YALPS! Yet Another Linear Power Supply

[pietr]

A few months have passed and I've finally found the time to prototype this design. The digital part seems to working as expected, but the analog power design is oscillating like crazy. I've built up the the current control, the voltage control and the transistor driver (see attached schematic) on a breadboard (see pics), using potentiometers to set the current and voltage (so the digital part is not at play). I measured the output voltage and voltage control op amp output signal with my scope (see attached screenshots: yellow is inverted output voltage, blue is voltage control op amp output voltage) and it clearly shows the oscillations. The circuit has a 68 ohm resistor as a load and is set to constant voltage mode (my multimeter indicates about 0.5V output voltage). If I change the constant voltage setting using the potentiometer, the oscillation frequency changes (which is expected, I suppose).

Any idea why the circuit is oscillating this badly? I've used very similar components as in the original Agilent PSU, so I would expect this circuit to be more stable. Can this heavy oscillation just be caused by using a breadboard instead of a proper PCB, or is there something more fundamentally wrong with the circuit?

[sync]

Which op amp do you use?

The E3610A uses two pass transistors in parallel. You only one. Maybe this causes instability.

Try shorting R11. The voltage error amp compensation of the original E3610A has some problems. See https://www.eevblog.com/forum/testgear/hp-agilent-e3610a/

Try about 1uF and 75 ohm in series across R20. Look at the E361x schematic. There is R10, C10 across the voltage pot.

[pietr]

Which op amp do you use?

An LF412CN, which is the dual version of the LF411 used in the E3610A.

The E3610A uses two pass transistors in parallel. You only one. Maybe this causes instability.

Maybe, but I don't have another 2N6058 to test this.

Try shorting R11. The voltage error amp compensation of the original E3610A has some problems. See https://www.eevblog.com/forum/testgear/hp-agilent-e3610a/

This improves the response a lot. It makes the output stable for voltages less than about 0.60V, but now I get a different kind of oscillation for higher voltages (see attached screenshot, notice the different voltage scales as opposed to the screenshots in my previous post).

Try about 1uF and 75 ohm in series across R20. Look at the E361x schematic. There is R10, C10 across the voltage pot.

This doesn't seem to work. Doing this causes a lot more oscillation and turning the voltage adjust pot doesn't influence the output voltage anymore. I'm using a electrolytic 1uF cap, maybe I should use a tantalum instead (but I don't have one laying around for the moment)?

[sync]

This improves the response a lot. It makes the output stable for voltages less than about 0.60V, but now I get a different kind of oscillation for higher voltages (see attached screenshot, notice the different voltage scales as opposed to the screenshots in my previous post).

Try lower values for C2. Maybe a couple of 100pf. Keep R11 shorted.

This doesn't seem to work. Doing this causes a lot more oscillation and turning the voltage adjust pot doesn't influence the output voltage anymore. I'm using a electrolytic 1uF cap, maybe I should use a tantalum instead (but I don't have one laying around for the moment)?

The voltage pot should still working. Maybe i described it poorly. Look a the attachment.

Edit: Does your original circuit oscillates only at low voltages?
At 0.5V only 5uA flow through R20. Maybe this is to low for breadboarding. In the E3610A there is a constant current through the resistor (voltage pot).

[pietr]

No progress so far. Here's what I've tried:

This improves the response a lot. It makes the output stable for voltages less than about 0.60V, but now I get a different kind of oscillation for higher voltages (see attached screenshot, notice the different voltage scales as opposed to the screenshots in my previous post).

Try lower values for C2. Maybe a couple of 100pf. Keep R11 shorted.

Lowering C2 doesn't seem to have any measurable effect.

This doesn't seem to work. Doing this causes a lot more oscillation and turning the voltage adjust pot doesn't influence the output voltage anymore. I'm using a electrolytic 1uF cap, maybe I should use a tantalum instead (but I don't have one laying around for the moment)?

The voltage pot should still working. Maybe i described it poorly. Look a the attachment.

The attachment describes how I interpreted it. I double and triple checked the connections I've made, but adding this resistor and capacitor over R20 totally messes up the response: the output oscillates wildly. I tried using a 1uF tantalum instead of a 1uF electrolytic, but this doesn't seem to have any effect.

Edit: Does your original circuit oscillates only at low voltages?
At 0.5V only 5uA flow through R20. Maybe this is to low for breadboarding. In the E3610A there is a constant current through the resistor (voltage pot).

No, the original circuit oscillates at basically any voltage except 0V. After shorting R11 the output is stable for low voltages only. I've tried using 3.3K and 10K for R14 and R20 respectively (10 times lower than their original values) but this seems to slightly lower the voltage at which the output becomes unstable (about 0.4Vinstead of 0.6V).

I'll do some more measurements and experiments tomorrow. Thanks a lot for the help so far!

[megajocke]

Shorting out the 1M5 resistor is probably a good start, but that 1 MHz oscillation needs to be fixed.

There does not seem to be any decoupling caps close to the opamp on the breadboard. That might be a problem. Also, I'm a bit suspicious about those 1N400x diodes. They might have a little too much capacitance at the opamp inputs.

Is the output capacitor really properly connected? The previous scope shot showed the output voltage (I assume) shooting up to 16V in less than a microsecond which doesn't make much sense if the capacitor is properly connected.

[pietr]

I managed to fix the oscillations by adding a 100nF capacitor from the op amp's non-inverting input to ground. I noticed that when I probed this input with my scope, the oscillations disappeared up to about 1.5V on the output. The cap seems to fix the oscillations completely.

Shorting out the 1M5 resistor is probably a good start, but that 1 MHz oscillation needs to be fixed.

There does not seem to be any decoupling caps close to the opamp on the breadboard. That might be a problem.

I added a decoupling cap shortly after taking the picture of the breadboard.

Also, I'm a bit suspicious about those 1N400x diodes. They might have a little too much capacitance at the opamp inputs.

You are right. If I remove the 100nF cap that fixes the oscillations and remove those diodes as well, then the oscillation is still gone. I guess I will leave the diodes and the cap to ground in (because the diodes are there to protect the op amp) and see if the circuit is still fast enough to get a decent transient response.

Is the output capacitor really properly connected? The previous scope shot showed the output voltage (I assume) shooting up to 16V in less than a microsecond which doesn't make much sense if the capacitor is properly connected.

Well the cap is connected straight across the output, without much resistance in its path, so I guess it can charge up very quickly.

Thanks for the help! Now I'm on to test the constant current part...

[sync]

I managed to fix the oscillations by adding a 100nF capacitor from the op amp's non-inverting input to ground.

This will make the control loop really slow.

You are right. If I remove the 100nF cap that fixes the oscillations and remove those diodes as well, then the oscillation is still gone. I guess I will leave the diodes and the cap to ground in (because the diodes are there to protect the op amp) and see if the circuit is still fast enough to get a decent transient response.

Replace the diodes at the op amp inputs with small signal ones like 1N4148. And in the E3610A the diodes are connected differently. Like the attachment.

D27, D28 should also signal diodes. 1N400x are too slow.

[pietr]

Replace the diodes at the op amp inputs with small signal ones like 1N4148. And in the E3610A the diodes are connected differently. Like the attachment.

I will try replacing them with 1N4148's. Note that my voltage control circuit is slightly different from that of the E3610A, by design. The reason is simply that the E3610A uses an analog pot to set the voltage, while my circuit is designed to take a 0-5V signal (from an MCU) as input to control the output voltage. One thing I only noticed now is that R13 in the E3610A schematic is rated for 3W, even though there will be hardly any current flowing through it (if I'm not mistaken). I wonder why it needs to be a 3W resistor... I suppose this resistor plays a different role in the E3610A schematic than R15 does in my design (where it is only used as part of the op amp input protection).

What would be the advantage of connecting the diodes like in your attachment?

D27, D28 should also signal diodes. 1N400x are too slow.

These are already signal diodes, for some reason the part numbers didn't show up on the schematic  .

[sync]

Note that my voltage control circuit is slightly different from that of the E3610A, by design. The reason is simply that the E3610A uses an analog pot to set the voltage, while my circuit is designed to take a 0-5V signal (from an MCU) as input to control the output voltage.

Yes, i noted that already. It's maybe a problem. Your circuit is high impedance compared to the E3610A. Perhaps it's prone to oscillation.

One thing I only noticed now is that R13 in the E3610A schematic is rated for 3W, even though there will be hardly any current flowing through it (if I'm not mistaken). I wonder why it needs to be a 3W resistor... I suppose this resistor plays a different role in the E3610A schematic than R15 does in my design (where it is only used as part of the op amp input protection).

The E3611A uses the same 3W resistor. When a voltage source (battery charging, paralleling PSUs) is connected to the output and the voltage is turned down then the current flows from S+ through CR6, R13, R37 (= 0 ohm) to the negative terminal. At 40V (E3611A) there will be about 1.5W dissipated in R13.

In your design this doesn't happen of course.

What would be the advantage of connecting the diodes like in your attachment?

This is how HP done it since ages. I'm not an EE and perhaps wrong or miss something. If the diodes are connected like in your circuit then the 40V from above will going from S+ through R8, CR6, R13, R37. There would be about -20V at the op amp inputs - ouch!

CR7 is (only?) for discharging C10 if the output are shorted, etc. The PSU is already protected against reserve polarity by CR3.

Your circuit may oscillates because of the diodes capacitance and a additional cap to ground help. In the HP design the diodes goes directly to ground (S+). This maybe prevent your oscillation problem.

[pietr]

The E3611A uses the same 3W resistor. When a voltage source (battery charging, paralleling PSUs) is connected to the output and the voltage is turned down then the current flows from S+ through CR6, R13, R37 (= 0 ohm) to the negative terminal. At 40V (E3611A) there will be about 1.5W dissipated in R13.

I see. You are absolutely right!

This is how HP done it since ages. I'm not an EE and perhaps wrong or miss something. If the diodes are connected like in your circuit then the 40V from above will going from S+ through R8, CR6, R13, R37. There would be about -20V at the op amp inputs - ouch!

I don't see why there would be -20V at the op amp inputs. CR6 would ensure there is max 0.7V (1 diode drop) across the inputs, right?

Strangely I can't seem to reproduce the results I previously had when leaving the diodes out. The circuit still oscillates with the protection diodes removed. It also oscillates with 1N4148 diodes instead of the big 1N4004's. Changing the diode connections like in the E3610A circuit doesn't help either.

I now use a 2.2nF capacitor from R15 (attached to the non-inverting op amp) to ground, instead of the 100nF one I used in my previous post. This seems to be enough to make the circuit stable. I will have to measure whether it messes up the transient response. I also changed R14 to 10K and R15 to 33K, which should provide the same amplification but is lower impedance (but still high enough not to load the PWM low pass filter too much). I'll update the schematics in my opening post shortly.

[sync]

I don't see why there would be -20V at the op amp inputs. CR6 would ensure there is max 0.7V (1 diode drop) across the inputs, right?

Strangely I can't seem to reproduce the results I previously had when leaving the diodes out. The circuit still oscillates with the protection diodes removed. It also oscillates with 1N4148 diodes instead of the big 1N4004's. Changing the diode connections like in the E3610A circuit doesn't help either.

Maybe the breadboarding causes all that trouble.

I now use a 2.2nF capacitor from R15 (attached to the non-inverting op amp) to ground, instead of the 100nF one I used in my previous post. This seems to be enough to make the circuit stable. I will have to measure whether it messes up the transient response.

With this capacitor you basically have a RC low-pass. R is about 26k (R14 and R20 parallel + R15). With 2.2nF the time constant is 57us. I don't think you can get the performance of the E3610A with it. But experiment with it's value in your final build.

I also changed R14 to 10K and R15 to 33K, which should provide the same amplification but is lower impedance (but still high enough not to load the PWM low pass filter too much).

Do you mean R14 and R20?

[pietr]

Right! I was only thinking about the voltage between the op amp inputs, not their absolute value. I should indeed prevent the inputs from getting out of range.

Do you mean R14 and R20?

Yes, indeed. I must have been breathing too much solder smoke while writing my previous post...

[pietr]

It's been quite some time, but I found some time for this project again. I've had some time to think about the problem described by sync above, and I've come to the conclusion that this problem does not occur in my circuit. If 40V is placed between the output terminals in my circuit, the current will go through R8, D7, R15 and R20 (I'm referring to the components in my power.png schematic, not the Agilent schematic). Since R8 is only 1K and R15+R20 is 34k, there will only be about -1.14V at the op amp inputs, which is OK. Even if I would lower R20 by an order of magnitude to 3.3K, there would still only be -9.30V at the op amp inputs.


I do have another problem, or there is at least something about the E3610 schematic I don't understand. In the E3610 schematic, you can see the current shunt resistor R2 on the output path. Between the current shunt and the positive output terminal, there is a ground connection (the +S label). Now, doesn't this mean that current can flow from this ground connection into the output terminal? Such current does not pass through the current shunt, and hence the total current out of the output terminal will deviate from the current set point.

Why isn't this a problem in the E3610 power supply? Is it simply because the current flowing from the ground connection to the output terminal is too small to cause any significant offset, or is there a flaw in my reasoning?

[sync]

It's been quite some time, but I found some time for this project again. I've had some time to think about the problem described by sync above, and I've come to the conclusion that this problem does not occur in my circuit.

Right.

Why isn't this a problem in the E3610 power supply? Is it simply because the current flowing from the ground connection to the output terminal is too small to cause any significant offset, or is there a flaw in my reasoning?

There is a small current. I measured it over the current shunt while nothing is connected to the output. It's about 0.5mA.

[pietr]

Why isn't this a problem in the E3610 power supply? Is it simply because the current flowing from the ground connection to the output terminal is too small to cause any significant offset, or is there a flaw in my reasoning?

There is a small current. I measured it over the current shunt while nothing is connected to the output. It's about 0.5mA.

OK, so my reasoning was correct. But how can this 'current leakage' be calculated and controlled? I suppose my circuit can potentially leak much more current, because all the digital logic I use to drive the main power circuit draws a relatively large amount of current?

[sync]

Your digital logic doesn't matter. It must have it's own separated power supply (E3610A's reference and bias supply). So the current of your logic doesn't flow through the shunt.

[pietr]

But what theoretically prevents the current from going through the load? Does it just come down to board layout? That is, should the board layout ensure that the digital logic ground path to the reference and bias supply has lower resistance than the path through the load?