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

[bloguetronica]

Hi,

I'm currently thinking of doing a variable power supply. In order to be more efficient, I wanted to use a variable DC-DC converter before my pass Darlington element, basically to keep the input voltage a couple of volts above the output voltage, and to reduce the dissipation on that pass Darlington. I don't have a circuit yet, but it would be something in the likes of the one attached, but adding lots more power besides that DC-DC converter before the pass element (in this case will be a Darlington and not a simple NPN).

I wonder, is there any DC-DC converter controlled directly by voltage? Say, a tracking DC-DC converter? I think I can make a control circuit that just adds 2 or 3 volts and outputs the voltage to be tracked, no problems there. I want this to be very accurate, and that's why I'm not planning to use an off the self DC-DC converter + LDO in the same package or something.

Kind regards, Samuel Lourenço

[David Hess]

DC-to-DC converters which use an external feedback divider can be used to do this.  One complication however is that the DC-to-DC converter will need to operate over a wide voltage range.

[bloguetronica]

Thanks David. But using a DC-DC converter like this wouldn't have the opposite effect? I mean, if the DC-DC converter sees a too low voltage on its feedback loop (from the DAC, or the output - to be decided), won't it attempt to ramp up its output voltage to compensate?

Kind regards, Samuel Lourenço

[David Hess]

Thanks David. But using a DC-DC converter like this wouldn't have the opposite effect? I mean, if the DC-DC converter sees a too low voltage on its feedback loop (from the DAC, or the output - to be decided), won't it attempt to ramp up its output voltage to compensate?

Isn't that what it is suppose to do?  The idea is to use the DC-to-DC converter to regulate the voltage across the linear regulator while the linear regulator regulates the output voltage.  If the output voltage rises, then the output voltage from the DC-to-DC converter rises to compensate and maintain the same voltage across the linear regulator.

[bloguetronica]

Hi David. That was the idea. The DC-DC converter has to keep a certain voltage difference from the linear output, or from the DAC. I see that my logic was flawed. I was thinking about the feedback return seen on many DC-DC converters. So, what I have to use is a variable DC-DC converter that has its voltage set by a voltage divider, right?

Kind regards, Samuel Lourenço

[intabits]

See Dave's video #260 & #329:-

[bloguetronica]

Thanks, intabits. The circuit with the PNP transistor looks like a solution that I can use. It seems I can use any boost DC-DC converter. I wonder if I can use a buck converter. Anyway, if I can find a variable isolated DC-DC boost converter, I can ditch the buck converter after the fixed isolated converter and use it as pre-regulator.

Kind regards, Samuel Lourenço

[prasimix]

Yes you can use a buck converter. Such solution I've used in EEZ H24005 based on LTC3864 (see Power board). Regarding "PNP tracker" don't forget to add 470 pF or so as suggested in Dave's video. Pre-regulator use 48 Vdc on its input and gives about 3-43 V on output.

[bloguetronica]

Thanks prasimix,

It seems that I'll have to use a buck converter. Can't find any isolated DC-DC converter with a suitable output voltage and/or with external feedback.

I've noticed that the boost or buck converter have a fixed output voltage of 5V, on their standard application. Is there any reason? Does that have any influence? This is hard to understand, BTW.

Kind regards, Samuel Lourenço

[T3sl4co1l]

Or -- stick with me here -- how about making a DC-DC converter quiet enough that it doesn't need a postreg?

I guess this is the beginner's dual-control project, in which case, by all means proceed.  I don't see that it's ever beneficial or capable of delivering on its stated or implied goals as well as a simpler solution though.

Tim

[bloguetronica]

Hi,

I can't use a off-the-self DC-DC converter without any kind of "postreg", since I want both low noise and accuracy. Hence the topology with the Darlington being served with a precision DAC and a precision op-amp. The voltage after the DC-DC doesn't need to be accurate at all. It only needs to have a certain difference to mitigate the dissipation on the Darlington., yet still providing enough overhead.

Kind regards, Samuel Lourenço

[T3sl4co1l]

Ah, so you're not going to attempt an SMPS design yourself, only looking for off-the-shelf?

There's still a lot you can do (e.g., servo the output of an inaccurate module with a more accurate reference and opamp), though the noise levels of COTS modules are probably going to suck no matter how you cut it (passive or active filter).

Tim

[prasimix]

Thanks prasimix,

It seems that I'll have to use a buck converter. Can't find any isolated DC-DC converter with a suitable output voltage and/or with external feedback.

I've noticed that the boost or buck converter have a fixed output voltage of 5V, on their standard application. Is there any reason? Does that have any influence? This is hard to understand, BTW.

Kind regards, Samuel Lourenço

Yes, in almost 100% of all app. notes and eval. boards output voltage is fixed since it's easier in that case to ensure stability of the circuit despite the load connected and input voltage variations. But, many of them can be make stable regardless of broad output voltage range. So far, I didn't find any example of converter presented as pre-regulator. Usually people mentioned a couple of Jim Williams LTC app. notes that include some basic switching pre-regulator but nothing based on "modern" SMPS controller. If you're going to design your own SMPS pre-regulator take care about PCB layout and try to use as much as possible SMT parts since they are smaller and can be better organized for lower EMI and circuit stability. If possible (don't know your budget and final goal) try to use 4-layer PCB instead of min. two layers.

[bloguetronica]

Well, I will have to weight all the options carefully. Probably a self made DC-DC converter will do, although I saw an application based on the LM2576 from Texas Instruments. In my application, I don't require a sophisticated DC-DC converter. However, I was not expecting that the DC-DC converter would be one having a fixed voltage, which is counter intuitive. With that feedback loop, the output voltage is now variable?

Difficult to grasp for me, IMO, for now. I want to get there and fully understand this concept from head to toe. Only then, when I understand the whole concept, and when I can customize it, I can call myself a designer of this project. Probably, a discretely designed DC-DC converter, customized to the task at hand, having minimal parts, would be ideal. A design simple, elegant, and one that I can understand. I have to research, research, research... and reinvent the wheel if necessary.

As for the PCB, you gave a good advice. A 4-layer board is definitely in the requirements for this project.

Kind regards, Samuel Lourenço

[bson]

You could relay switch between a few different converters for different ratios.  Then each of them can be optimized for that voltage and a specific load current.  For very low currents you can bypass all of them.

[T3sl4co1l]

Well, I will have to weight all the options carefully. Probably a self made DC-DC converter will do, although I saw an application based on the LM2576 from Texas Instruments. In my application, I don't require a sophisticated DC-DC converter. However, I was not expecting that the DC-DC converter would be one having a fixed voltage, which is counter intuitive. With that feedback loop, the output voltage is now variable?

Easy -- start with an adjustable regulator.  The feedback pin ties to a voltage divider sensing the output.

So, suppose we tie another resistor to the feedback pin.  If we tie the other end to ground, it's in parallel with the lower resistor of the divider, and the ratio drops, so the output voltage rises.  Simple enough?

Suppose we tie the other end to a voltage reference.  If that voltage is 0V (GND), we have the above situation.  If VREF = V(FB), nothing happens, because there's no voltage across the resistor, no current flow and no one's the wiser.  If we tie it to a higher voltage, then some of the current into the bottom divider resistor comes from this one, and some comes from the original (top divider) resistor.  The bottom resistor always draws the same current (because V(FB) is kept constant by the internal control), so that current gets divided between the two top resistor.

This is a linear circuit, so there's no distinction between any of these cases, they're all equivalent.  In short, biasing the FB pin down causes the output to rise, and vice versa.

Or, overall, the FB pin is the -in of an inverting opamp, so putting current into that pin (with the feedback divider connected as usual) causes the output to fall proportionally.  A simple geometric leverage problem.

So it's very easy to set the output with a pot or opamp or DAC.

Tim

[coppercone2]

I thought about doing this a few times but I never saw the benefit of not just using a large transformer and heat sink. Only got interested when I thought about running power MMIC circuits. But they are so expensive I am scared of not just using a commercial PSU because I might not anticipate a failure mode. But with a 800 $ IC you can throw together a giant heat sink and the most inefficient regulators imaginable and it will still be worth it........

[bloguetronica]

Well, I will have to weight all the options carefully. Probably a self made DC-DC converter will do, although I saw an application based on the LM2576 from Texas Instruments. In my application, I don't require a sophisticated DC-DC converter. However, I was not expecting that the DC-DC converter would be one having a fixed voltage, which is counter intuitive. With that feedback loop, the output voltage is now variable?

Easy -- start with an adjustable regulator.  The feedback pin ties to a voltage divider sensing the output.

So, suppose we tie another resistor to the feedback pin.  If we tie the other end to ground, it's in parallel with the lower resistor of the divider, and the ratio drops, so the output voltage rises.  Simple enough?

Suppose we tie the other end to a voltage reference.  If that voltage is 0V (GND), we have the above situation.  If VREF = V(FB), nothing happens, because there's no voltage across the resistor, no current flow and no one's the wiser.  If we tie it to a higher voltage, then some of the current into the bottom divider resistor comes from this one, and some comes from the original (top divider) resistor.  The bottom resistor always draws the same current (because V(FB) is kept constant by the internal control), so that current gets divided between the two top resistor.

This is a linear circuit, so there's no distinction between any of these cases, they're all equivalent.  In short, biasing the FB pin down causes the output to rise, and vice versa.

Or, overall, the FB pin is the -in of an inverting opamp, so putting current into that pin (with the feedback divider connected as usual) causes the output to fall proportionally.  A simple geometric leverage problem.

So it's very easy to set the output with a pot or opamp or DAC.

Tim

Oh, now I see! Hence the PNP transistor controlling the FB pin. So, essentially, if the output voltage (after the linear pass element) drops, the base of the PNP is pulled up, then the collector current rises and pulls up the FB pin (as if the output voltage of the DC-DC was too high), and then the voltage of the output of the DC-DC converter drops accordingly. That's actually simple and brilliant. Essentially, the PNP inverts the compensation effect of the FB pin, right? In this case, it shouldn't matter if the DC-DC converter is fixed or not.

Thanks Tim!

I thought about doing this a few times but I never saw the benefit of not just using a large transformer and heat sink. Only got interested when I thought about running power MMIC circuits. But they are so expensive I am scared of not just using a commercial PSU because I might not anticipate a failure mode. But with a 800 $ IC you can throw together a giant heat sink and the most inefficient regulators imaginable and it will still be worth it........

It is worth it, IMHO. No relays or transformer taps, no relays, you can use a smaller pass element and much smaller heatsink, and less heat drifting the precision of your voltage reference (but that last reason is debatable, and I'm nitpicking here).

Kind regards, Samuel Lourenço

[prasimix]

I thought about doing this a few times but I never saw the benefit of not just using a large transformer and heat sink. Only got interested when I thought about running power MMIC circuits. But they are so expensive I am scared of not just using a commercial PSU because I might not anticipate a failure mode. But with a 800 $ IC you can throw together a giant heat sink and the most inefficient regulators imaginable and it will still be worth it........

Using commercial PSU is no warranty that something does not go wrong. @dmg learned that in hard way. Neither large transformer and heat sink (i.e. classical linear regulator) alone is enough that sensible load can be safely powered (just imagine for example a large power up overshoot). You need a properly designed voltage and current control loops combined with various protection mechanisms to decrease chance of failure to the minimum.

[T3sl4co1l]

Oh, now I see! Hence the PNP transistor controlling the FB pin. So, essentially, if the output voltage (after the linear pass element) drops, the base of the PNP is pulled up, then the collector current rises and pulls up the FB pin (as if the output voltage of the DC-DC was too high), and then the voltage of the output of the DC-DC converter drops accordingly. That's actually simple and brilliant. Essentially, the PNP inverts the compensation effect of the FB pin, right? In this case, it shouldn't matter if the DC-DC converter is fixed or not.

Not quite.  I think their intent was to have a current source into the pin, but the current is proportional to output voltage minus control voltage.  It's a high-side current source, and therefore referenced to the output rail.

So the current mirror really doesn't accomplish anything, and it looks like another resistor in parallel with the top divider resistor.

I suppose that's not inherently a bad thing, it just reduces the gain.  The current mirror does invert the control signal, though, so it acts as a follower rather than an inverter.

If the control input is in turn set by the output (postreg) voltage, it should track, give or take gain and offset, which I'd have to write down and figure out to be sure about.

I think I would prefer to use the two-way resistor divider, and drive the control node with an op-amp that computes the required gain and offset.  At least for starters.

I might then decide to optimize it, potentially to as simple as a single transistor -- in which case, it would be roughly equivalent to what's shown, give or take resistor values and offsets and all that.  Again, I'd have to write it down.

The rest of the circuit looks a bit sausage to my eyes, it could be simplified, and linearized better for more stable control and probably lower noise at the same time, or something like that.  That's just more to the point that almost all lab supply circuits on the internet are crap, just to varying degrees.  I've only seen one that's mostly good (unfortunately, the author got very defensive when given constructive criticism).

I mean, I'd design the perfect supply to end all supply design threads -- but I don't need one, and no one's giving me any money to.

Tim

[bloguetronica]

Thanks Tim! I'll have to consider all the options. Gain is not important. Probably, I could sort of track the DAC voltage to have a faster response (so that the voltage after the pass element doesn't have to depend on the voltage of the DC-DC converter, which in turd depends on the former).

However, I'll add current limiting, and I have to create an unusual circuitry to compute the "should be" output voltage (which is limited to a value when the current is "limited" to a set value, according to the load resistance - or is equal to the DAC voltage when the current is not limited). Probably, on second thought, I'll track that computed voltage indirectly without the use of the funny transistor, although tracking the output voltage directly via that transistor would be much simpler. So, if I add a voltage to that computed value, via an op-amp or even a Zener pulled up, I'll have the tracking reference for the DC-DC converter. Thus, the computed value would be the reference for the op-amp that controls the pass element, and the added value (via the Zener) would be the reference for the DC-DC converter.

Lets see where this goes.

Kind regards, Samuel Lourenço

[T3sl4co1l]

You don't want the SMPS output to track DAC/setpoint output, because it won't follow under a current limit condition.  That's why it's made to follow the actual output instead.

(But you can get just as stable DC with a servo amp setting the SMPS correctly; and, a reasonably fast, stable and well-filtered SMPS can do as well on AC noise, as the postreg can, without the extra loss).

Tim

[bloguetronica]

I'm planning to do something as represented in the attached sketch.

There are some things that were not represented there:
- The amp circuitry that controls the pass transistor, in this case a Darlington, will include lead impedance/resistance compensation (and the possibility of having an external voltage sensing at the load). This is fairly easy to implement, and it only requires 6 precision resistors and a few protection diodes.
- The DC-DC converter control is not that straightforward, and probably is grossly misrepresented here. The voltage that controls this stage is the same voltage that controls the final stage, but with a Zener drop added. Probably instead of the Zener, I'll use the PNP transistor and drive its base with the voltage from the "computation block".
- The "computation block" is fairly simple. You have two DACs, one that sets the voltage, and one that sets the current. It the voltage resulting from the current sensed at the output equals or exceeds the voltage given by the "Iset" DAC, the compute block will output the voltage so that these previous voltages are equal. If not, the "compute block" will output the exact same voltage as the "Vset" DAC.

This will require some extra precision op-amps and comparators, but it allows for a very precise voltage and current control, which is the main goal of this project. As for taking out the post regulator altogether, and replacing it by a filter, well, you will need a huge filter. I want to keep the noise as low as possible, and on any load conditions.

Kind regards, Samuel Lourenço

[prasimix]

- The amp circuitry that controls the pass transistor, in this case a Darlington, will include lead impedance/resistance compensation (and the possibility of having an external voltage sensing at the load). This is fairly easy to implement, and it only requires 6 precision resistors and a few protection diodes.

Hm, if you are talking from experience than you'll be well aware that is not "fairly easy to implement" since there is a lots of places where things can goes wrong. Many of them are even easily visible in simulation and you can spent hours and hours in front of simulation just to address various "use cases" and still not be 100% sure that everything can be fine with real circuit.

- The DC-DC converter control is not that straightforward, and probably is grossly misrepresented here. The voltage that controls this stage is the same voltage that controls the final stage, but with a Zener drop added.
- The "computation block" is fairly simple. You have two DACs, one that sets the voltage, and one that sets the current. It the voltage resulting from the current sensed at the output equals or exceeds the voltage given by the "Iset" DAC, the compute block will output the voltage so that these previous voltages are equal. If not, the "compute block" will output the exact same voltage as the "Vset" DAC.

This will require some extra precision op-amps and comparators, but it allows for a very precise voltage and current control, which is the main goal of this project. As for taking out the post regulator altogether, and replacing it by a filter, well, you will need a huge filter. I want to keep the noise as low as possible, and on any load conditions.

It seems that you're really interesting in making a pre-regulator (mind the prefix in that name ) control overcomplicated. All what you need is a simple PNP tracker connected to output as you are already instructed in some of previous posts. That tracker will faithfully follow output voltage regardless of mode of operation, i.e. CV-constant voltage or CC-constant current when due to current limitation voltage starts to drop and your pre-regulator output voltage will follow that change!

[bloguetronica]

- The amp circuitry that controls the pass transistor, in this case a Darlington, will include lead impedance/resistance compensation (and the possibility of having an external voltage sensing at the load). This is fairly easy to implement, and it only requires 6 precision resistors and a few protection diodes.

Hm, if you are talking from experience than you'll be well aware that is not "fairly easy to implement" since there is a lots of places where things can goes wrong. Many of them are even easily visible in simulation and you can spent hours and hours in front of simulation just to address various "use cases" and still not be 100% sure that everything can be fine with real circuit.
...

Well, I've used the concept already in another project, but I'm yet to test it, truth to be told. Not complicated at all. If you open the schematic attached to the OP, you'll see four precision resistors near the op-amp. That is for internal lead compensation.

...

- The DC-DC converter control is not that straightforward, and probably is grossly misrepresented here. The voltage that controls this stage is the same voltage that controls the final stage, but with a Zener drop added.
- The "computation block" is fairly simple. You have two DACs, one that sets the voltage, and one that sets the current. It the voltage resulting from the current sensed at the output equals or exceeds the voltage given by the "Iset" DAC, the compute block will output the voltage so that these previous voltages are equal. If not, the "compute block" will output the exact same voltage as the "Vset" DAC.

This will require some extra precision op-amps and comparators, but it allows for a very precise voltage and current control, which is the main goal of this project. As for taking out the post regulator altogether, and replacing it by a filter, well, you will need a huge filter. I want to keep the noise as low as possible, and on any load conditions.

It seems that you're really interesting in making a pre-regulator (mind the prefix in that name ) control overcomplicated. All what you need is a simple PNP tracker connected to output as you are already instructed in some of previous posts. That tracker will faithfully follow output voltage regardless of mode of operation, i.e. CV-constant voltage or CC-constant current when due to current limitation voltage starts to drop and your pre-regulator output voltage will follow that change!

Probably I'm over-complicating. But having a voltage (from the computation block) that is equal to the output voltage, and deriving the DC-DC control from there provides faster response, right? Then I could drive the PNP tracker with the voltage from that "computation block" instead of using the output voltage for that. Yes, I dropped the Zener idea, meanwhile.

Without using any "computation block", I could track the output voltage directly, and implement CC-constant by starving the base of pass transistor. I'm not sure about the precision if I do that. Probably I can get away with just another DAC, an instrumentation amplifier for the current sensing (or a dedicated amplifier), a comparator and another NPN transistor. The NPN transistor would be wired in an analogous fashion as depicted in the OP schematic, but its base would be connected to the comparator. I think this last approach is more simple and could be tested first.

Kind regards, Samuel Lourenço