Next Gen DRSSTC

[Hydron]

Yes, QCW mode uses phase shift control

[Hydron]

A quick heads up for anyone trying to use QCW single-shot by setting the qcw_repeat parameter below 100. As the current github code stands, this has a bug that will result in full-CW operation (!!!) - you need to add a

Code: [Select]

qcw_reg = 1; after line 271 (https://github.com/Netzpfuscher/UD3/blob/master/common/ud3core/qcw.c#L271) in qcw.c, otherwise the handler that checks when the QCW pulse should finish is never called.

[TMaxElectronics]

I've been making some big improvements to the UD3's hardware in the recent months, and am about to release that version as the UD3.1

• LED outputs are now open drain, allowing use of 24V indicators
• Voltage sense amplifiers moved to single 5V rail
• Updated gate drive structure to allow more wide range of N&P Fet combo packages
• Added protection resistors to Relay and LED pins, added protections resistors in series with mosfet gates
• Added buck regulator for VDrive, adjustable between 10V-24V
• Added more indicator LEDs for debug purposes
• Added supply voltages to more expansion headers, external gate header can now directly support 24V gate driver modules
• The Design can now be almost entirely built with the JLC assmebly service. Only the psoc needs to be hand soldered
• Schematic cleanup... its now on more pages but each one is smaller

I Think the biggest change is the fact that it has now been optimized for jlc pcb assembly. Especially since the board got quite crowded since the last version



I still need to do the silkscreen layout, then I'll be ordering a first batch of test boards.

Here's the link to the github repo if anybody is interested: https://github.com/TMaxElectronics

[Hydron]

Great timing on the update - I'm in the middle of laying out a GDT board to be driven from a UD3.x controller, so while I won't be using your new PCB layout some of the schematic changes are certainly of interest. Once done and tested I'll be uploading my designs btw, though they are going to be pretty customised for my specific project.

The thing I'm most interested in is the GDT mosfets and their drive circuit:
- is there a preferred output P+N-ch mosfet you're aiming at and have tested? I've got some AOD609s and the good old FDD8424Hs, but those aren't always easy to find, and the power-SO-8 package style you're using here seems more standardised. I might see if I can hack a footprint up enough to allow compatibility with the SO-8 as well as the TO-252 styles in my design.
- What sort of gate threshold variation do you think requires the passive deadtime network? I'm assuming that gate charge variation between P and N could also have an effect too if the difference is large enough? I must admit that I haven't looked much at this part of the gate drive circuit, as with the FDD8424Hs it's basically just always seemed to work OK - my measurements and simulation have mostly been of the GDT itself and the IGBT drive network.
- Have you noticed any performance differences with 5V vs higher (e.g. 12V) UCC drive IC voltage? It's certainly a bit lower power (and easier, given 5V is already needed) to just run it on 5V and skip the intermediate supply between that and the IGBT drive voltage.

Finally a comment about the bipolar LEM current sensors - if all they are doing is looking at the current from rectified mains charging the bulk bus capacitors, then there may be an easier and cheaper way of doing this. If you use a 50/60Hz capable current transformer on the incoming AC line (i.e. before the rectifier) then you can rectify it's output using a small bridge and feed this into the input (with burden resistor), with no power supply required at all - as it's on the AC side then a simple CT is enough. If a voltage doubling rectifier is used for the bus then the burden resistor value set in the UD3 cli/gui should be entered as twice the real value though (OR the CT ratio as half the real ratio), as the CT measures twice the actual DC bus current with the doubling rectifier, so it needs to be scaled to the real value and this is an easy way to do it.

[TMaxElectronics]

is there a preferred output P+N-ch mosfet you're aiming at and have tested?

The point of that deadtime network is to allow using more or less any mosfets you can find. I haven't tried any specific one but built my gatedriver project that that network is inherited from with quite a few different ones without issue.

What sort of gate threshold variation do you think requires the passive deadtime network?

Not sure, but the old design relied on the 5V driver supply to not have shoot through problems. That apparently just worked with those particular fets. I just wanted to have more flexibility with parts choice for the next version.

Have you noticed any performance differences with 5V vs higher (e.g. 12V) UCC drive IC voltage?

No, since I haven't built a board yet
5V seems to be fine for the fets on the older UD3 model, but as i said I want to make the new version compatible with more devices. Possibility of using higher drive voltages is part of that.

Finally a comment about the bipolar LEM current sensors

Interesting Idea, but that gets a little cumbersome to do with three phase supplies. Not impossible, but just puttign a hall in between the rectifier and the bus seems a little easier to me

I also just prepared the order of the board, and including assembly of most smd components the boards cost less thatn 150$ for 5. I personally find that crazy, especially with all of the specialized non-basic parts in the design.
They are early prototypes and will likely have some bugs, but if anybody wants to have a board or two (without the psoc though!) just send me a pm and we can work something out.

[ZeroGee]

What wonderful changes you're making with UD3.1. I've been infrequently checking to see some updates on the UDx boards over the past several years. I've got a few of the UD1.3s back when they were hot off the press... I recently had an idea for a project that the UD3, and now the UD3.1 board would be perfect for.

... I Think the biggest change is the fact that it has now been optimized for jlc pcb assembly. Especially since the board got quite crowded since the last version ...

Here's the link to the github repo if anybody is interested: https://github.com/TMaxElectronics

I'm wanting to spin up a few boards via jlcpcb w/bottom assembly, and I don't think I see the BOM and CPL in the repository. Just checking if this is intentional. Is this something you're anticipating people to generate in Altium via the repo files? I haven't had Altium on my machine in ages, since changing occupation to one that doesn't need it...

[Hydron]

Finally a comment about the bipolar LEM current sensors

Interesting Idea, but that gets a little cumbersome to do with three phase supplies. Not impossible, but just puttign a hall in between the rectifier and the bus seems a little easier to me

I also just prepared the order of the board, and including assembly of most smd components the boards cost less thatn 150$ for 5. I personally find that crazy, especially with all of the specialized non-basic parts in the design.
They are early prototypes and will likely have some bugs, but if anybody wants to have a board or two (without the psoc though!) just send me a pm and we can work something out.

Thanks for the reply, I think it's probably a good idea to have a look at shoot through before ordering my boards - the FDD8424 is pretty well matched but the AOD609 isn't quite so close. Could also just add the optional parts if I'm not too lazy.

RE three phase input - yes that certainly changes the equation about using the LEM sensor, however I've never had the pleasure of living somewhere with ready access to three phase power, so I honestly didn't even think of that :'(

150$ for 5 is indeed crazy cheap, I really need to try the service at some point (though I'd lose the luxury of parts on both sides of the board). Are there many SMD parts they couldn't manage? I didn't find any P+N FETs for the output in their available parts list so I assume this is one of the ones you have to add? Though it would be easy to miss something with the terrible parts search system.

[TMaxElectronics]

I'm wanting to spin up a few boards via jlcpcb w/bottom assembly, and I don't think I see the BOM and CPL in the repository.

tbh I'd wait with that until I got back my version and fixed the unavoidable mistakes I probably made on the PCB ^^ The assembly files will of course be in the repo then too

... I've never had the pleasure of living somewhere with ready access to three phase power,...

Here in germany pretty much every houshold actually gets three phase power (we for example have 3x63A). Then again we have to pay like 35ct's/kWh now so it doesn't come cheap

150$ for 5 is indeed crazy cheap

I made a few optimizations and now its actually more like 100

(though I'd lose the luxury of parts on both sides of the board)

no they actually do that too. Just costs a bit more

I didn't find any P+N FETs for the output in their available parts list so I assume this is one of the ones you have to add?

nope, they have quite a selection actually ^^
In fact they have all the parts needed for the UD3 except for the psoc. No need to hand solder any other smd parts.
 
I suggest not using the parts browser from jlc because it is indeed horrendous. There is an open source website thats a lot better. Though the poorly maintained part parameters does still cause problems sometimes.

Of course there is the difference between basic and extended parts. Resistors, caps, diodes and the such like are pretty much all available as basic ones, but almost everything else on the UD3 is an extended one. Could certainly do with a bit more optimization, but at this point I don't really care to be honest. 100$ for 5 boards is cheap enough for me to just throw money at the problem
Extended parts eacht cost 2$ setup fee extra. When making more boards that makes less of a difference. Buying 20 UD3s would for example decrease the price to 15$/board including all specialized parts such as the drive voltage regulator. Without that it would probably get down to something like 10$

[TMaxElectronics]

Yay update time: the boards finally arrived

Soldering from JLC seems ok, though there were two dry joints, and quite a few crusty looking ones.




Looks pretty cool if I may say so myself

I Hand soldered only the PSOC and the through hole parts. Total cost was pretty high at around 40$/board but just because I ordered only two

Hardware seems to be working nicely, but a few issues are present:

• +5/+12V selector for hall effect labeling swapped
• UVLO threshold divider needs to be changed, to allow for gate drive voltages down to 10V
• gate discharge resistor in the GDT powerstage was too small, needs changing to 10k to keep the mosfet from turning off during a long on time
• mislabelled fan pinout
• slow response of th egate drive voltage regulator... causes about 3V of droop for the first 400us of a pulse. Some compensation network tunings still required

I will be running some tests with the small coil I am building at the moment in the coming week, once those are passed I will ifx the errors in the board files and publish the gerbers, boms and pick&place files. I'll also be purchasing some more boards then.

In addition I will probably also end up with a few leftover populated boards from the second order, that i will probably sell on my website.

We will also add some way for the firmware to figure out which board revision it is running on, so there won't be two different firmware versions we need to maintain.

[Kizmo]

This is super interesting project!

What kind of gate drive power this driver can support? My big coil (3.5 meters, ~30kW) is running some pretty big silicon, UD2.5 will handle it with 2 GDTs.

I have been looking at UD3 drivers for a while and it seems like PSoC availability has improved.

[Hydron]

What kind of gate drive power this driver can support? My big coil (3.5 meters, ~30kW) is running some pretty big silicon, UD2.5 will handle it with 2 GDTs.

Will support the same GDT drive power as UD2.x - the drive stage is basically the same (albeit with a different P&N FET footprint, though it is possible to make a combined one that will take any of the TO-252/Power-SO8 options). You could even double up the PCB and only populate the gate drive section if you wanted to use 4 GDTs. Each of the two outputs is separate btw, which allows for freewheeling, phase-shift QCW etc.

Note that in the case of the coil I'm building now I'm using the UD3 code but not the PCB, and have located the gate drive circuitry on another board much closer to the GDTs/IGBTs (GDTs are planar transformers built right into the board!), will try and post some pics of that shortly.

With JLCPCB assembly prices what they are it might make sense to go all the way to isolated direct per-IGBT drivers though (at which point you can get desat detection etc as well to detect if one of those big IGBTs fail) - these days GDTs may not be the best choice for a more advanced coil and I'm considering whether to change tack with my design. Needs much more PCB design work though, so not the right choice for those trying to do a simpler coil and avoiding PCB layout work.

TMax - quick question on this topic, am interested in the choice of 100R on the output of the MOSFET driver, it seems like a fairly high value, have you noticed any slow-down of the gate drive because of it? Or are you using fets with significantly less gate charge than the FDD8424Hs?

[FonziDaytona]

I’d be interested in buying a pre populated board.

-Matt

[TMaxElectronics]

What kind of gate drive power this driver can support?

As hydron said, the powerstage is pretty similar to other UD drivers, though the modifications I made to still need some verification. JLCPCB unfortunately doesn't seem to have too great of a selection of N&P channel fet combos and I managed to kill the ones I selected when driving four TO247s in cw at 200kHz... could well be the large-ish 100Ohm resistor causing large switching losses. Will investigate further.

(GDTs are planar transformers built right into the board!)

That sounds like a great idea. I've been wanting to try planar Transformers for a while but was always stopped by the layer requirements. I assume you managed to fit it onto just two or four?

With JLCPCB assembly prices what they are it might make sense to go all the way to isolated direct per-IGBT drivers though

Absolutely. Especially with faster and faster IGBTs (or even fets if I ever get my hand on some SiC parts) the stray inductance starts to become a real issue. The mini coil I'm building rn would kill the IGBTs if it wasn't for the TVS Diodes on the gates...
I've been meaning to play around a bit more with those all in one IGBT driver ICs, and even have a ready made board for driving bricks around, but haven't put it to a real test yet.

I’d be interested in buying a pre populated board.

Once I fixed the issues with the boards I will add the UD3 to my webstore as a pre-order to better judge the total demand and then order a batch. I'll post about that once thats happening. Might even make a jlc-assembly friendly version of the fibernet to go along with them.

[Intra]

Great to see the UD is lives on. Thanks, Thorben!

Kizmo said important thing. And I also should mention FDD8424H, which was good in 2009 and with kind of coils like in that past, in next two specific cases could be unreliable.
1. Some hard-gate-capacitance igbt like 1000A+ with freewheeling.
2. Some QCW like 50%+ of duty cycle with more than 2 bridges in parallel.

That is of course not a common cases, but I think that it will be good to get away from balancing on the probability that something not covering the continuous-wave-mode will fail by betting on the probability that it will not fail.

I understand, the goal of this was got packed with jlcpcb assembling, but if you have some bright thoughts about how to get this onboard, it will be nice to move this prj on this way.

My forthcoming QCW gate in attachment. TO-220 only because it has copper-plates to transport heat to radiator much more faster than fdd8424h. If there is something way to combine this case to jlcpcb assembling case it will be beautiful to add it to UD pcb prj, if you have interest to do that.

[TMaxElectronics]

Update time

I've fixed the issues I found with rev A of the board and will be releasing the new version in the coming days.

I'm especially impressed by the performance of the tiny gate drive voltage regulator. The new compensation has made it crazy fast and it now only drops around 200mV from a load transient of 0A-4A in 16us. It manages to supply around 7A peak at 19V, which is limited by the maximum dutycycle. Average I wouldn't pull more than maybe 65W due to thermal limitations, less if the input voltage isn't quite 24V or the ambient temperature is over 25°C. Oh also keep in mind, that the PCB gets to around 40°C over ambient under full load even with free convection. Could cause trouble if you have a 3d printed enclosure
Then again any IGBT setup that requires more power will likely not care about the added stray inductance from a 10:10:13 GDT and could thus be used with the VReg bridged.

Some firmware/psoc hardware changes also happened in the wake of my mini coil IGBT explosion debugging.
There was a glitch that could potentially cause the first four zero crossings after the start oscillator to be significantly wrong:

sorry about the scribbles in german and screen-photos, thats what I could find on my phone

This was caused by a zcd period of 0 counts being measured if there was a zcd pulse right after the first switch transition of a interrupter pulse. This now gets ignored.
Furthermore I changed the start oscillator to where it is now active indefinetely until enough current is detected in the primary resonant circuit. This threshold is adjustable and can thus be fine tuned to suit every coil. I have it set to 50A for my large DRSSTC and 25A for the small one.
The start_cycles now act as a limiter rather than the trigger to switch to feedback. If there is still not enough current after the number of start_cycles has elapsed the pulse will be stopped immediately and a feedback error thrown. You can also set how many of those can occur every second before a sysflt is triggered to protect IGBTs if the circuit doesn't start oscillating as it should.

I'm also about to make a version of the fibernet that is JLC-assemblable. I'll be releasing updated data of that once its done too. Oh and I might have figured out why that sometimes didn't boot after a long runtime... I was missing some pullups on I2C datalines that aren't in use, but it turns out they might be required anyway.

And finally:
I'll be accepting pre-orders of UD3s starting today. If you want one of the first batch please place a pre-order now, so I know how many boards I'll have to buy
It is going to come partially assembled: SMD components all soldered on, THT for you to solder
Link (if it's ok for me to post it in a non-marketplace thread): https://tmax-electronics.de/product/ud3-1b-kit-preorder/

[Hydron]

Looks good - I'd certainly noticed some nasty switching early in a burst, though I'm not sure whether it was due to this bug (I assumed it was from mismatch between start_freq and actual resonant freq). Will certainly need to give the fix a go myself.

Are your changes in the digipot branch of Netzpfuscher's repo?

[TMaxElectronics]

Are your changes in the digipot branch of Netzpfuscher's repo?

jep, but you might get some weird behaviour with the bottom exp header, as there are the connections for the digipot controlling the VDrive regulator on it. You'll also want to change ivo_led to 1 after installing or the LEDs will be the wrong state. Currently the version detection is hardcoded to always detect a new UD3 as we haven't actually implemented that yet (rev b will have a 5 bit binary coded version indicator on the pcb.. resistors to be precise ), though that should really only affect the default states of the LED and the exp_header behavior.

[TMaxElectronics]

TMax - quick question on this topic, am interested in the choice of 100R on the output of the MOSFET driver, it seems like a fairly high value, have you noticed any slow-down of the gate drive because of it? Or are you using fets with significantly less gate charge than the FDD8424Hs?

I've actually just verified this, the gates to be turned off are discharged in around 20ns and the ones to be turned on charged in 100ns, both not very fast but in my opinion perfectly ok. I just really wanted to make sure there was enough deadtime between the two mosfets. Due to the low threshold the mosfet also already starts to turn on after only 40ns, again perfectly ok in my opinion.

I have actually changed to some TO252-4 mosfets for the rev.b. but might revert that change due to their probably slightly higher gate charge. The SO-8 Mosfets seem perfectly ok now that I'm using some with a power-pad to dissipate the heat.

[TMaxElectronics]

Good news (no not the dacia sandero),

The Rev. B PCBs have arrived and are working well. I'll be shipping out the boards to those who pre-ordered during the next week or so (once I had the time to solder on the PSOCs).
I've also pushed the latest Gerber files to GitHub if you want to order the PCBs yourself.



The V2 of the fibernet that is JLC-Asssembly optimized still needs some work with sizing of the footprints, but is working well otherwise too. I'll be releasing the gerbers for that too, but you will have to hand solder the PIC aswell as the LAN9250 (the mac chip).

[TMaxElectronics]

I made some more small additions to the UD3.1.

First of all I added a resistor to the IBus input filter to add a voltage divider to that input. This now allows using Voltage mode current sensors, that output more than 5V. I also added a solder jumper that means one no longer needs to de-solder the smd shunt for use with the voltage mode sensors.


I also removed the little 5V regulator daughter board and replaced that with a smd 5V regulator. That way you don't need to solder that anymore and it will come pre-assembled when using the SMD assembly service. It is a little bit more expensive, but I think it's worth the reduction in soldering effort.

I might also just remove the position for the vdrive THT trim-pot, as the digiPot is working very well and I doubt the trimmer would ever be used. This will free up the space necessary to fit the 5V regulator

I'll be releasing that version as the UD3.1c once I verified the changes. The next batch of UD3's in my shop will be that version.