William ENKI

[lfarrand]

I've got a small update. I finally finished the MOSFET driver boards and assembled the other discrete parts of the system. I just need to connect them together.





I assembled a 27.3nF capacitor from 4x 2000V 6.8pF silver mica Soviet capacitors.





I formed a low inductance, low resistance coil from 10mm copper pipe. This was my first copper pipe coil and I did it by filling it with salt and taping up the ends, before rolling it around a cardboard tube former. It turned out pretty well, although there is one small kink, but it shouldn't affect performance.

The inductance is 140nH, but you can adjust it by moving the output connection point. I'm aiming for ~0.93nH, because when combined with a 27.3nF capacitor that gives a 1MHz resonant frequency which is the frequency that I'm aiming to run the system at.



I assembled the brain of the device using a Teensy 4.1 running Arduino.



I spent the best part of 2 days hacking together the PWM code using hardware PWM timers. There are a bunch of limitations around these timers that weren't super obvious initially. There are limitations to the pins used (some use the SD card pins, so aren't easily accessible), frequencies can't be different across sub-modules, different modules can't be easily synchronized etc.

In the end I've nearly got there. I have 3 distinct PWM outputs operating at different frequencies and duty cycles. The first two are synchronized and have a 50ns dead time period between PWM1 fall & PWM2 rise and PWM2 fall & PWM1 rise to prevent both being turned on at the same time. They operate at 1MHz. The third one is the output stage and that runs at 50Hz.

[lfarrand]

Here are some scope shots of the wave forms for PWM1 (discharge, yellow) & PWM2 (recharge, blue).

1MHz







10kHz



The pulse gets a bit distorted as the frequency increases, so I've included a shot running at 10kHz to better illustrate the wave forms.

PWM1 is simply a 10% duty cycle pulse.

The PWM2 wave form was the trickiest to implement because I had to use a 90% duty cycle and then invert that on channel B to get a 10% duty cycle with the pulse at the end.

[gyula]

Hi Lee,

You have been doing awesome work!

It is very good that the copper pipe coil has extra self inductance so that with the use of the 27.3 nF capacitor assembly you will surely find resonance at 1 MHz by moving the coil connection point (it would be around 930 nH inductance https://www.allaboutcircuits.com/tools/tank-circuit-resonance-calculator/   but when you couple it to another coil,  resonance will need little readjusting as you surely know).

Hopefully the resonant peak voltage across the paralleled capacitors will not exceed their rated 2000 V breakdown voltage. Normally such capacitors can withstand little higher than their rated specification but not too far. The resonant current will be enormous too.

Regarding the distorted pulse, it will surely get restored to sharp rectangular waveforms via the MOSFET driver ICs. Regarding the driver ICs, they may have develop excess heat at the 1 MHz switching frequency, you are surely aware of this.

Thanks for showing your progress.

Gyula

[lfarrand]

Hi Lee,

You have been doing awesome work!

It is very good that the copper pipe coil has extra self inductance so that with the use of the 27.3 nF capacitor assembly you will surely find resonance at 1 MHz by moving the coil connection point (it would be around 930 nH inductance https://www.allaboutcircuits.com/tools/tank-circuit-resonance-calculator/   but when you couple it to another coil,  resonance will need little readjusting as you surely know).

Hopefully the resonant peak voltage across the paralleled capacitors will not exceed their rated 2000 V breakdown voltage. Normally such capacitors can withstand little higher than their rated specification but not too far. The resonant current will be enormous too.

Regarding the distorted pulse, it will surely get restored to sharp rectangular waveforms via the MOSFET driver ICs. Regarding the driver ICs, they may have develop excess heat at the 1 MHz switching frequency, you are surely aware of this.

Thanks for showing your progress.

Gyula

Hi Gyula,

Thanks! Yes I'd noticed that the gate driver ICs got quite hot at high frequencies, so my plan is to mount some 140mm computer fans to the sides of the MOSFET stack in a push/pull arrangement. I was toying with the idea of mounting mini heatsinks on the tops of the gate driver ICs, but I'm not sure if this would compromise the low/high voltage isolation by providing a bridge over which high voltage might travel across. It's a big enough problem that PCBs have slots drilled out to reduce this risk, so I think adding a heatsink might add another plane for the voltage to travel over.

I'm not sure what to expect when running the system at it's resonant frequency. I'm aiming to cancel out capacitive and inductive reactance, and be left with only DC resistance. In a series resonant arrangement the voltage will rise, so I need to be mindful of this and protect the MOSFETs, capacitors and diodes from an overvoltage situation.

Also, small correction: when I said 4x 2000V 6.8pF, I meant 4x 2000V 6.8nF (6800pF).

[gyula]

Hi Lee,

Yes, the distances count a lot at higher voltage levels and a protruding metal surface may reduce sparking-distance. Hopefully the air blowers will help reduce heat on the driver ICs.

It would be a good approach to start the tests with a much reduced supply voltage for the switching MOSFETs, and doing initial tunings and check voltage peaks with a HV probe if you have such. And then increase supply voltage gradually and doing fine tuning accordingly.

At resonance, the reactive voltage across the L and C and the current will be at its maximum values and may change drastically during the tuning in any direction, so doing small steps at a time is a good approach.

Yes, I noticed the 6.8 pF typo, and it did not cause confusion, a coil like you made should need several  nF capacitor to make the circuit resonate at 1 MHz (namely 27.3 nF).     The 2 kV limit remains, unfortunately for the paralleled caps, hopefully it will be enough.

Gyula

[gyula]

Are you building the circuit as you showed here: https://www.overunityresearch.com/index.php?topic=4467.msg104929#msg104929    ?

ADDITION   You wrote:  "...(you) have a 50ns dead time period between PWM1 fall & PWM2 rise and PWM2 fall & PWM1 rise to prevent both being turned on at the same time. "

Can you vary the dead time to decrease or maybe increase it if needed ?  Just to find the best adjustment possible, the MOSFETs rise and fall time specifications have a certain range and it is better to have means for some correction.

[lfarrand]

Are you building the circuit as you showed here: https://www.overunityresearch.com/index.php?topic=4467.msg104929#msg104929    ?

ADDITION   You wrote:  "...(you) have a 50ns dead time period between PWM1 fall & PWM2 rise and PWM2 fall & PWM1 rise to prevent both being turned on at the same time. "

Can you vary the dead time to decrease or maybe increase it if needed ?  Just to find the best adjustment possible, the MOSFETs rise and fall time specifications have a certain range and it is better to have means for some correction.

Yes I'm building that circuit with a few adjustments. For example, the FWBR as drawn won't work because the primary capacitor will simply drain into the output cap directly. I need to think about how to stop that happening and only make the FWBR active after the primary cap has discharged.

And yes, the dead time is configurable. It can be anything you want, so long as dead time + pulse time are less than the period length.

[lfarrand]

I've revised my circuit to fix the FWBR problem, and also replaced the manual startup process with an automated solution so that things can be coordinated.

[gyula]

Hi Lee,

Maybe your tests will show as a no problem,  I would mention the secondary coil of the step up transformer is always in parallel with C1 as you drew, so the winding resistance may discharge C1 rapidly after the charge is done, In fact the coil may form an LC circuit with C1 and the input charge would oscillate between them till full ring down. 

All I mean is that the charge you initially input to C1 may be dissipated unwantedly in a quicker way via the secondary coil than it would in case the secondary coil would be separated from C1 just after the charge up process.
Note that the flyback spike the 9V also creates across the step up transformer primary coil would also be transformed by the secondary coil with its own polarity (the polarity the 9V battery establishes during the on time of Q1 will be "overwritten" by the much higher opposing flyback spike).

I do not have a simple solution yet, will be thinking on this.  You may wish to avoid pushing once a simple push button which would charge up C1 from a HV DC source and then you release the button (at 2000 V level this may be cumbersome).  Maybe with a mechanical relay the push button could be replaced?

Gyula

[lfarrand]

Thanks for your feedback Gyula, you make an important point. I hadn't considered that the charge in C1 might dissipate in the windings of the transformer. This is actually a harder problem than it originally appeared to be. The devil is in the detail as they say! Only having a small capacitance really constrains the time window within which I can charge and discharge C1 at start up, otherwise the charge leaks out.

Do you think adding a diode might help? It would enforce the unidirectional flow of current from the step up transformer secondary and prevent the secondary and C1 from forming an LC circuit.

[lfarrand]

I also did some more research into the current handling capabilities of TO-247 MOSFETs and found that my original assumption that they were limited to 70A is not correct. It depends on the resistance and how quickly you can dissipate heat. In addition, the bonding between the package and silicon die is the weakest link, not the leads as I had originally thought.

It turns out that Infineon use something called .XT interconnection technology, where they bond the die to the package using a diffusion soldering process in order to reduce the thickness of the solder layer. This results in a higher current carrying capability and lower operating temperatures.

I've soldered thick tracks to the MOSFET leads and they're mounted to heatsinks which will be actively cooled by fans. I'm hoping this arrangement will allow heat to dissipate quicker. When I was soldering the tracks they were doing a good job of dissipating the 300°C from the soldering iron.

I think it's safe to assume that the current ratings on the datasheets are achievable, so long as the heat is dissipated quickly.

[forest]

nice. we need power mosfet of really low Rds according to Ismael Aviso. Did you managed to work with mosfets in Mhz range effectively ?

[gyula]

...
Do you think adding a diode might help? It would enforce the unidirectional flow of current from the step up transformer secondary and prevent the secondary and C1 from forming an LC circuit.

Hi Lee,

Well,  perhaps the simplest solution for isolating the secondary coil of the transformer from the capacitor C1 in your schematic would be to use a GDT in series with the coil (instead of a diode) which fires around 1800- 2000 V and could charge up C1 when fires once. 

All the initial charging circuit you show around the 9 V battery would remain as is, except for this suggestion: connect a diode in series with the 9 V battery with a polarity to conduct when the MOSFET is ON, this way the flyback spike of the 9 V coil would be blocked to reach the MOSFET. 

Regrading the use of a diode in series with the secondary coil I am not sure it would give good isolation.

If you agree to use a GDT, see this link: https://www.littelfuse.com/products/gas-discharge-tubes/high-voltage-gdt.aspx 

Perhaps it would be wise to start the tests at a much lower than 2 kV, to prevent the expensive components from getting smoked. (This would involve to have GDTs at that lower voltage levels too, https://www.littelfuse.com/products/gas-discharge-tubes/medium-to-high-surge-gdt.aspx )

Gyula 

[Excelsior]

Just thought I'd add this for what it's worth:

Would it be a good idea to use a peltier module on each FET?
It'd also require thermal pads/paste, oil, pump, good fans and a decent size "water block" (+ a stable DC supply)
But this could be a good way to keep temps under 175°c (Rule of thumb for FET's iinm)

Then use a laser temp gun to probe hotspots mid operation

I imagine them being powered separately to the device itself, for clarity (until loop is achieved)

I'm still educating myself on the principles involved so can't add much technically speaking for now,
But I can vouch for these being an effective (if inefficient) solution to thermal issues I have encountered.

There is of course the Delta T within the switching devices in mention, to be observed,
And the fact that these things can generate a considerable amount of condensation,
But the only other option I see to do this would be to gut a small fridge and adapt the innards..
(Tough job to do without venting the pressure. Requires a recovery unit etc, etc,)

But you could then stop worrying about melting expensive hardware
or at least mitigate the risk.

[lfarrand]

nice. we need power mosfet of really low Rds according to Ismael Aviso. Did you managed to work with mosfets in Mhz range effectively ?

I've tested the Infineon MOSFETs up to 1MHz, which is what the gate driver is rated to on the data sheet. When I was testing using the Microchip SiC MOSFETs I was able to get up to 5MHz before the signal got distorted. I haven't tried the same with the Infineon ones yet though.

[lfarrand]

use a GDT in series with the coil (instead of a diode) which fires around 1800- 2000 V and could charge up C1 when fires once.

Good idea! I have various GDTs in my collection, ranging from 200V to 2000V so I'll try to find them and put them to good use.

[lfarrand]

Would it be a good idea to use a peltier module on each FET?

That's a good idea and an excellent way to keep them cool, but I imagine they consume a fair amount of power.

I've got the MOSFETs mounted to the heatsink using a clip and a thermal pad.

I'll monitor the temperatures and see if more extreme cooling is required. No point over complicating things until it's a problem. If I do get to the stage of requiring better cooling then that means a lot of power is being switched, which would be a good problem to have!

For monitoring, I'm going to use some temperature sensors connected to the Arduino. I'll monitor the MOSFET heatsink and gate driver IC initially. I'm also going to add voltage & current sensors to each MOSFET driver PCB to monitor the DC input. I'm also going to monitor the voltage on the output capacitor(s) and connect it to the Arduino via a resistor divider (to map high voltage to a lower voltage range) and into an optoisolator. Finally, I'll add voltage and current sensors to the batteries.

In order to view the real-time status of the sensors I'm going to expose a web server on the Arduino. I'm also planning to have a rudimentary UI to allow adjusting of PWM frequency, duty cycle and dead time on the fly. The value will be exported via MQTT to InfluxDB (timeseries database) and I'll visualize the output using Grafana, so that I can track things over time.

Can anyone think of any other metrics I should monitor?

[Excelsior]

Broad spectrum RF receiver maybe?
Gauss meter / Geiger counter / Tri-field meter
or apparatus that will detect similar phenomena?

We don't know what other effects this could incur.
Would be a shame to not log them if they are present.
So long as these measurements don't detract from the main focus.

I like the idea of remote operation via Arduino - That's really cool

[lfarrand]

Broad spectrum RF receiver maybe?
Gauss meter / Geiger counter / Tri-field meter
or apparatus that will detect similar phenomena?

I've got a Geiger counter, a Tri-field meter and a compass that I usually have running when experimenting. You never know what might turn up! It's a shame that I won't be able to store those readings over time. There's no way to extract the readings on the meters that I have unfortunately. I think being able to correlate various metrics over time will prove to be invaluable when looking at the results of tests.

[Excelsior]

Idea for cooling switching devices:

This is roughly how I might approach an active cooling system which could allow for better switching performance.
$100 - $150 to set up if you're thrifty about where you get your stuff from.

Power consumption should be irrelevant in a test model / prototype
(I'd guess a system like this could use probably about 200w or more)
Easily covered by a loop if it self powers, as we hope.

Condensation would be the only issue I see manifesting .. not so good with HV nearby
I'm sure a "workaround" exists. (Insulate the pins, bend them upwards and away from the Delta T perhaps?)
Might not even be an issue if the full rated power is not delivered to them
(Driving the Peltier devices at <50% may even suffice)

Oil could be used instead of water, provided the rubbers / plastics involved in fluid transmission won't be affected.
A smear of high temp thermal silicone could be used to bed IGBT/FET/SCR onto the "cold face" of the Peltier chip.
Then the same to bed the Peltier "hot face" onto the block. (Thermal silicone withstands 1000+ ^oC)

Might be worth considering using multiple blocks, one for each device, for the potential "arcing" from the rear plate.

Some of the components listed are less than ideal
Just wanted to put the idea out there in its most basic form for others to elaborate on.
Chill your beans!

[lfarrand]

Condensation would be the only issue I see manifesting .. not so good with HV nearby
I'm sure a "workaround" exists. (Insulate the pins, bend them upwards and away from the Delta T perhaps?)

That's a nice cooling setup. It would keep the FETs cool for sure!

Regarding the condensation problem - liberally spraying the FETs and any electronic equipment likely to be affected with conformal coating should fix that. I've sprayed my boards with MG 422B, mainly to prevent HV arcing, but it also provides moisture resistance.

[lfarrand]

I've got a web server running on the Arduino, as well as some temperature sensors hooked up. I can display the current sensor values and also update the duty cycle, PWM frequency and dead time.

This is my to do list and the items I've completed so far:

1. Solder Ethernet pins and attach ethernet port
2. Use QNEthernet library for Ethernet
3. Use EthernetWebServer library for web server. Also refer to this link for Teensy 4.1 specific info.
4. Add temperature sensors to MOSFET heatsinks, gate driver IC & DC/DC converter.
5. Add voltage & current sensors to batteries and DC input of each MOSFET driver board.
6. Add voltage sensor via voltage divider to capacitors. Use Zener to protect optocoupler.
7. Publish sensor data via MQTT using AsyncMQTT library
8. Build a simple web page to allow setting PWM frequencies, duty cycle, dead time. Also display real time sensor values and current settings values.

I also had to prepare a new Arduino board with new Teensy 4.1 board because I'd soldered the last one directly and I couldn't access the under side in order to solder the Ethernet pin header. I tried to desolder and remove it, but it broke. I had to sacrifice it for the greater good. I added some pin header sockets to the solderable breadboard, so that I can easily remove the Teensy 4.1 board in the future.

The new board turned into a mini project because there are so many wires underneath for the PWM outputs, and a couple of things didn't work due to small errors when wiring them up. It took a disproportionate amount of time to debug and fix them. This wasn't unexpected as I'm a software developer by profession, so finding and fixing bugs is a daily activity for me!

[gyula]

Hi Lee,

Hats off to your devoted activity with all the "bells and whistles"!
   (I mean I hope your arduous additions of features will be justified by the overall result of having a COP > 1 setup.) 

Keep it up.

Gyula

[Excelsior]

I second Gyula - Keep at it !
Awesome stuff

Can't beat a well thought out to - do list with these projects.
Frees up so much headspace

[lfarrand]

I revised my do list to prune out the completed items as well as add some new ones.

1. Set current time using NTP server (this is required so that I can timestamp the sensor data)
2. Publish sensor data to InfluxDB via HTTP using EthernetHttpClient library
3. Solder breakout board for the DS18B20 digital temperature sensors
4. Add voltage sensors to switching power source batteries
5. Add current sensors to switching power source batteries
6. Add voltage sensors to DC input of each MOSFET driver board
7. Add current sensors to DC input of each MOSFET driver board
8. Add voltage sensor via voltage divider to input capacitor(s). Use Zener to protect optocoupler.
9. Add voltage sensor via voltage divider to output capacitor(s). Use Zener to protect optocoupler.

I've pretty much got the software side nailed down now and I've got something working end to end.

I just need to add the sensors above, mount the various bits onto a more permanent platform and then take it for it's first test flight.

I'm itching to get started and see some meaningful results (good or bad), but I'm mindful about jumping the gun so want to make sure I get everything on my list completed before turning it on for the first time.