Very Low Noise Preregulator for Benchtop Power Supply

[Liv]

Downprogrammer must be controlled by the error amplifier instead of a separate comparator.

[Kleinstein]

Many output stages also like to have a minimum current, because transistors usually get much slower al low currents. So the downprogrammer is not just for bringing the volatge back down fast, but also to give a minimum current, even at low output voltages.  A minimum load alows a much faster reaction of the regulating loop as the properties of the output stage does not change that much with load current.

[Liv]

In theory, you're absolutely right. But as practice shows, and modeling for the output stage need not load more than giving feedback divider. Improvements in performance are not visible. You can certainly do some quiescent current, but its value should be much lower than the downprogrammer current for reasons of heating.

[ludo3232]

Hi,

Very interesting schematic

It's possible to use NMOS in place of PMOS with this pre-regulator? I have a lot of NMOS and they are cheaper and are high current rated then PMOS

Thanks

[Boschi]

hi,
yes, it would be possible to use a NMOS, but it would be a  messy.
 you should charge a capacitor at least 5/10V more than the source and with some transistors switch on and off the gate, but you lose like 5/10V and the pre regulator would become useless.

sorry for the bad english, but i think you get the point 

[Marco]

You need a small boost/switched capacitor power supply to get some extra voltage headroom to drive the NMOS efficiently.

On the other hand, the NMOS will attenuate some more noise so there is value in it.

[David Hess]

in a typical opamp type positive drive BJT configuration, i tried in a simulation where in a "weak" load situation (voltage tending to overshoot), if the op amp also "drives" a current limited PNP/PMOS setup to drain the output (maybe 100mA drain), it seems to help "stabilize" the output, yet to try in practical

Many output stages also like to have a minimum current, because transistors usually get much slower al low currents. So the downprogrammer is not just for bringing the volatge back down fast, but also to give a minimum current, even at low output voltages.  A minimum load alows a much faster reaction of the regulating loop as the properties of the output stage does not change that much with load current.

Often when I see this problem it is caused by lack of a base-emitter shunt resistor to remove charge from the base.  It helps a lot to lower its value below that which is necessary to absorb collector-base leakage so that charge is removed more quickly and the faster responding drive transistor takes up more of the load.  Alternatively make the output from the error amplifier stiffer and allow it to pull charge out of the base of the pass transistor directly.

I suspect quasi-saturation of the output transistor at high currents and low voltage drops may explain mysterious transient response behavior I have seen in some designs.

Test the transient response over the entire output current and voltage range and then compensate it for where it is worst.  If the transient response varies a lot, then find out why and fix it.

Referring back to the original discussion post, 1 amp at up to 24 volts is well within the power handling capability of a single output transistor linear design.  Attention will need to be paid to the reference and error amplifier to keep noise low.  If necessary, maximum power dissipation can be cut in half without adding noise by automatically selecting the secondary tap with hysteresis.

[xavier60]

Turn on the transistor at zero crossing and off when the desired voltage across the capacitor provides a very sharp shape of the current and a large voltage spike due to the leakage inductance.

Turn on the transistor to the middle of the half-period provides a smooth current waveform but provides current spike in the beginning.

Slowing opening of the transistor eliminates the current spike.

Therefore, the transistor need to turn on in the middle of the half-period and turn off at zero crossing.

I think it would be even better if the transistor was turned on for some variable time period during the rising part and then during the falling part of the rectified half cycle. It is likely to need a micro-controller to produce the drive timing.

[not1xor1]

Turn on the transistor at zero crossing and off when the desired voltage across the capacitor provides a very sharp shape of the current and a large voltage spike due to the leakage inductance.

Turn on the transistor to the middle of the half-period provides a smooth current waveform but provides current spike in the beginning.

Slowing opening of the transistor eliminates the current spike.

Therefore, the transistor need to turn on in the middle of the half-period and turn off at zero crossing.

I think it would be even better if the transistor was turned on for some variable time period during the rising part and then during the falling part of the rectified half cycle. It is likely to need a micro-controller to produce the drive timing.

I guess you mean switching the transistor at 200/240Hz (depending on country AC line frequency).

I cannot see any advantage in that.
If you switch-off an inductor (in this case the secondary winding of a transformer) while current is flowing, you always get huge voltage spikes.

The advantage of SCR-style circuits is that current decreases naturally, thanks to the decreasing value of the rectified sine wave.
You do not have to waste any power in a "lazy" mosfet switch, working in its linear region to reduce just a bit  transformer voltage spikes.

You just need a large (it has to withstand high current without saturating) inductor after the SCRs (or mosfet switch) to slow down the current charging the levelling capacitor.

The tricky part is synch-ing the switch to turn on at the right time, according to output voltage and load, every 100-120ms.

Bad SCR-style circuits do not care of synch-ing and just switch-on the SCR as soon as the capacitor voltage gets too low, so are quite noisy and inefficient since  the dropout voltage varies a lot and needs a higher threshold to ensure proper regulation.

[xavier60]

So the best is to turn on during the falling side and not have to worry about turn off voltage spikes. High transformer leakage inductance would help soften the current pulse also.
I have been trying to figure out the schematics for the Agilent U8002A. I see that it has a 900uh inductor between the secondary and the bridge. I would like to think that it turns on the MOSFET during the falling side also.

[not1xor1]

So the best is to turn on during the falling side and not have to worry about turn off voltage spikes. High transformer leakage inductance would help soften the current pulse also.
I have been trying to figure out the schematics for the Agilent U8002A. I see that it has a 900uh inductor between the secondary and the bridge. I would like to think that it turns on the MOSFET during the falling side also.

I downloaded that schematic a while ago, but so far have not fully understood how it works, but have not spent much time on it...

In any case I made some modifications to one of those J. Williams circuits to make it work with mosfets.
I did not make many tests, but I think it needs some further development to make it work with any value of output voltage and load and to be able to properly respond to transient load.

I'm attaching the asc file in case anybody is interested in playing with it.
BTW it wastes quite a lot of power in the transformer... it might be more efficient with a huge capacitor across the secondary winding...

It uses a quite rough transformer model I wrote about here: transformer.zip

[T3sl4co1l]

I can think of several reasons why this is a bad idea to begin with, and a few more reasons why it's simply, unconditionally, worse (in performance) than an alternative arrangement of the same parts, give or take... 

Tim

[xavier60]

It shouldn't be any worse than the SCR version. There is a risk of the MOSFET not getting proper Gate drive at low pre-regulation voltages. I'm not certain if D11 is needed.
I once saw the big bites taken out of the secondary waveform in an SCR controlled battery charger and wondered how efficient the idea really was with that large voltage drop being imposed on the secondary voltage.

[T3sl4co1l]

It shouldn't be any worse than the SCR version. There is a risk of the MOSFET not getting proper Gate drive at low pre-regulation voltages. I'm not certain if D11 is needed.
I once saw the big bites taken out of the secondary waveform in an SCR controlled battery charger and wondered how efficient the idea really was with that large voltage drop being imposed on the secondary voltage.

Well, that and the fusing rating of the MOSFET.  Among other things.

Or more to the point, what's the reason that SCRs have such a rating and MOSFETs never do?

Tim

[xavier60]

I wasn't aware of the fuse rating. It may not be important. I wish I had some time to experiment. I would like to put a large power choke into the pre-reg circuit to stretch out the current pulse.  This would be be kinder for the transformer and might reduce noise by reducing current transients.

[not1xor1]

I can think of several reasons why this is a bad idea to begin with, and a few more reasons why it's simply, unconditionally, worse (in performance) than an alternative arrangement of the same parts, give or take... 

Tim

I'm not suggesting to build that circuit, but to use that in simulations to see how an SCR-like circuit works.
BTW the mosfet doesn't suffer any stress at all. Its losses are just 3-5W, because, thanks to the inductor, there is not much current on switching on, and no current on switching off as the half sine wave is already below the capacitor voltage.

xavier60, you are right about the gate drive at low output voltage.
BTW I added D11 because the path from the levelling capacitor to ground, through the PMOS body diode, prevented the circuit to work in the first tests I ran.

It should work better with NMOS switch and control circuit bootstrapped by the positive output voltage.
A differential amplifier connected to Vpre-reg,Vout, with a much lower RC constant and with a reference changing according to the output voltage would also help to deal with large variations of that (output voltage), load and transient loads.

In simulations of a different circuit of this kind, I made last year, I found that a large capacitor, resonating at about AC-line frequency with the input inductor, had a huge impact on efficiency, bringing it almost at the level of switching pre-regulators.
If I get enough spare time, I'll try to find (or re-draw) it, in the next few days.

BTW the other kind of pre-regulator, that one that switches-off at a given output  voltage threshold, has a really poor efficiency at low output voltage.

[not1xor1]

I wasn't aware of the fuse rating. It may not be important. I wish I had some time to experiment. I would like to put a large power choke into the pre-reg circuit to stretch out the current pulse.  This would be be kinder for the transformer and might reduce noise by reducing current transients.

The inductor is already there, and even without inductor the peak current would not be so high to destroy a power mosfet.
The real problem might rather be the inductor itself.
In some cases the circuit might work better with an inductor as high as a few mH and able to withstand 10-20A without saturating.
One should probably build it with those E-I ... pieces... I can't recall the proper English term  but I guess you understand what I mean.

[xavier60]

Yes, the E and I pieces are stacked with their own kind rather than interleaved so that and air gap can be placed between the E stack and the I stack.

[jaycee]

On the subject of downprogrammers, I once heard them described as "something a little bit like a Class AB output stage". So I tried that in my bench PSU design... seen here as Q5 biased by Q6. Ordinarily it functions as a 30mA constant current load, but can either increase to pull the voltage down, or reduce to allow the voltage to ramp up quicker. If more grunt was needed then it would be trivial to add to Q5 with further NPN devices much like a quasi-complimentary output stage of an amplifier.

[David Hess]

On the subject of downprogrammers, I once heard them described as "something a little bit like a Class AB output stage".

That is about right although some may be class-A or class-B.  The variable power supplies I commonly use just have small constant current or constant resistance loads on the output.

Another way to do it is to drive the sink stage in current mode from the error amplifier as shown in National Semiconductor linear brief LB-28 shown below.  It struck me as an odd feature considering that this 10 amp design with careful frequency compensation only has 0.22uF of output capacitance but it is designed to drive any amount of additional load capacitance.

[jaycee]

That is about right although some may be class-A or class-B.  The variable power supplies I commonly use just have small constant current or constant resistance loads on the output.

Yep most of the cheaper supplies I've seen just have a dumb power resistor across the output. I figured I'd try something with a little more finesse in my own design I also designed the PCB so that an extra BC556 can be fitted across Q5 to form a simple constant current sink, in case the class AB biased downprogrammer didnt work out

[David Hess]

That is about right although some may be class-A or class-B.  The variable power supplies I commonly use just have small constant current or constant resistance loads on the output.

Yep most of the cheaper supplies I've seen just have a dumb power resistor across the output. I figured I'd try something with a little more finesse in my own design I also designed the PCB so that an extra BC556 can be fitted across Q5 to form a simple constant current sink, in case the class AB biased downprogrammer didnt work out.

It does not necessarily mean cheap; it may just mean that nothing else was required.  The Tektronix power supplies that I like draw a small current of 3 to 6 milliamps to a point below ground simply so that the output can reach ground without any ambiguity.  Due to a deliberately designed in offset, the outputs actually go a few millivolts below ground to make sure it is possible to reach zero.