Lab supply using Lateral MOSFETs

I tried vertical FETs for this application in this unfinished project:
0-70V, 0-5A Lab Power Supply Design - Page 2

From an SOA standpoint, they're great at these voltages. Not so great is trying to get them to current share without huge source resistors. They really do need to get matched, and it really is a tricky game.

Ultimately, I came to the conclusion that short of lateral FETs (which are NOT easy to get here in the States and are very expensive), about the best thing to use are regular old BJTs. I used NJW21194s because I got a really good deal on them (I bought 30 of them for $.80 each) but there are lots of other economical choices.

Ultimately my project got shelved when I got distracted / busy with other projects. Well, that's how it goes sometimes. I have all the boards made, I just need to order the transformers and stick everything in the box.

In my case I got a pair of Crest CA9s that were butchered by another tech. I managed to get one working amp (using parts from both) and the heatsinks and chassis from the scrapped one will end up being used for this power supply when I get around to finishing it.
 
Hi Suzyj,

I really admire your power supply design!

If matching MOSFETs remains problematic, you might consider driving each FET with its own opamp to servo out the variations, as sketched below.

Best,

Steve
 

Attachments

  • thumbnail_IMG_1806.jpg
    thumbnail_IMG_1806.jpg
    185.5 KB · Views: 248
Okay, I think I'm converging on a design that I can build...

I've swapped the output devices for BJTs (2SC5242, as I have a pile of these in the 'O' sorting, with pretty good beta). I just used one of the six as the driver. This arrangement simulates well. Output resistance is approximately equivalent to the vertical MOSFET case. Setting the output to 50V and connecting a 5A load, I need just 51.2V on the input.

I also did a board re-spin to knock the length down a little, as I'd like to build these into a 1/2 width 19" chassis, so they need to be shorter than 210mm. The board is now 200mm x 70mm. That necessitated going for slightly smaller diameter main capacitors (30mm vs 35mm), plus substituting a different (smaller package) bridge rectifier.

Essentially dropping one output device means each device now has to dissipate 20% more power. Worst case (80V in, output shorted, set to 6A), that's 96W each, which is simply excessive. So then I thought, hey, I can add another 4 output devices, by sticking them underneath, but I'll need to use SMD emitter resistors. So I did that. Current per device now drops to 670mA, and max dissipation is now just 53W, so much betterer.
 

Attachments

  • Screen Shot 2021-01-28 at 6.55.54 am.png
    Screen Shot 2021-01-28 at 6.55.54 am.png
    961.9 KB · Views: 146
  • Screen Shot 2021-01-28 at 6.56.15 am.png
    Screen Shot 2021-01-28 at 6.56.15 am.png
    657.6 KB · Views: 102
  • Screen Shot 2021-01-28 at 6.56.38 am.png
    Screen Shot 2021-01-28 at 6.56.38 am.png
    570.7 KB · Views: 201
  • Screen Shot 2021-01-28 at 6.56.54 am.png
    Screen Shot 2021-01-28 at 6.56.54 am.png
    529.6 KB · Views: 252
  • Screen Shot 2021-01-28 at 6.58.04 am.jpg
    Screen Shot 2021-01-28 at 6.58.04 am.jpg
    352.2 KB · Views: 243
I've ordered 5 boards. I also threw together a design for a five-digit unipolar DVM (0-5V input), as I just couldn't find anything at a reasonable price that's do that online. I'm hoping to display voltage and current with mV/mA accuracy. that means 60,001 counts (0-60V) for voltage, and 6,001 counts (0-6A) for current. The usual 4.5 digit DVM only does up to +19.999, which won't quite do the deed.

Anyway, I judge a combination of MAX6350 reference (~1ppm), plus 24 bit LTC2400 ADC will get me there. I've used a PIC16F876. Despite my frustration in programming this guy for my latest preamp, I have the tools and programming hardware for it, so I know it'll do the job.

As I don't need the fifth digit for current, I also knocked together a 4-digit version, substituting a cheaper ADR02 reference and LTC2420 20 bit ADC, and using ceramic caps to filter the Vref rather than the expensive MKS film caps. It saves about $20 on the BOM, and is sized appropriately.

I also further tweaked the regulator board, using Molex Micro-fit connectors for all the low current stuff just to clean things up. I've upgraded to KiCad 5.1, and the renderings from this are nothing short of superb. Indeed everything about KiCad 5.1 is just incredible. It's a pleasure to route with.
 

Attachments

  • Screen Shot 2021-02-02 at 3.49.12 pm.jpg
    Screen Shot 2021-02-02 at 3.49.12 pm.jpg
    278.5 KB · Views: 283
  • Screen Shot 2021-02-02 at 3.43.00 pm.png
    Screen Shot 2021-02-02 at 3.43.00 pm.png
    846.9 KB · Views: 282
  • Screen Shot 2021-02-02 at 3.42.13 pm.jpg
    Screen Shot 2021-02-02 at 3.42.13 pm.jpg
    116.2 KB · Views: 293
  • Screen Shot 2021-02-02 at 4.43.24 am.png
    Screen Shot 2021-02-02 at 4.43.24 am.png
    241.7 KB · Views: 301
Well it went together nicely.

I might want to find a slightly shorter heatsink for the regulators, as the 38mm tall one is going to make access to one of the mounting screws for the output transistors a bit of a pain. The current sense shunts are skinny little blighters.

I have no heatsink as yet, so I thought I'd just apply power to the 12VAC input from my old Thurlby lab supply, and probe around. +/-12V rails come up nicely. It draws 48mA from the +ve rail and 41mA from the -ve rail, which smells about right, and the 5V reference is, well, 5V, near as I can tell with the 3478A. I'll need to borrow a better multimeter to really set it up accurately.

I've been a bit obsessive about accounting for temperature coefficients of the various resistors, plus of course the reference, in the hope I can get true 5 digit accuracy in voltage and 4 digit accuracy in current. I figure if I'm going to build a supply, it should at least be settable to millivolt and milliamp accuracy (unlike the old Thurlby, which is pretty approximate). Not metrology grade, but not inaccurate either.

I need to buy or make some adapters between the non-shrouded sockets on my 3478A and my shrouded leads. I refuse to cut the shrouds off.

So far so good!
 

Attachments

  • IMG_1676.jpeg
    IMG_1676.jpeg
    369.9 KB · Views: 255
Last edited:
Good and bad news.

Firstly the bad. The idiot who drew the schematic wired the current gain feedback network to the wrong side of U8. Did I mention she's an idiot? Cut tracks and wire links are required to get the prototype going. I'll have to do a board respin.

Good news: Once it's going it goes. The voltage stability is really pretty nice. I put 1V into Vset, and I get 19.9999V out (Vout = Vset * 20). And it stays there.

I'm wanting another 3478A, as my Fluke 87 is missing a couple of digits and disagrees with the 3478A by 1 LSD.
 

Attachments

  • IMG_1678.jpeg
    IMG_1678.jpeg
    413.4 KB · Views: 174
Member
Joined 2004
Paid Member
suzyjAlso said:
That's why the vintage analog instruments had switches/number wheels (Lambda LR-613-DM Precision Power Source | eBay) and the current instruments have keypads. A 10 turn pot may give you better than 1 part per thousand setability but you are looking for more like 10 PPM setability. I have an HP supply/amp (HP467a https://www.pearl-hifi.com/06_Lit_Archive/15_Mfrs_Publications/20_HP_Agilent/HP_467A_Op_Serv.pdf ) with a novel coarse/fine pot setup where the single knob controls both with the fine tweak at its end moves the coarse pot. It works fine (the supply is 50+ years old) but I have no idea if anyone knows how to make such a pot today.
 
I’ve added extra complexity to make it so I can control it from a 0-3V voltage, both for current and voltage, so if I want to do a front panel with keypad and DAC I could. I like the lack of fuss that simple pots give. I figure I can get within 50mV with the 3 turn pot (0.9 degrees), so if I have a second single turn pot with 200mV range, that should allow close to mV setability.
 
Good and bad news.

Firstly the bad. The idiot who drew the schematic wired the current gain feedback network to the wrong side of U8. Did I mention she's an idiot?


I would NEVER make a mistake like that! Not once. You'd have to be a total child-like brain-dead IMBECILE to do something like that!


Stares nervously at box of extra prototype PCBs with layout mistakes


All kidding aside, what is the transient response like in this design? It looks really nicely done.