create a high speed pulsed (greater than 20 MHz) 5 Amp current source

[super7800]

so I'm trying to create a high speed pulsed current source to test some current sensing magnetics. I created a prototype that works up to around 750 KHz, but has substantial issues beyond that. At this point I think my approach is flawed. I was hoping to kind of "jurry rig" a solution as accuracy isn't an issue, as the ultimate application is for ultra fast overload current detection, not for taking usable current measurements. There also is no load.

Important:
-high speed (>20MHz)
-5 Amp or greater

Not important
-accuracy (+-1MHz stability, +-500mA etc. is OK)
-"clean edges"

my solution was two of these circuits in parallel. there are two used so that the current source (keithley 228 in this case) always "sees" a constant 5 amp load. driven with inverted square waves with some overlap. Results are verified using a TCP 202 Probe.


The issue is that at higher frequencies (in this case 15MHz) , the mosfets heat up (with big heatsink cant run for more than a few minutes), and the current waveform (measured by the tcp202 and my sense element) is not pulsing, whereas around 750KHz (around the maximum where the output is usable) it works fine. The mosfets also die at these higher frequencies seemingly at random. The gate drive is working fine, I measured and verified that.

So my question is how flawed is my approach (and can it be bodged to work?), and if my current approach is hopeless, what might be a better overall "architecture" to accomplish this goal?

This topic discusses a similar project, but their goals include accuracy, and a fairly complex design. I was hoping for something simpler https://www.eevblog.com/forum/beginners/ac-constant-current-source/

[PartialDischarge]

Keep in mind that most of the specs of a mosfet will get worse as temp rises, specially Rdson which means more losses and more heat.
Even lower frequency active loads use mosfets in parallel, so with your setup you could estimate how much current you can switch at what max frequency and then calculate how many mosfet you are going to need.
It is not an easy project, many trial and errors for sure.

[redkitedesign]

Some things to try:
- Vary the current that is switched. If the current is low, Rdson losses will not be relevant at any frequency.
- Is there any inductance in the high-current path? Add some resistance to at the current source (to prevent it from eating the voltage spikes and measure with a scope
- Is there a ground bounce? The circuit you've posted doesnt show the FET's sources being at ground potential. That can really alter your gate voltage
- Is the FET actually switching as fast (edges!) as it should?
- is there garbage (ringing etc) on the gate at higher frequencies?

[super7800]

Thanks for the ideas @redkitedesign. Ive tryed both ways with the sources being at ground potential and without. seems to make little difference. Ill try your other suggestions and report back.

Attached are some waveforms. yellow/blue are outputs of drivers, green is tcp202, and purple is the sense element (ignore this trace). There is some noise on the yellow/blue. this is primarily due to poor probing (i suspect).

[Terry Bites]

For this kind of speed, check out SiC and GaN fets.

The current source will not be able to respond to the step load at these speeds.
The current will not be very constant (if at all) because the internal current control loop will not have the necessary bandwidth.
A current source should offer a high output impedance. It needs to be >> higher than the load to get any meanigfull result.
I had a quick RTFM of the 228, it's not designed or specified to operate into an inductive load.
They specify a capacitive shunt impedance of 200nF, if so the output impedance @ 20MHz is only 40 milli ohms.

My thinking would lead me to using a switching a fixed DC voltage into the magnetics via a reference inductor.
That will appear as a current source to your load. You'd likely use an RF CT to monitor the current.

[super7800]

sic and gan fets are nice, if you like soldering BGA . Im getting around the current source response by using the two sets of fets, so to the current source there is always a steady load present. Measuring this, according to the tcp202 probe the total current drawn is always what the 228 is set to. its not really an inductive load (of course its inductive, but enough to matter?), is it? its at most 6-8 inches of PCB trace + the connecting leads. Could you expand on "My thinking would lead me to using a switching a fixed DC voltage into the magnetics via a reference inductor"? thanks.

[Weston]

What are you requirements here? For testing a current sensor I would assume you want a known waveshape. A bit surpised that the waveform shape is not important. In the limiting case is a sine wave ok?

Does frequency need to be continuously variable?

Do you need the dv/dv on the sensing loop to be limited?

If you don't need a high duty cycle, a fast GaN fet (+flyback diode) pulling some low inductance power resistors and a minimally sized sensing loop would work. Size the resistance and supply voltage to achieve the current you want with a small enough L/R time constant. You can change the current by varying the supply voltage.

If a sine wave is ok and you don't need a continuous frequency range a tuned LC circuit can achieve a high circulating current with little drive power. 

[redkitedesign]

I see quite some sawtooth-like shapes in the green trace.

If its the current, inductance is determining the current.

I suspect spikes on the drains.

[Marco]

Since you can compromise on the edges, have some overlap in the on time so the MOSFETs spend less time avalanching?

Do a sanity check current measurement of the common leg too. If it's fluctuating too much, maybe add some inductance on the common leg?

[moffy]

BJT's might offer some advantage in switching because they require less voltage swing on the base, also a differential pair can be used as well as a cascode transistor to isolate the load from the switching transistor, which reduces miller feedback. The example I show illustrates this, hope it might help.

[jonpaul]

simple idea...

Use fast pulse gen, 50 Ohm load, and series coil to fit the CT or DUT eg 100T #22 Insulated solid wire in 35mm square loop.

With 100 T the AT is multiplied x100 so effect of 500 ma on CT is achieved with just 5 mA in the loops

Just like a power mains splitter for mains current.

résultats....no amps or FETS, just a good pulse gen at desired freq

Your thoughts?

Jon

[TopQuark]

Perhaps you can take a look at my project for some ideas (https://www.eevblog.com/forum/projects/gt1mhz-bandwidth-electronic-load/) . I can get 0A to 50A steps in 280ns (slightly more then 1MHz) BW. No where close to the 20MHz you want, but I think with a smaller, easier to drive MOSFET you might get close to 10MHz? One thing I learnt from this project is layout and probing technique is absolutely critical.

Also, according to the TCP202 manual page 19, (https://xdevs.com/doc/Tektronix/Probes/TCP202-Current-Probe-Instruction-Manual-070954203-RevA.pdf) you might not want to use the probe to verify the 5A 20MHz waveform, as the probe is derated to 2-3A peak at 20MHz.

At 20MHz, perhaps it makes more sense to use LDMOS FETs designed for power RF amplifiers, rather than power MOSFETs.

[TopQuark]

its not really an inductive load (of course its inductive, but enough to matter?), is it? its at most 6-8 inches of PCB trace + the connecting leads.

Loop area and thus inductance is absolutely critical, I had to reduce the distance between the power source (High power RC Lipo battery) and the load FET to 10-15cm to get clean-ish results, and that's for ~1MHz. Take a look at Jim Williams app note at page 6 (https://www.analog.com/media/en/technical-documentation/application-notes/an133f.pdf) for some understanding on the issue.

Power supplies are usually low impedance only for frequencies lower than 10s or maybe 100s of kHz. Above that the closed loop gain of the power supply are deliberately cut down, and you are at the mercy of the output caps for the higher frequency power delivery. That's before you add long wires between the supply and load. Just saw that a current source is used, my comments are not very relevant here.

[PartialDischarge]

At 20MHz, perhaps it makes more sense to use LDMOS FETs designed for power RF amplifiers, rather than power MOSFETs.

Can these be used in pulsed operation? The DS always mentions CW modulations...

[TopQuark]

At 20MHz, perhaps it makes more sense to use LDMOS FETs designed for power RF amplifiers, rather than power MOSFETs.

Can these be used in pulsed operation? The DS always mentions CW modulations...

Not too sure ether tbh, haven't used them in this manner. But regular power MOSFETs really struggle at 20MHz, thats for sure.

[Terry Bites]

I got that wrong. Thinking about another thing there.

I understood that you were testing magetics directly from a CCS, but the load is not really magentic to any extent?

Switching from a dummy load to the test load is a smart idea if the impedances match.
So the CCS seeing a very fast transient that it just ignores. Cool.

I'm not an RF warlok but LDMOS FETs do look interesting.
DS do show tests made under pulsed operation.

[nimish]

You need a Class D/E/F/DE/EF/phi RF amplifier tuned to whatever frequency you want, with a decent power FET. Even shitty mosfet's will work if their output capacitance is ok, and you'd actually want some in a D/DE half bridge design.

For high voltage, hard to drive FETs you'll also likely need a resonant gate driver which is a mini version of the RF amplifier you are building, but just for the gate of the power FET.

this dude's thesis has some pointers: https://etd.ohiolink.edu/apexprod/rws_etd/send_file/send?accession=wright1546870469456974&disposition=inline

"Resonant Wireless Power Transfer" will also have good info since that works at 13.56MHz, which is close enough

[Marco]

He wants something more akin to an inverter than an amplifier, but the class E/DE resonant switching inverters swing to a constant voltage, not a constant current.

There's probably some smart way to convert the topologies for constant current, but I think just overlapping the on times of the FETs will help a lot. It's probably the avalanching destroying them.

Switch hard enough and keep the current on the common leg constant, she'll be right. If it doesn't work, just means you have to switch harder.

[nimish]

He wants something more akin to an inverter than an amplifier, but the class E/DE resonant switching inverters swing to a constant voltage, not a constant current.

Potato, potahto, one man's inverter is another man's amplifier. The wireless power people seem to use the topologies interchangeably for what its worth.

[super7800]

thanks for all the replies!

@TopQuark for the jim williams design note. Seems their approach is similar to mine, but more advanced. I think ill take yours and jim williams and make a prototype off of that with a battery as a current source.

@weston sine wave is fine but not desirable. The sense element sucks for overall accuracy, but i'm hoping it works at very high speeds, so that's why absolute accuracy doesn't really matter.

@marco sadly overlapping the mosfets on-time is not (easily) possible. I have 2 Agilent 33220A function gens, but those sadly don't have the option to sync installed. Right now i'm using the inverting and non inverting outputs of an hp 8082A pulse generator.

its looking like the mosfet is not switching fast enoughph (see attached waveform + test schematic). yellow = Diff probe, blue = output from driver, green = tcp202, purple = ignore. Honestly i cant seem to figure out why. The parts are (according to their datasheets) rated for this speed. any ideas?

[TopQuark]

with a battery as a current source

When I mentioned battery as a source, I meant very high power Lipo packs for RC use, not the garden variety AA batteries. You'd be lucky to be able to source 5A out of an alkaline without the voltage dropping to nothing, but high power density RC Lipo packs can deliver >100A with maybe 10-20% voltage dip, those are the ones I used for testing.

Just note that these batteries can puff up, catch on fire or blow up if you short it with a low enough impedance path. Make sure you are somewhat confident your circuit behaves properly before attempting Lipo as a source.

edit: When testing with high power Lipo as a source, inserting a switch between the battery and the load may melt the switch, fuse are preferable but may add too much series resistance and finding one that will handle tens of A can be hard, grabbing a flaming pack to unplug it is not very OSHA friendly. I prefer to just have a side cutter in hand and snip the battery leads when things goes really south. YMMV.