fast switching high current power supply

[vixo]

I've been tasked with building a power supply, it must be 10A, switch at 200kHz into a low resistance / inductive load and be very accurate! What existing equipment uses a power supply like this? Does anyone know resources I could look at to get me started? Playing around in ltspice I can get near but finding it hard to fulfill all criteria.

 

[T3sl4co1l]

Well, start with numbers.

"Fast" -- how fast?  200kHz might be the repeat rate, but how quickly does it need to settle when switched?

"Low resistance / inductive load" -- how resistive, how inductive (range for each)?

"Very accurate" -- how much? And within what time frame?

Is it always 10.000..A exactly, or is this the maximum of a range? What's the minimum, and what error is allowed down there?

Likewise, when switched off, how fast and accurately does current/voltage need to go to zero?  What threshold counts as "off"?

This is a very unusual set of requirements, and it's likely you need something else entirely.  If you can start at the beginning -- describe what you are doing and why -- likely a better overall solution can be proposed.  For example, maybe output level doesn't need to be accurate at all, it just needs to be measured (either self-calibrated, or adjusted gradually over time as a feedback loop).  Maybe RF analysis techniques can be applied (e.g. to measure load resistance/inductance or real/reactive power), or other industrial solutions (e.g. induction heating power measurement).

Tim

[Marco]

I'm going to guess you want something like this :
https://www.researchgate.net/publication/263679196_Constant_Voltage-clamping_Bipolar_Pulse_Current_Source_for_Transient_Electromagnetic_System

The Us voltage in the paper needs to be an adjustable power supply and is calibrated to set the pulse current (will drift, needs a control loop). Making it unipolar isn't too difficult.

[vixo]

I'm going to guess you want something like this :
https://www.researchgate.net/publication/263679196_Constant_Voltage-clamping_Bipolar_Pulse_Current_Source_for_Transient_Electromagnetic_System

thanks! that's very interesting. I will take a closer look this week.

Well, start with numbers.

"Fast" -- how fast?  200kHz might be the repeat rate, but how quickly does it need to settle when switched?

"Low resistance / inductive load" -- how resistive, how inductive (range for each)?

"Very accurate" -- how much? And within what time frame?

Is it always 10.000..A exactly, or is this the maximum of a range? What's the minimum, and what error is allowed down there?

Likewise, when switched off, how fast and accurately does current/voltage need to go to zero?  What threshold counts as "off"?

This is a very unusual set of requirements, and it's likely you need something else entirely.  If you can start at the beginning -- describe what you are doing and why -- likely a better overall solution can be proposed.  For example, maybe output level doesn't need to be accurate at all, it just needs to be measured (either self-calibrated, or adjusted gradually over time as a feedback loop).  Maybe RF analysis techniques can be applied (e.g. to measure load resistance/inductance or real/reactive power), or other industrial solutions (e.g. induction heating power measurement).

Tim

I can't be too specific. I'm designing a unipolar voltage controlled transconductance stage which delivers between 0 and 10A. It needs to settle within 5uS and be accurate to.. hmm lets say ideally 1ppm, with very low noise and very low drift. My colleagues would get very impatient very quickly if I start trying to come up with other ideas, I'm just enquiring in case someone case seen or has experience with something in this ball park.

[Phil1977]

Laser diode controller?

Not just as a question, maybe also as an idea where to find these qualities.

[T3sl4co1l]

Well, sensing current to the ppm's, at thermally-relevant scales, is going to be somewhere between very expensive, and impossible.  The settling itself might not be entirely impossible, but you will need a fast and stable power amp, with considerable gain (about 1/error) at such time scales.  Analog solution, lots of heatsinks, let alone what compliance (voltage) range which you didn't mention, but whatever it is, it has to handle that continuously, even if it's only needed on peak.

The more you can push back on that requirement, or insist upon some kind of servo feedback mechanism, the better.

Laser diode controller?

Not just as a question, maybe also as an idea where to find these qualities.

More like laser galvo controller TBH, but it shouldn't need to be this powerful.  Unless it's steering a defense laser or something.  Or maybe hard drive voice coil actuators, but not modern ones, they're light enough not to need such ratings.

Tim

[victorb]

I've worked on a similar project before. You might want to look at power supplies used in RF amplifiers or motor drivers—they often meet those specs. For resources, I found Texas Instruments and Analog Devices really helpful. They have some solid application notes and design tools. When I was figuring things out, the datasheets for their switching regulators had some great reference designs.

[jonpaul]

budget? Need one pc or for product/prod/

j

[Phil1977]

More like laser galvo controller TBH, but it shouldn't need to be this powerful.  Unless it's steering a defense laser or something.  Or maybe hard drive voice coil actuators, but not modern ones, they're light enough not to need such ratings.

Tim

I thought about something like that:

https://www.fbh-berlin.de/fileadmin/downloads/III-V_Electronics/FBH_Datasheet_LiDAR_Drivers.pdf

– laser driver with GaN transistor in the final stage
with current pulses up to 250 A
– pulse amplitude and width can be controlled
– current pulse width 3 ns - 20 ns

But TBH I can't imagine the application anymore, maybe I overread something in the thread.

[vixo]

Well, sensing current to the ppm's, at thermally-relevant scales, is going to be somewhere between very expensive, and impossible.  The settling itself might not be entirely impossible, but you will need a fast and stable power amp, with considerable gain (about 1/error) at such time scales.  Analog solution, lots of heatsinks, let alone what compliance (voltage) range which you didn't mention, but whatever it is, it has to handle that continuously, even if it's only needed on peak.

The more you can push back on that requirement, or insist upon some kind of servo feedback mechanism, the better.

We have current sense resistors and they are very expensive, so that aspect is tackled, at least until it becomes the weakest link in the chain. Currently I am considering amplifier topologies and considering which would give me the best possible accuracy. At the moment I am using a howland pump with an added power stage but the accuracy is limited by resistor matches and op amp offsets. Is there anything I could use that might be better, or modifications I could make that might improve the performance?

I have been investigating whether its possible to split the job of charging the inductance and holding steady, accurate current to two different circuits driving the same load.

[PCB.Wiz]

I have been investigating whether its possible to split the job of charging the inductance and holding steady, accurate current to two different circuits driving the same load.

How long is the pulse ?  What duty cycle ?
With an inductive load, you will likely need some pre-driver assistance.
Higher voltage helps the dI/dT, but kills the power envelope at 10A, so a change in pre-voltage will help power loss for long pulses.

I can't be too specific. I'm designing a unipolar voltage controlled transconductance stage which delivers between 0 and 10A. It needs to settle within 5uS and be accurate to.. hmm lets say ideally 1ppm, with very low noise and very low drift.

What does settle mean here ? to 1%, 0.1%, or .00001%
The inductance will dictate the current slew rate, with your voltage headroom, and then when you hit the regulation corner, you will have a LCR resonant circuit to contend with.

1ppm is kind of a crazy target, but you can throw money at sub-ppm resistors and very high performance opamps to at least get repeatable.
You will need to calibrate this, somehow, as the best resistors are 0.01% ballparks.
 

[vixo]

How long is the pulse ?  What duty cycle ?
With an inductive load, you will likely need some pre-driver assistance.
Higher voltage helps the dI/dT, but kills the power envelope at 10A, so a change in pre-voltage will help power loss for long pulses.

not totally sure what you mean here by pre voltage, do you mean something like filtering the input control voltage to give it some q to compensate for the effect of the inductor?

The exact requirements are yet to be nailed down 100% but it's expected that the voltage needs to be held anywhere between 10uS and 10S. In case you missed it elsewhere in the thread, this is a voltage controlled transconductance stage so it needs to be anywhere between 0A and 10A.

1ppm is kind of a crazy target, but you can throw money at sub-ppm resistors and very high performance opamps to at least get repeatable.
You will need to calibrate this, somehow, as the best resistors are 0.01% ballparks.

Yep, shoot for the moon and if you miss you'll be among the stars 

[PCB.Wiz]

How long is the pulse ?  What duty cycle ?
With an inductive load, you will likely need some pre-driver assistance.
Higher voltage helps the dI/dT, but kills the power envelope at 10A, so a change in pre-voltage will help power loss for long pulses.

not totally sure what you mean here by pre voltage, do you mean something like filtering the input control voltage to give it some q to compensate for the effect of the inductor?

If you want faster ramp in the inductor you need more initial voltage, say 20V, but once that current hits 10A you need to reduce that input voltage or your current regulator will dissipate ~200W
So you have a slightly slower Vin control that manages power, feeding your current regulator stage.

The exact requirements are yet to be nailed down 100% but it's expected that the voltage needs to be held anywhere between 10uS and 10S. In case you missed it elsewhere in the thread, this is a voltage controlled transconductance stage so it needs to be anywhere between 0A and 10A.

At 10S, you will need thermal power envelope management as well as current regulation.
You need lowest Cin/Crss FETs that can spread the heat.
You should also find what overshoot can be tolerated.

[T3sl4co1l]

A reminder: S is siemens, the unit of [trans]conductance, the gain of the amplifier in question.  s is seconds, the unit of time.

Tim

[Marco]

I have been investigating whether its possible to split the job of charging the inductance and holding steady, accurate current to two different circuits driving the same load.

If you don't and you need significant voltage to push the inductance, you're going to be burning power in the amplifier.

So lets say you need 1V to keep the current steady, but 100V for rise time purposes ... a Howland is going to be burning off that 99V for most of the time, 1% efficiency. That's why the paper I posted had the secondary capacitor stage.

[vixo]

If you don't and you need significant voltage to push the inductance, you're going to be burning power in the amplifier.

So lets say you need 1V to keep the current steady, but 100V for rise time purposes ... a Howland is going to be burning off that 99V for most of the time, 1% efficiency. That's why the paper I posted had the secondary capacitor stage.

Yes thankyou that paper was quite good, so I have been trying to split the job of a) charging with higher voltage FETs and b) holding the voltage with lower voltage Howland. Sort of works, not there yet

[T3sl4co1l]

It may be worth knowing that the envisioned schemes are just special cases for class G/H amplifiers: the supply voltage is variable as needed, stepwise or continuously, reducing power dissipation of the inner linear section.

CRT vertical deflection driver ICs had such a scheme, where the extra supply is single-sided, because the waveform is asymmetrical (flyback/retrace) and the voltage or current ramp is otherwise carefully defined.  I suspect one could use them as quirky audio amplifiers if really wanted.

Tim

[vixo]

coming back to this - I am researching two possible designs

It may be worth knowing that the envisioned schemes are just special cases for class G/H amplifiers

1) Howland pump containing a precision op-amp buffered by a class g power stage. However I'm having a hard time getting the feedback right - either undershoot or overshoot or complicated feedback paths that add offsets / have gain dependent on resistors.

So lets say you need 1V to keep the current steady, but 100V for rise time purposes ... a Howland is going to be burning off that 99V for most of the time, 1% efficiency. That's why the paper I posted had the secondary capacitor stage.

2) a ramp and hold circuit, like in the paper posted. I like this idea but I'm not sure how to make this precise - I can "charge" the inductance by switching high / low voltage either end of the inductance, but in the steady state I'm not sure how to hold that voltage with any accuracy. In the paper the steady state is just defined by the battery voltage and the load, which isn't really what I'm looking for. I would like to use an op-amp to hold the steady state to maximum accuracy, but I need to somehow turn it off in the charge / discharge periods.

maybe someone can suggest something to look at which will get me a little further?

[Marco]

The problem with an amplifier is not so much that it's hard to add a second rail to overcome the inductance, the problem is that it doesn't have a lot of time to stabilize.

A system with a voltage which stays constant between pulses and capacitor to absorb/release the inductive energy can be controlled much slower and thus more precisely ... and you wanted precision.

[vixo]

The problem with an amplifier is not so much that it's hard to add a second rail to overcome the inductance, the problem is that it doesn't have a lot of time to stabilize.

A system with a voltage which stays constant between pulses and capacitor to absorb/release the inductive energy can be controlled much slower and thus more precisely ... and you wanted precision.

Yes, stabilising the (class G) system is the hard bit.

In a ramp / hold circuit, how do I control the steady state? In that particular paper, the current through the load is just dependent on the battery voltage / load. I am looking for a transconductance stage, the output current can be anywhere between 0-10A. I need not only to be able to charge the inductance to the desired voltage, but also to hold it at that voltage. I haven't been able to do this with an op amp, as switching the op amp feedback loop means the loop has to resettle every time it's switched. It also adds offsets which degrade the accuracy

[mtwieg]

I'm guessing this is for some sort of NMR/MRI application.

First of all, accuracy of 1ppm... forget that, not even the super-high end amplifiers achieve that. They may have precision/repeatability around 1ppm (and even that's highly debatable), but not absolute accuracy. Static offsets don't matter (so long as they don't drift over short time scales), those get calibrated out at some point.

Even so, I'm not aware of any current sense resistor technology which can give 1ppm level precision, especially when including shunt amplifier drift (unless you have all of it oven-controlled). If I really needed to meet that spec I would go for a good fluxgate transducer.

For the power amplifier topology, it's hard to recommend something without knowing more about your load (resistance and inductance) and desired slew rate and duty cycle. Class G is an option I've worked on before, but it can be very difficult to get the rail transitions to behave nicely. You fine tune it, and it seems to work well, but then someone says "oh I need to adjust amplitude/slew rate/pulse duration a bit" and suddenly it acts up again.

I don't know what you're referring to when you say "switching the op amp feedback loop". In a class G circuit you only switch the supply rails for the amplifier, not its signal path.

I'm wondering why you're using a Howland current source, as opposed to a simple noninverting amplifier configuration (with the current shunt connected the GND, and the load being in the feedback path). AFAIK the Howland circuit is chosen in cases where the load absolutely has to be connected directly to circuit GND. But it has more error sources than a simple noninverting amp circuit.

[Marco]

In a ramp / hold circuit, how do I control the steady state?

If you go digital, you can use the system from the paper and just replace C1 with a high voltage source.

Use a comparator with an interrupt to make sure voltage always gets removed a bit above the current level you want and a diode so it can't go below 0.

That will prevent damage, but will not keep current steady on the top of the pulse and will have some overshoot and ringing. You can then just slowly dial in the timing of the high voltage pulses and the simmer voltage for an exact pulse shape.

[vixo]

I don't know what you're referring to when you say "switching the op amp feedback loop". In a class G circuit you only switch the supply rails for the amplifier, not its signal path.

I'm wondering why you're using a Howland current source, as opposed to a simple noninverting amplifier configuration (with the current shunt connected the GND, and the load being in the feedback path). AFAIK the Howland circuit is chosen in cases where the load absolutely has to be connected directly to circuit GND. But it has more error sources than a simple noninverting amp circuit.

first point - I am pursuing two different options here, the first is a class G stage, but the second is a ramp and hold circuit adapted from a paper posted by Marco which is what I was referring to there.

I have so far preferred to use a Howland pump because if I were to use an inverting op amp I would have to have 10A in the feedback loop, but if there is something ive overlooked ill change it. We have thought about fluxgate transducers for a future version of this, but right now we have some fairly good sense resistors to use up.

In regards to the duty cycle, Im looking for something that will hold a DC current for as long as it needs to - 1 day? 1 second? The spec is changing but I'm not expecting the time between state changes to be less than 1mS. It just needs to settle within 5uS and hold that current for as long as needs, so I am less worried about pulse timing, other than getting it to be fast enough. I hoped to build a class G stage that is DC coupled and can be used in a PID loop - I think I need overshoot of around 60V for only a couple of microseconds to get the current through the inductor to 10A, then the power rails switch to low power to hold the steady state. As far as I can see most hifi amps (which is where Im looking for class G references) are AC coupled. I need a voltage amplifier stage with high enough gain and slew rate that I can drive with DC. Do you have any suggestions on the feasibility of this? Any references you can share?

If you go digital, you can use the system from the paper and just replace C1 with a high voltage source.

Use a comparator with an interrupt to make sure voltage always gets removed a bit above the current level you want and a diode so it can't go below 0.

That will prevent damage, but will not keep current steady on the top of the pulse and will have some overshoot and ringing. You can then just slowly dial in the timing of the high voltage pulses and the simmer voltage for an exact pulse shape.

this is not a technique I am very familiar with, how accurate would be able to get the steady state voltage? If I am understanding you correctly it sounds like it would produce a lot of noise?

[Marco]

If you replace C1 with a high voltage source and without thermal drift, there are three parameters needed to drive your pulse. Rise time, simmer voltage, fall time. Get those right and you have perfect pulses of arbitrary length with next to no noise (just the simmer voltage power supply, an adjustable power supply has a lot less noise than an amplifier).

So I'm suggesting to just have enough analog circuitry to make sure the pulse gets somewhat close (comparator cut off for pulse height, diode cut off to prevent negative current) and measure the pulse with an ADC. Then determine rise time, simmer voltage, fall time programmatically.

[Marco]

I guess you don't even really need a simmer voltage, a cascode linear current regulator can be fast and handle things away from the edges.

With something like this you just need to dial in the timing for the rising and falling edge. Q1, Q2 and Q3 on for rising edge, Q3 on for top of pulse, Q2 on for falling edge, Q2&Q3 on for zero current. The current supply will be burning continuous power, but if it is a problem there are ways around that (turn it off during zero current).