Is there any variable DC-DC converter suitable for pre-regulation?

[bloguetronica]

Hi Tim,

In my case, I need to control the current more or less precisely to the milliamp. The current limit is set by a DAC, and it has to be set proportionally and precisely. That's whay I'm using a full op-amp as a comparator. I can't dispense the voltage control op-amp either for a similar reason (because I want a very precise voltage controll, even far more precise than the current control).

So, I'll use the resistors around the op-amp as you suggested. Also, the voltage divider to control the base of the transistor is also a good idea to diminish the sensitivity (I could use a MOSFET, but unfortunatily it has a too high of a Vth gs).

Kind regards, Samuel Lourenço

[Wolfgang]

Hi Sam,

are you sure you have understood how control loops work ? What you do here is running them uncompensated, which is a major design flaw.
You should think about your gain and phase margins under several load and current/voltage mode conditions.
A good idea is to simulate load step responses and tune compensation so that overshoot and settling time is minimized and no "ringing" occurs.

The voltage/current steering solution is also not optimal. Normally you try to use switching diodes to priorize the active op amp.
In order to speed up the switchover, some minimum amp loading is also beneficial.

[bloguetronica]

Hi Wolfgang,

For the voltage control op-amp, I make sure to use a unity stable op-amp. The op-amp used for the current control is the loose end here. The voltage control part works very well and also compensates the output lead inductance (the reason behing the odd resistor configuration around the op-amp).

Kind regards, Samuel Lourenço

[Wolfgang]

Hi Sam

this is a basic misunderstanding. Even if your op amp is unity gain stable, the control loop as a whole can still oscillate. I am afraid you need to dig yourself a little bit more into the theory of loop stability. Why not try some Jim Williams Appnotes from LT ? Or even Wikipaedia is a good place to start.
Why do you think that even a methusalem like the LM723 has a frequency compensation pin ?

I am absolutely sure that lead inductance is of no importance here. The delays created inside the op amps and in the power stages including filtering are much more important for stability. Not all people like math, but here its needed to understand what is going on, sorry.

Another very good reading is Horowitz and Hill Art of Electronic 3rd Edition Chapter 9 about power supplies and regulators.

regards
  Wolfgang

[bloguetronica]

Hi,

I've been simulating ways to solve the oscillation issue. First, I've implemented the regeneration resistor plus the voltage divider. It improved the performance a bit, but not solved it. Then I proceeded to implement the op-amp compensation circuit, but saw no significant difference. The circuit is still oscillating. It is better to consider taking the whole current control circuitry to the garbage and implement a solution that I can understand - one that only implements the voltage control plus the well needed lead impedance compensation circuitry (aka, remote sensing).

Tested this scheme without the current control and worked very well. I'm ready to trow the towel on this CC crap thing and just use a fixed SC protection as I did before.

What is the point of implementing someone else's circuitry if it was done already? If I'm to reinvent the well, I'll not copy it, especially if I don't understand it. The simple solution, on the other hand, I understand it very well. I can call it a solution of my own.

The only thing "new" in this circuit will be the DC-DC pre-regulator, but if that fails as well, I'll ditch it too. No problems with that.

...
I am absolutely sure that lead inductance is of no importance here.
...

Lead impedance (not inductance) is important and has to be compensated. Remote sensing is the key word. Anyway, it is not the lead compensation circuitry that causes the oscillation. It is the damn current control circuit (unless the simulator is wrong).

Kind regards, Samuel Lourenço

[T3sl4co1l]

R16 is probably too small.  Sanity check: making C1 ever larger, won't do any good when C1 * R16 is already much longer than the duration of the oscillation.

Tim

[xavier60]

Why not just have a diode between the CC op-amp and the Base of T1? Or a PNP Emitter follower. Then there will not be extra gain between the op-amp and the Darlington.

[T3sl4co1l]

Why not just have a diode between the CC op-amp and the Base of T1? Or a PNP Emitter follower. Then there will not be extra gain between the op-amp and the Darlington.

I guess that's not original enough.

Tim

[xavier60]

Why not just have a diode between the CC op-amp and the Base of T1? Or a PNP Emitter follower. Then there will not be extra gain between the op-amp and the Darlington.

I guess that's not original enough.

Tim

This is. In my Agilent U8002A, the CC op-amp, via a diode, is able to pull down the Comp2 pin on the CV op-amp which is actually a connection between the transconductance stage and the input to the op-amp's output stage.  But HP thought of it first.

[bloguetronica]

Why not just have a diode between the CC op-amp and the Base of T1? Or a PNP Emitter follower. Then there will not be extra gain between the op-amp and the Darlington.

I guess that's not original enough.

Tim

Never though of that.

Kind regards, Samuel Lourenço

[xavier60]

There could be complications. The Base of T1 needs to be pulled down to 1v. If you try the PNP follower idea, the op-amp might need a load resistor on the output pin to help get the voltage down.
The transistor's B-E junction will also need to be protected from more than 6v reverse bias somehow.
There is no negative control rail?

[bloguetronica]

Why not just have a diode between the CC op-amp and the Base of T1? Or a PNP Emitter follower. Then there will not be extra gain between the op-amp and the Darlington.

I guess that's not original enough.

Tim

Never though of that.

Kind regards, Samuel Lourenço

On second though, bad idea to use the diode. The op-amp will not not drive the pass transistor properly. Had to replace the op-amp with a 24V part, and still didn't work. Anyway, if HP, aka Agilent, aka Keysight (they are the same) uses this, it is probably patented.

Kind regards, Samuel Lourenço

[bloguetronica]

Hi,

Did a new simulation using different values. Seems to be stable if I do a transient analysis for DC point operation. It shows oscillation if I do transient analysis based on the initial values or on zero values. At least, there is promise.

Anyway, if I do a longer analysis, say, 100ms, I see that the voltage drifts upwards from the CC setpoint of 5V to the CV setpoint of 8V.

Kind regards, Samuel Lourenço

[xavier60]

I doubt that HP would mind someone using their idea for a one off project. It may not have originally been their idea anyway.
I'm mainly curious as to how the simulator deals with it, the re-purposing of the Comp2 pin in that way.

[bloguetronica]

Well, the simulator is not making any sense. There is no reason why the voltage should drift from 5V to 8V in the first place. I'm not blocking any DC with that added capacitor, except for the feedback loop. Thus, DC wise, the op-amp still works as a comparator of sorts. Thus, that behavior is impossible in real life.

It seems I'll have to build this circuit myself and then tweak some component values. In the simulation I forgot to add a capacitor to the output, which is, in any case, essential. The power supply will oscillate without one.

Attached is the circuit I'm going to test. The voltage generators will be replaced by DACs. There is no better simulator than real life.

Kind regards, Samuel Lourenço

[T3sl4co1l]

An unstable system, initialized to a stable point (this is what the DC analysis does), will not move from that point unless perturbed by noise.  Transient analysis does not include noise unless you add it yourself.  So, simply, it's stable because nothing has pushed it off.

It's like stacked bowling balls, except those can remain upright because they have rough surfaces.

The zero-initial-conditions result is the correct one here.

Tim

[bloguetronica]

Hi,

More testing with TINA-TI made me to conclude that the simulator is doing AC analysis to evaluate the transient response. Anyway, didn't make any changes to the circuit, and found out that it doesn't oscillate, no matter how I run the simulation. The drifting was caused by the AC analysis.

Thanks to T3sl4co1l for suggesting a series of measures, such as the degeneration resistor, the voltage divider and the feedback network. The feedback network makes sense, because the op-amp has a gain very close to 1 in AC while acting as a comparator regarding to DC voltages. The transistor has its gain reduced, but the current never falls out of regulation since the op-amp doesn't hit the rails, while still showing a wider variation on its output.

Kind regards, Samuel Lourenço

[bloguetronica]

Hi Tim,

Can I exclude R15? I mean, the value is so small in comparison to R16. What is the cost of doing so?

P.S.: It seems to be stable now, even when simulating with zero initial conditions.

Kind regards, Samuel Lourenço

[T3sl4co1l]

Yes, that is a correct insight.

However, did you try larger values and smaller C values?

The step response is going to be ass with a cap that big.  (Try putting a pulse current in parallel with the load resistance.)

Tim

[bloguetronica]

Hi Tim,

Indeed I find the capacitor too large. The inverting input series resistor (R16) is also a tad large and noise inviting, but still OK.

I think it is better to simulate this on a real simulator, also known as a PCB. TINA-TI is just unrealistic, and is showing abnormal behavior on the CC circuit part, that at least no longer oscillates (not that its feedback loop is closed and there is no chance of having positive feedback locally there). The CC control transistor is working better than ever.

But first, I'll have to implement the pre-regulator. I don't want to have excess dissipation on that pass transistor.

Kind regards, Samuel Lourenço

[bloguetronica]

Hi,

I finally had time to simulate the pre-regulator circuit using a buck DC-DC converter with the PNP. I suspect that the model is incomplete, and that is why I probably can't simulate this.

Could anyone check if the circuit is done properly? I will probably use the LM2592HV instead of the TPS62140. It has a lower switching frequency that is easier to filter out and it withstands 24V input voltage. The topology should be similar, save a few pins. The circuitry attached to the FB pin is the one I want to analyze and check for correctness.

Kind regards, Samuel Lourenço

[bloguetronica]

Hi,

Doing late hours results in mistakes. The circuit was all wrong. Attached is the correct one, which makes far more sense. Still, I can't run a simulation.

Anyway, how can I set the overhead voltage, that is, the voltage difference between the DC-DC output and the one after the pass element? I want to set this voltage difference to 3 or 4V above.

Kind regards, Samuel Lourenço

[prasimix]

If your can't run simulation due to circuit complexity you can try to remove everything except SMPS controller/regulator and PNP tracker and apply voltage source on the PNP tracker as for example in this simulation where Vout_sim is 22 V that give 25 V on the pre-regulator output.

[bloguetronica]

Hi prasimix,

I can't simulate your circuit, since it doesn't have a schematic. However, my simulation shows Vout skyrocketing.

On another note, it seems that the datasheet of the DC-DC converter I'll end up using (which is the TPS54233, instead of the LM2592HV or the TPS62140) is very pick about maintaining the top feedback resistor a constant 10K. Is there any way we can do this, and put the transistor below? I think the alternative is to reduce all resistors tenfold, guessing that the problem is a too high impedance on the feedback circuit.

Anyway, how is the voltage difference set? Can I set it to 4V? How can I calculate the resistors?

Kind regards, Samuel Lourenço

[prasimix]

I can't simulate your circuit, since it doesn't have a schematic. However, my simulation shows Vout skyrocketing.

The schematic (in .png format) is included into zip file if you followed mentioned link.