400V/10A-100W mini dynamic electronic load

[ElectronSurf]

I bought an old AMD CPU heat-sink years ago in the hope of designing a small pure analog electronic load around it for draining batteries etc.
I've designed several electronic loads in the past (up to 400W) but each had their own disadvantages and problems, this time I'm going to do it right hopefully with the help of you guys.

I have several old K2372 MOSFETs laying around, it's adequate for this job and I'm not going to pay $ for IXYS linear MOSFETs!

The main problem was differential current reading, I used to use single ended current reading but the wire length adds resistance to the shunt value and cause error in the readings.

Then I tried differential current reading and fed the output of that op amp to the error amplifier, it was accurate but a bit slow. after some research I found an article showing how to drive a MOSFET with a single differential amplifier, I'm going to use that design as a base.

The other problem was when there's no input voltage connected, op amp output clips to positive rail and making the MOSFET fully ON, causing a huge current surge when input voltage applied.

To fix that I used another op amp to sense input voltage and if it's not connected, pull the voltage reference to negative supply so the error amplifier (now differential amplifier) clips to negative rail thus making the MOSFET fully OFF.

The question is; can I replace the extra op amp with some discrete circuitry? if so how?

There are a few more problems but we'll discuss them as design progress.

Link to the simulator if you wanted to play with it.

[Kleinstein]

One can replace the OP-amp for the voltage detection with a discrete circuit, but it would not be very simple if the threshold is supposed to be small. An OP-amp is the proper solution, but it would still need something like a diode.

A suitable idea would be to add some kind of automatic cross over between 2 modes, a bit like with a lab supply:
If the voltage is high enough, the load would operate in constant current mode.
If the voltage is small the load would work in a kind of constant resistance mode - so clamp the set voltage (usually before a divider) of the current regulator to a value proportional to the voltage.
The simulated resistance has to be larger than the shunt and MOSFET R_on combined.  A little offset can turn of the load all the way if the voltage is very low (e.g. < 10 or 100 mV).
So the current circuit is missing some kind of diode for clamping. As shown it does some linear combination, so a mix in between constant current and constant resistance, but not switch over.

I would consider 400 V and 100 W a bit optimistic for the MOSFET. The FET type is pretty old and may thus handle power dissipation reasonable well.  As there is a thermal transient curve in the datasheet, I would be a bit sceptical about the SOA curve. It is not sure that the cuves are real or just calculated from the transient termal impedance.
As the gate capacitance is already relatively large, it would be more separate channels with a seprate OP-amp per FET if more power is needed.

[ElectronSurf]

One can replace the OP-amp for the voltage detection with a discrete circuit, but it would not be very simple if the threshold is supposed to be small. An OP-amp is the proper solution, but it would still need something like a diode.

A suitable idea would be to add some kind of automatic cross over between 2 modes, a bit like with a lab supply:
If the voltage is high enough, the load would operate in constant current mode.
If the voltage is small the load would work in a kind of constant resistance mode - so clamp the set voltage (usually before a divider) of the current regulator to a value proportional to the voltage.
The simulated resistance has to be larger than the shunt and MOSFET R_on combined.  A little offset can turn of the load all the way if the voltage is very low (e.g. < 10 or 100 mV).
So the current circuit is missing some kind of diode for clamping. As shown it does some linear combination, so a mix in between constant current and constant resistance, but not switch over.

I would consider 400 V and 100 W a bit optimistic for the MOSFET. The FET type is pretty old and may thus handle power dissipation reasonable well.  As there is a thermal transient curve in the datasheet, I would be a bit sceptical about the SOA curve. It is not sure that the cuves are real or just calculated from the transient termal impedance.
As the gate capacitance is already relatively large, it would be more separate channels with a seprate OP-amp per FET if more power is needed.

If it's going to over complicate the circuit, as you said it's better to keep the op amp operate there. Thank you.

[ElectronSurf]

Just finished the schematic, used TL431 as voltage reference even though I had better options at hand like LM4040. I think it's good enough for this little project.

The range switch, switches from ~22A to ~2.2A. it then goes to a multi-turn potentiometer and buffered to set the current.

U1D "sense" the input voltage and if there's nothing connected it will pull the MOSFET gate to negative voltage.
U1C is a simple relaxation oscillator with 50% duty cycle and frequency can be set from ~120Hz to ~12Hz, I had this idea of adding variable duty cycle but then ditched it.
U1B is overheat protection, the limit is going to be 70℃.

The shunt have very low value (10mΩ) and only precision op amp that I have in hand is OP07, I'm not sure if the low slew rate can cause a problem in dynamic behavior.

[Kleinstein]

The capacitor C6 is likely a bit on the small side. I would expect more like 1-10 nF. It may be a good idea to run a simulation to get the right size.
There is no need for R6, the voltage is small enough already.

The voltage set by R8 and R9 is too small - the TL074 can have more offset. One would need a better amplifier or just set the voltage limit higher.
The buffer U1A is not really needed and with it's offset and drive probably doing more harm than good.

The LEDs usually have a quite limited maxium reverse voltage. So one should somehow limit the maximum output voltage of the OP-amps, e.g. with an extra feedback path via diode and series resistor for the set voltage. Or maybe just with a lower supply voltage. U1D is the only part that may want a high voltage so that not that much of a divider is needed.

The shunt for the current is rather small and the MOSFET would not really work well at very high current.

p.s. The MOSFET seems to be an old type, which is good in the sense of a good chance to have an OK SOA. However the R_on is relatively large (0.2 ohm typ) by modern standards. So there is little sense to use it with more than some 5 A, maybe 10 A at the very most. The gate capacitance is already quite large. Not sure how well the OP07 can handle this. Additional FETs would need it's own driver / regulator.

For stablility one would likely need an RC snubber at the output, at least as an option.

[ElectronSurf]

The capacitor C6 is likely a bit on the small side. I would expect more like 1-10 nF. It may be a good idea to run a simulation to get the right size.
There is no need for R6, the voltage is small enough already.

The voltage set by R8 and R9 is too small - the TL074 can have more offset. One would need a better amplifier or just set the voltage limit higher.
The buffer U1A is not really needed and with it's offset and drive probably doing more harm than good.

The LEDs usually have a quite limited maxium reverse voltage. So one should somehow limit the maximum output voltage of the OP-amps, e.g. with an extra feedback path via diode and series resistor for the set voltage. Or maybe just with a lower supply voltage. U1D is the only part that may want a high voltage so that not that much of a divider is needed.

The shunt for the current is rather small and the MOSFET would not really work well at very high current.

p.s. The MOSFET seems to be an old type, which is good in the sense of a good chance to have an OK SOA. However the R_on is relatively large (0.2 ohm typ) by modern standards. So there is little sense to use it with more than some 5 A, maybe 10 A at the very most. The gate capacitance is already quite large. Not sure how well the OP07 can handle this. Additional FETs would need it's own driver / regulator.

For stablility one would likely need an RC snubber at the output, at least as an option.

Thanks for the suggestions, I did the changes you mentioned;

- Increased C6 to 1nF, though in the falstad simulator (which is far from reality) it needed a 22nF cap but we'll see how it goes.
- R6 is needed otherwise the differential amplifier doesn't work.

- Voltage set by R8 and R9 increased.
- I have to buffer the set current, connecting a potentiometer directly to Iset will add resistance to R17 and mess up set current. but I agree with you about offset of TL074, I have to change it to a precision op amp.

- To keep the LEDs reverse voltage below 5V I added another regulator to lower the negative voltage to -5V, also did a zener trick to keep the output voltage of U2D down to about 4.5V.

- I have no other option to change the shunt at the moment, yet "one" did limited the sink current from 25A to 10A. now the range switch, switches from around 10A to 1A.

- Also added another pot to change the duty cycle of relaxation oscillator.

Please tell me more about the snubber you talking about.

[Kleinstein]

For the buffer for the set voltage, one could have the pot directly at the 2.5 V and do range switching and division down only after that.
The buffer would mainly be an issue if a 10 turn put with scale is used. Without a buffer the scale would be slightly nonlinear and range dependent.

The output impedance of the constant current source circuit is somewhat capacitive. In conbination with an DUT / voltage source that is inductive (many voltage regulators are) this can make a prone to osciallation.  An additional snubber at the output (e.g. some 100 nF and 100 Ohm in series) can help to avoid oscillation. How much is needed depends on details like the C6 the FETs used and similar.

Testing a power supply with an electronic load is somewhat tricky, as both the supply and the load can be responsible for oscillation. So one should make sure the electronic load is well behaved and not at the edge to instability by a phase shift of more than 90 degree. This may than still work with a very well behaved supply, but not with all. One may get around some of the problems by adding some actual resistance in series to the electronic load.
At least one should do a good (try to find good OP-amp and FET models)  simulation to check the output impedance.

[ElectronSurf]

I did a few changes and optimization;

- Replaced TL074 with precision quad op amp LT1014.
- Replaced TL431 with LM385 1.2V and removed the range switch.
- Removed overheat protection, it's really not needed as long as power is below 100W.
- Removed negative -5V voltage regulator, (it was added to prevent too much reverse voltage over LEDs).
- Removed zener and replaced it with an LED instead, now I have an indication to show input voltage is connected or not.
- Replaced oscillator LED indicator with a 4148 diode and placed the LED after 4148.
- Removed set duty cycle pot.

And a few minor changes...

[Kleinstein]

If it is about dynamic tests, one may have to look a bit more on the detail of the compensation and check it with a simulation to at least get a reasonable fast response and also to get a truely stable output.
The LT1013/1014 are relatively slow OP-amps and thus no super fast load.

When testing a lab supply with an electronic load any overshoot is a combination of both - a nasty load can make a well behaved supply ring and maybe even oscillate.
An electronic load usually behaves kind of capacitive over a large frequency range. However the phase shift should not go beyond 90 degree, to represent a realistic laod to a supply to test.

For additional trim options one can consider a resistor in series to C3.
There should be some RC series element at the output, so that for the very high frequencies the impedance is set by this RC combination and no longer the active circuitry.
As a MOSFET tends to fail short, a fuse for the output is a good idea too, just in case.

[ElectronSurf]

If it is about dynamic tests, one may have to look a bit more on the detail of the compensation and check it with a simulation to at least get a reasonable fast response and also to get a truely stable output.
The LT1013/1014 are relatively slow OP-amps and thus no super fast load.

When testing a lab supply with an electronic load any overshoot is a combination of both - a nasty load can make a well behaved supply ring and maybe even oscillate.
An electronic load usually behaves kind of capacitive over a large frequency range. However the phase shift should not go beyond 90 degree, to represent a realistic laod to a supply to test.

For additional trim options one can consider a resistor in series to C3.
There should be some RC series element at the output, so that for the very high frequencies the impedance is set by this RC combination and no longer the active circuitry.
As a MOSFET tends to fail short, a fuse for the output is a good idea too, just in case.

In falstad with 0.4-V/μs slew rate at 10A the response is pretty sharp, reducing it causes a bit of overshoot to a few hundred mA.

But with 10nF at lower currents it's a smooth transition instead, I think there's a trade off. just have to find the middle point by changing the C and/or adding R.

Thanks.

[ElectronSurf]

Finished building the project, There's no details on the response time etc. because ATM I don't have any scope at hand.

The schematic is the last one that I have posted and the only changes are removing relaxation oscillator's indicator LED and reduced integrator capacitor to 1nF.

The board itself is 11x6 cm and weights about 300g, the current can easily be set with than ~5mA accuracy.
There's no spark when connecting the source, for a test I set the current to 10mA and source voltage to 30V, then connected an LED in series and it survived.

The dynamic response is yet to be tested with an scope but it will probably work as intended.

My goals was:

1. Eliminating the surge current when source voltage is connected. ✔
2. Be able to set current accurately. ✔
3. Small form factor. ✔
4. Using the parts that I have in hand without spending extra money. ✔

Achieved my goals and satisfied with the result.

[ElectronSurf]

There's a minor problem; when Iset is 0V it's sinking about 5mA and it doesn't change much with voltage variations. only when heatsink temperature goes above ~40℃ the 5mA disappears and actually goes to 0mA.

The heat is not spreading to the control circuitry, maybe just a little. my guess is it's the MOSFET that causes the drift/leakage but why and how? is it the body diode?

BTW I did a few minor changes to the schematic.

[Kleinstein]

Leakage current of the MOSFET should not be that large (more like 1µA range often smaller) and the leakage goes up with temperature.
An odd behaviour could be from oscillations of some kind, possibly going in an out of off. Another point could be main hum effecting the circuit.

[ElectronSurf]

Leakage current of the MOSFET should not be that large (more like 1µA range often smaller) and the leakage goes up with temperature.
An odd behaviour could be from oscillations of some kind, possibly going in an out of off. Another point could be main hum effecting the circuit.

I tried increasing the integrator capacitor to 1uF but it didn't have any effect.

I think it's the shunt! I warm it up with a lighter or even put my finger over it and 5mA offset goes to zero as the shunt temperature rises. and then blowing on it brings back the offset...


Any solution?

[ElectronSurf]

Leakage current of the MOSFET should not be that large (more like 1µA range often smaller) and the leakage goes up with temperature.
An odd behaviour could be from oscillations of some kind, possibly going in an out of off. Another point could be main hum effecting the circuit.

I tried increasing the integrator capacitor to 1uF but it didn't have any effect.

I think it's the shunt! I warm it up with a lighter or even put my finger over it and 5mA offset goes to zero as the shunt temperature rises. and then blowing on it brings back the offset...


Any solution?

The problem was the Iset buffer and pot not going to absolute 0V, outputting with less than ~500uV offset.

That aside I've came to conclusion that avoiding drift with temperature in a hobby project is inevitable and ~5mA drift is perfectly acceptable.

Update:

For a test I pulled 30W of power and current drifted around 7mA, but after temperature reached equilibrium the drift was negligible.

[T3sl4co1l]

For offset: put a small divider from VREF to +IN, which is to say, a large value resistor from U2-2 to U1B-5. Zero doesn't have to be zero, and indeed you might not want it to be, for exactly reasons like this.

If it's not oscillation of course, drift is most likely tempco of R13 or insufficient Kelvin connection thereto.  R10/R14 are differential inputs so wire them to R13 separately from the main current path, or use a 4-terminal resistor.  U2 and RV1 drift, and Q2 and D1 leakage, are also contributors.  In particular, LEDs are also photodiodes, which is to say, D1 leakage depends on ambient light.

Tim

[ElectronSurf]

For offset: put a small divider from VREF to +IN, which is to say, a large value resistor from U2-2 to U1B-5. Zero doesn't have to be zero, and indeed you might not want it to be, for exactly reasons like this.

If it's not oscillation of course, drift is most likely tempco of R13 or insufficient Kelvin connection thereto.  R10/R14 are differential inputs so wire them to R13 separately from the main current path, or use a 4-terminal resistor.  U2 and RV1 drift, and Q2 and D1 leakage, are also contributors.  In particular, LEDs are also photodiodes, which is to say, D1 leakage depends on ambient light.

Tim

The shunt tempco is terrible, I did tried to have separate paths for differential inputs though:



Thanks for the info, I don't think I can make it better than what it is because of the limited budget and equipment.

[ElectronSurf]

I did a few changes to the circuit to have faster response;

- There's no LEDs in the circuit anymore.
- The op amp (A) was clipping to the negative rail when no supply was connected, by adding a diode I reduced that to -0.7V.
- Op amp (B) also use to swing all the way from positive rail to negative rail, now the voltage swing equals zener voltage 5V and -0.7V.
- Also instead of pulling reference voltage down, now it's rising the inverting input of op amp (A).
- Found a 5W non-inductive 20mΩ resistor in my stash so Increased the shunt value to 20mΩ. it has better temperature coefficient.
- Increased voltage reference to 2.5V, LM336.
- Increased integrator RC value, it has smoother response now.
- Decreased input voltage sense voltage divider value to have more current flowing, it was too sensitive before that.



I'm also working on a fan controller circuit, after building this project I realized that the fan shouldn't work always at full speed. by adjusting the fan speed I can keep the heatsink at almost stable temperature which helps with stability and drift.

[ElectronSurf]

The prototype that I made works but have two main problems;

1- Current sink doesn't go to absolute 0A
2- LT1014 is slow

To solve first problem I want to use OP07 to buffer the voltage reference and use the offset null feature to zero the offset.

Want to replace LT1014 with NE5532 which have higher slew rate but It's not a precision op amp, so I had to increase the shunt value to 100mΩ and this force me to reduce the current sink to 5A max to keep the shunt power dissipation lower than 5W.

This is the design I end up with:



Also I want to change the oscillator op amp to TL074 and add fan control feature:



Please let me know if you see any error(s) or have any improvement ideas.

[blackdog]

Hi,

The speed of your current loop depends on your opamp, the MOSFet used and your compensation.
The LT1014 is not a fast opamp but your compensation with the 10nF capacitor might be a bit too much.

A buffer behind the current loop opamp usually helps, so the opamp sees a much lower load and you have more current to drive the Gate properly.

Keep in mind, that the buffer should be wideband enough.
I still have a small collection of LT1010 buffers lying around that I then use for that kind of application.

Not everyone has a collection of IC buffers in stock, but for experimentation you can also put the two opamps of the NE5532 in parallel.
Those then come inside your DC loop of the opamp.
Because of this, you will not suffer from offset errors of the NE5532, resulting in at least 60mA of peak current for the Gate of your Mosfet.
You will have to experiment with the 10 Ohm resistor, and the combination of 3K3 and the 10nF capacitor.

Make sure you pay attention to the build-up, especially if the MOSFet is reasonably fast.
You may need a Snubber from the Drain to the ground point of your 0.1 Ohm resistor.

Happy experimenting! 

Kind regards,
Bram

[ElectronSurf]

Hi,

The speed of your current loop depends on your opamp, the MOSFet used and your compensation.
The LT1014 is not a fast opamp but your compensation with the 10nF capacitor might be a bit too much.

A buffer behind the current loop opamp usually helps, so the opamp sees a much lower load and you have more current to drive the Gate properly.

Keep in mind, that the buffer should be wideband enough.
I still have a small collection of LT1010 buffers lying around that I then use for that kind of application.

Not everyone has a collection of IC buffers in stock, but for experimentation you can also put the two opamps of the NE5532 in parallel.
Those then come inside your DC loop of the opamp.
Because of this, you will not suffer from offset errors of the NE5532, resulting in at least 60mA of peak current for the Gate of your Mosfet.
You will have to experiment with the 10 Ohm resistor, and the combination of 3K3 and the 10nF capacitor.

Make sure you pay attention to the build-up, especially if the MOSFet is reasonably fast.
You may need a Snubber from the Drain to the ground point of your 0.1 Ohm resistor.

Happy experimenting! 

Kind regards,
Bram

Thanks for the input.

[LinuxHata]

So how this project goes on?
Worth repeating or not yet?