Update 2021-03-01: New pictures added further downUpdated bom 2021-02-24: https://docs.google.com/spreadsheets/d/1ThZLzR7Nggm0Z4HKQb2HZf72yWK6yGwCSyFWkue1RE0/edit?usp=sharingPictures (edit: out of date, see thread for changes)! https://imgur.com/a/tWysv8mSome of you may remember this project that got some attention a couple years ago:
https://habr.com/en/post/451246/I searched and while it's entirely possible I missed something, I couldn't find any evidence that someone other than the guy himself had actually built one. Therefore, I decided to try and be the first (maybe?) I can do up a more detailed thing if people want me to, but for now I'll just dump the major points.
Nothing but praise for the man who built it, and he isn't beholden to anyone to make his design easy to replicate, but it's certainly not a polished, easy-to-follow build. There are a few strange parts you can't buy in most of the world, some mistakes in the CAD, and the BOM is quite large and varied for such a small project. It comes across like it was designed for the specific set of random parts the guy had kicking around, which is totally fine and I won't blame him for it, but it did make replicating it from the opposite side of the world a challenge.
The first thing I'll say is if you live in North America, Japan, or somewhere else with a lower AC mains voltage, don't even bother trying to build his AC-DC converter circuit. He used a design that is intended to run off of 180VDC or higher from an active PFC, but he cheats and just uses rectified mains. If your rectified mains won't reach 180VDC, you're out of luck. I just left that entire part of the board unpopulated and used off the shelf supplies to provide the 30-40V and 10-11V inputs that the design needs. This had the bonus of eliminating like 35 of the 100+ unique BOM items.
Here's my BOM (lacking some components that went unused due to the above):
https://docs.google.com/spreadsheets/d/1ThZLzR7Nggm0Z4HKQb2HZf72yWK6yGwCSyFWkue1RE0/edit?usp=sharingThe second challenge is that I could not source the exact same cores and magnet wire as he did. Two of the inductors and the current transformer are built with "K16x8x6" cores that as far as I can tell were only made and are only found in the former eastern bloc (I substituted more standard T60-26 which are of similar size). It also seems like the cheap T130-6 iron powder cores he bought must have less permeability than the ones I bought. Finally, he uses magnet wire specified in millimeter diameter while I can only find magnet wire in AWG.
Update 2021-03-01: The T60-26 cores worked okay until they fried. Don't trust Amazon listings--they said they're ferrite but they were iron powder! The cores that Sergey originally used were K16x8x6 М50ВН (careful! That's Cyrillic, not Latin! "M50VN" can sometimes find them) which are readily available in most of the former east bloc but not easy to get in the US. Thankfully I have a Ukranian friend who arranged to ship me some. I'm currently planning on trying to find a Fair-Rite/Amidon or ferroxcube substitute.
Note that Amidon is a reseller: Their Iron Powder cores are mostly made by Micrometals with the same part number, while their ferrite cores are made by Fair-Rite, but with different part numbersI used 22AWG magnet wire for the smaller coils and 16AWG for the larger. Note that since the holes in the PCB are made for the millimeter-sized ones, this is a tight fit (it almost doesn't fit at all with the enamel still on the wire) and you might be able to get away with the next size down. I tried just following his winding count but the resulting build cooked the main power transistor and I ended up pulling out all of the inductors and re-winding them with a borrowed LCR meter from work. However for the current transformer, you should just be able to follow his turn counts since it's only the turn ratio that matters.
Update 2021-03-01: 20AWG also fits for the two drain chokes. I used it for L6 on my second unit because that one carries more current. On my first unit I ended up using TDK/EPCOS solenoid VHF chokes, which worked alright but I think the open magnetic circuit caused all the glitchiness I experiencedUnless you can get the exact inductance measurements that are shown on the schematic, you will have to revise the sixth order (I think it's sixth order? pls correct me if I'm wrong) LC filter by changing the capacitance using the measured inductance of your coils. The LTSpice file for the simulations I ran is attached. There are basically only two things you care about in the frequency response from what I can see:
- The operating frequency 13.56 MHz should not be too close to a resonant peak and should be boosted about 5-6 dB Note: as simulated with an ideal source and load! Not measured with a VNA
- All harmonics of that frequency should be attenuated at least 20 dB (27.12MHz is the first one, and unless you royally screwed up, each successive harmonic should be more attenuated)
I used 1% resistors and 10% capacitors for most of the circuit because there was not a tolerance spec on most of them--but if anyone wants to do a tolerance analysis and come up with a guide for a lower-cost BOM, that would be great!
Some guidelines for part purchasing that may not be obvious (and other stuff):
- The variable buck converter that powers the RF section runs off of a 30V input in Sergey's design, but can go up to 40V. I used 36V. It outputs 22V at up to 4.5A so your input should be >4A capability
- L1 (the aforementioned buck converter's main inductor) is specified as a Yageo part that, as far as I can tell, is not orderable and may in fact be discontinued. I did my best to try and match its specs but without knowing the actual requirements I could only go for 'at least as good' which due to the tight space it needs to fit in, left me buying a Wurth surface mount inductor that you can just barely bodge onto the thru-holes on the top side of the board
- The "+10V" supply can go up to I think about 11.5V, but for reasons I haven't quite figured out the system will start to misbehave if you try to go closer to the 14V maximum of the MOSFET driver.
- For U9, the part specified is just the one Sergey could most easily get. In that same family are other parts with different channel inversions, and since the board only uses one channel it
doesn't matter which one you useYes it does! Q1 gate must be LOW when the system is in a disable state - Metcal handpieces use an F connector. The connector on the handpiece lead is a slip-on clamp-down type and it may fit better on some F connectors than others.
- Because of how close Q1 and Q2 are to other components, some heatsinks might be too wide (like the ones I bought).
- Q1 and Q2's tabs must be insulated, they cannot be grounded. 2021-02-03 -- recommend connecting a nearby ground to Q2's screw to shunt the capacitive coupled noise that will otherwise return through the housing and output jack
- I did manage to find somewhere to buy the WinStar OLED that he mentions using, but it was expensive and shipped from Poland. Luckily, the 16x2 LCD display that Adafruit sells will connect right up with the same pinout. You can omit the middle four pins on the display header (D0-D3) as they are unused. Note that the microcontroller operates at 3.3V so other 16x2 displays that run on 5V may not work, but the adafruit one seems to work just fine 2021-02-03 -- I managed to make a significantly cheaper, though not as cool display from BuyDisplay/EastRising work. It requires FW changes which I created a PR for. Note that you can't just run a display off of the 3.3V regulator in the design, as the board layout just isn't made for it. I ended up adding a second one just for the display.
- For an enclosure I got the BUD Industries CU-476, which has similar dimensions to the enclosure Sergey used (which is not easily available in the US). One major difference is that it has a center screw post on both sides which you will have to cut away to fit the display.
- As far as I can tell there is no good reason for C4 to be thru-hole. The requirements are only modest with respect to voltage and stability. I bodged in the same X7R SMT cap used elsewhere in the design
- The board fits tightly in the enclosure, but it has no mounting features. You will have to make sure to sand off any high spots on the edges or it won't fit. I just put rubber feet on some safe spots on the bottom of the board and put kapton on some parts that stick out. I haven't finished putting it in the enclosure but will probably add some makeshift way of keeping the board in place
- The thru hole LED footprint (D13) will not fit a 'standard' full size LED. The holes are too small
Some other random things to look out for:
- Going by the blog post, L8 is intended to be adjusted by hand (by manipulation) "for maximum efficiency". I assume this means you are trying to maximize the slew rate on Q2's gate, but I am not sure. It is also a bit difficult to manipulate due to being surrounded by other components. I would recommend stretching it out as far as it will go before installing it as I think compressing it will be easier than trying to spread it. 2021-02-03 -- once I actually had a way to measure efficiency I found that it makes a difference, but not a game-breaking one--75% vs 80% eff or something like that
- Some of the refdes are obscured, both on the schematic and on the PCB. Beware!
- L7 and L8 ended up being closer to 10uH for me, I don't know if missing 9uH high or low is the safer option, but my setup seems to work. 2021-02-03 -- much higher is almost certainly ok.
- The firmware still builds fine with the free version of embedded workbench. I can try building a hex file if people would rather not install that (d'oh, there's one in the repo). A random ebay knock-off STLink programs it just fine
- The blog post recommends bodging the crystal load caps directly across the crystal and running a ground wire across to the cap on the STM's ground pin. The pictures also show additional capacitors bodged in parallel there--I have no idea what value those are but my board works fine without them.
- For the capacitor banks in the LC filter, you want to use as many caps in parallel as you can. I used C0G/NPO tight tolerance caps just to be safe. I only ended up changing one value: the first bank from 600pF to 496pF by removing two 100pF caps and replacing them with 47pF
- The values for R6 and VR1 seem wrong unless I'm missing something. You would not be able to reach the nominal 22V as they are. I subbed 22k for R6 but using a larger value pot would probably be the better choice. 2021-02-03 -- I think that 10k works, the SWR feedback circuit affects the node as well
- In his photos he installed L12 raised up off of the board. I'm not sure why. I didn't do this and I do see some erroneous "tip failed" messages sometimes. 2021-02-03 -- was almost certainly unrelated
And improvements that could be made to this design:
- The refdes are in no particular order. They should be renumbered
- Porting to KiCAD would be nice. Diptrace is not very popular and kinda sucks.
- There are probably some reductions to be made in the unique component count
- A lot of the SMT components are 0805 for no good reason. Even just shrinking them to 0603 would free up a lot of board space
- The AC/DC converter should be replaced with a more standard and universal design or just eliminated and the board shrunk and designed to run off an external DC supply
- The design would probably be a lot easier on a four-layer board
- display footprint doesn't fit 100 mil headers
- Pin 1 mark missing on a lot of the headers (notably the programming header).
Let me know if you want more detail! I'll probably reply to this thread with pictures and videos and more stuff. Also, if I missed anything big or said something incorrect please correct me! This is definitely out of my comfort zone, I just did my best.
2021-03-01 -- These pictures are out of date, see the top of this post for newer ones. I ditched the cheap power supply and used a decent quality 36V 4A brick. When using an external power supply, you do not want the housing to be tied to DC- at the input, as that causes a ground loop (which is not present in Sergey's design). In my current setup I have a separate mains earth wire running to the housing. Housing front
Housing back
Inside
Board top
Board bottom
Power supply (cheap chinese amazon special in a random box. Yes that is a macbook power cord)