Hybride power supply / load / charger

[tuxstef]

Hello,

I am working in my spare time on a huge power supply project with the output specs of the input voltage (DC) from 12 to 24 V and a output voltage of 0-36 (0-4A). To drive higher output voltages than the input voltage a boost converter is added after the power input. The feedback loop's voltage divider is influenced by a digi pot.
The main control loop is designed with a standard op amp and the output stage driven by these opamps is a bipolar Sziklai pair. The Sziklai pair was chosen because of using existing the control loop as an electronic load. A minimum load current of 1 A at 1 V should be possible (saturation voltage, no PMOS/NMOS because of complex transfer function and the need of gate drivers).


But there is only one huge problem left and one annoying:

The output voltage of the system does not go under about 300mV. What can I do against that?

I have attached the (partly) schematics

signal description:

LIN_IN: power stage input voltage (from direction relais)
LIN_OUT: power stage output voltage (to direction relais)
LIN_V_SP: power stage setpoint voltage
LIN_V: power stage output voltage
LIN_I_SP: power stage setpoint current
LIN_I: power stage output current
VIN: supply voltage, depending on mode and input voltage

The other thing is to reduce the 5 MLCCs at the power stage output (C1-4,C13). FRA was successfully without these in LTSpice but under some load conditions there was a huge ringing at the output without the MLCCs. Also a removing of the boost converter did not solve the problem and also a LM317 powered power supply did not help. With the MLCCs a measured FRA within closed loop gave me the same gain bandwidth as the simulation in LTSpice.
A output load of 1kOhm is permanentely added to discharge the MLCCs.

What parasitics could excite this annoying behaviour?

[Kleinstein]

The Sziklai stage alone might be unstable under certain conditions. Without a shunt in the "emitter" path chances are it can oscillate. Also the relatively high input impedance could promote instability.
The current measuring shunt is missing the plan and it is thus hard to tell how the actual circuit looks like.

For the control loops it depends on the parts without a values.

With a Sziklai output stage the OPs have a hard time to turn the output stage all the way off. As this might need to bring the OPs output slightly below 0. It might just work if Schottky diodes are used. So with a single supply one might be better of with a darlington output stage.

The output capacitance for a linear supply is often working better as a combination of low ESR capacitance to improve the step response and a capacitance with some ESR (e.g. 0.1-1 Ohms range) to ensure stability. Low ESR capacitance only, like the MLCCs is not effective to dampen possible oscillations. Again the position of the current measurement is important to as it's impedance can help.

[tuxstef]

Hello,

@Kleinstein
Thank you for your answers!
The diodes are Schottky ones - so ok - I can live with the 300mV.
What do you mean with "shunt in the "emitter""?

So there is no change to remove this capacitors on the output of the power stage?

I have attached now the complete control loop.

[David Hess]

The MC33072A is basically a faster 324/358 with a similar emitter follower output stage which cannot sink significant current at low output voltages.  The added voltage drop of the diodes makes things worse so the operational amplifiers cannot pull the output of the Sziklai pair to ground.

The old Tektronix PS501 and PS503 power supplies use a very similar configuration but have a -5 volt bias supply so the 741 operational amplifiers can pull their outputs significantly below ground cutting off the Sziklai pair.

If a separate bias supply is unacceptable, then I would consider adding a discrete diode or two to the Sziklai pair to raise its effective base-emitter voltage.

I might also replace the 10k bias resistor to the base of the Sziklai pair (Marked R10?  It is difficult to tell.) with a constant current source.  On the Tektronix power supplies I mentioned, a resistor is used from a much higher voltage bias supply so the impedance can be much higher.

That is a nice job on the frequency compensation and decoupling.  I usually add pads for the frequency compensation parts whether they are needed or not.

[Kleinstein]

The current sensing part looks complicated. It also seems like the current sense is at the supply side, before the linear regulator, so it will not help in stabilizing the output stage. So one might need something like a 0.1 Ohms resistor between the output stage and the output capacitors. So one might as well have the shunt behind the the output stage (usually low side) instead.

For the capacitors at the output, it is hard to get away without it, but there is a chance to use less output capacitance. The faster the regulator stage, the smaller the caps can be. Something like 2 µF plus another 1 µF in series with 0.1 to 0.5 Ohms might be realistic. To a certain degree the output capacitance (especially the low series resistor part) is a compromise in on how much overshoot / drop on load changes is acceptable. To get a supply with essentially no capacitance a negative supply and than a class AB like output stage would be the obvious path.

However there is also a kind of additional effective capacitance from the constant current loop - depending on the speed of that loop, this can add a simulated capacitance in the µF range. As far as I can tell the current loop as shown is very slow and thus the simulated capacitance could reach the mF range. In addition the slow current limit can in case of a sudden short add an extra current spike before current limiting sets in. This can be as bad as damaging the output.

To get lower in voltage a different OP with a more powerful output stage might be an option (e.g. OPA171).

[tuxstef]

Hello,

The constant current supply did the job for setting the base current instead with a single resistor! Great idea - thank you very much - so I switched back to a standard Darlington circuit for the pass element. Also the MC33072A can stay where i have placed it - great there is a min voltage of 28mV if setting the output to zero - so only an offset dac/op amp problem!