA. I have no heard of a push-pull servo amplifier but I will check it out.

If it's only 5mA then then why not simply power the load driectly from the output of the op-amp?

[Fulcrum]

The current requirements are low, only +-5mA. I have no heard of a push-pull servo amplifier but I will check it out.

If it's only 5mA then then why not simply power the load driectly from the output of the op-amp?

Then I'd need three op-amps. One for faux_gnd, one for 15V and one for 3.3V, since they all need to be shifted at the same time. Seems unecessary.

[Zero999]

The current requirements are low, only +-5mA. I have no heard of a push-pull servo amplifier but I will check it out.

If it's only 5mA then then why not simply power the load driectly from the output of the op-amp?

Then I'd need three op-amps. One for faux_gnd, one for 15V and one for 3.3V, since they all need to be shifted at the same time. Seems unecessary.

It seems like a sensible solution to me. Quad op-amp IC packages are widely available and are cheap too.