(Now with pictures!) I built Sergey's 80W metcal-compatible soldering station.

[quadtech]

That's really compact - if you could make 100mmx50 , you can fit two
of them in usual 100mmx100mm China PCB fab size.
Eagle freeware would be awesome - I could help as well, as I am more comfortable with Eagle.
I guess the 2nd output would need firmware changes, unless it is switched so only one is active at any time.

[t0m]

As an aside, I recently bought a SP200 power supply (SP-PW1-10) on fleabay for $40. I'll start a new thread with pictures of the insides. It'll be interesting to see the differences. I suspect it is slightly easier to design a 470kHZ system than a 13MHz one.

[rfmerrill]

If someone can port this design to KiCAD that would be really cool... so far we have one design in Eagle and one in Diptrace, but none in KiCAD

One thing I think could really be improved is better decoupling between the RF and digital sections, and in general better grounding (like you have a thin trace between two ground pads on the bottom layer that are both just attached to a ground flood on the top layer... wtf.

For a particular example, on the bottom layer around the uC, the microcontroller's one ground pin is located on one side, connected to the top layer via a trace that snakes up to the 5V regulator and cap, and also a thinner trace that goes down to the display pcb header. You can see in these scope measurements why I think that's a problem:

(click for gallery)

Path to from GND pin to the top layer ground fill. Signals on the opposite side of the chip are referenced to a bottom layer GND fill that is only distantly connected. Note particularly the BOOT pin (pin 1)


Probing between "ground" and "ground"


resulting scope trace


I noticed this issue when I was trying to figure out why the display power indication was jumping all over the place despite measuring a constant power output.

[t0m]

Ugh. It looks like 2 main frequencies in the noise. Presumably one of them is 13.56MHz?

I'll see what I can do with the layout. I haven't used KICAD, only gEDA and Eagle.

[rfmerrill]

Bit of a false alarm: noise doesn't show up nearly as high when I use a low impedance probe (coax probe) so I think it might have been coupled noise from the RF circuitry into the probe.

Got some better 3.3V measurements and I'm fairly certain it was just noise on AVCC from the OLED display. I think I might just stick down a second 3.3V regulator rather than running the display off the same one as the uC.

[rfmerrill]

Yeah that was the ticket. Display is a lot smoother now. I put the LD2985BM33R back in place and used the MaxLinear SPX part to just power the display, and now the power meter output is very stable.

So yeah, I think that's the way to do it if you're going to use a 3.3V display. If anyone is spinning the board maybe put a stuffing option for a larger 3.3V regulator with better routing to avoid that noise?

Also to t0m and Prasad and anyone else who may spin a board based on this design: The display header has a 5V supply but 3.3V IO. This is technically not supported by most displays, but you get away with it with the adafruit display. If you want to use a 5V display adding a level shifter might make it more reliable. If you want to use the BuyDisplay/EastRising OLED that I'm using, you might want to change the pinout otherwise it will require a lot of cross-wiring.

[t0m]

I'm not clear which is the MaxLinear part, is that on your BOM?

The display is certainly part of the design that needs sorting out. Can you post a link to the OLED you used? Can we get rid of the 5V regulator altogether?

[rfmerrill]

I'm not clear which is the MaxLinear part, is that on your BOM?

It was. I just removed it as it didn't work how I intended. The part was SPX3819M5-L-3-3/TR

The display is certainly part of the design that needs sorting out. Can you post a link to the OLED you used?

This is the product: https://www.buydisplay.com/i2c-16x2-oled-serial-character-display-module-screen-yellow-on-black

The choices in the dropdown I believe just change the stuffing of some setting resistors on the display, which you can fix yourself (and they may goof and ship you the wrong one). 6800 4-bit is the correct interface.
It's not as sexy as the Winstar module that Sergey used, but it's cheaper (at least if you're in the US--winstar module goes for like $80 on ebay and the other choices are shipping from poland).

Requires modified firmware that repurposes the LED pin to drive the RESET of the display, instructions are in my pull request here: https://github.com/SergeyMax/SolderingStation/pull/1

If you can get the winstar module (WEH001602AGPP5N00001) for a price that you're ok with, you could easily build the board to support both with just different firmware for each. Beware that the display on the one I linked above is upside down relative to the Winstar and Adafruit modules (in terms of where the io header is) and they're very mechanically different (overall size is the same but display cutout and hole positioning are different).

Can we get rid of the 5V regulator altogether?

The 5V regulator only directly drives the display if you are using a 5V display. Otherwise if you can derive 3.3V directly from the 10V rail, that is also an option. The 3.3V regulator sergey used does nominally support 10.5V input but it will dissipate 4x the power of course (and it might only barely be able to drive enough current to power both the display and the uC).

I chose to keep the 5V reg as it's a much bigger package and can easily tank the waste power compared to the 3.3 regulator. If you're going to spin a new board, though, you could substitute a bigger 3.3V regulator that can dissipate all the power you need.

[t0m]

Thanks. I'll try to spin the new board so that different configurations are possible, but I'm starting to think the Adafruit display is not a bad option.

[rfmerrill]

Thanks. I'll try to spin the new board so that different configurations are possible, but I'm starting to think the Adafruit display is not a bad option.

I suspect that the Adafruit 16x2 display actually contains the exact same LCD that Metcal puts on their current MX-500 line.



It's similar size and the colors and general look are very reminiscent.

Cursory testing indicated you might be able to get away with 3.3V IO but a level shifter isn't too expensive and is good insurance.

I like the OLED because it is much easier to read but ultimately it's not like you spend a lot of time looking at the display, and the OLED displays are pricey.

[t0m]

The display is an optional extra in my book.

I'm also thinking it might not be the worst idea to put the microcontroller on a separate PCB and make it actually, rather than theoretically, optional. If the signals to and from the microcontroller need to be shielded you could use copper tape, or wire braid, or even coax.

With stacked PCBs in mind, how high do the tallest components stick up off the board? I'm guessing the T130-6's need headroom of 4cm or so?

[rfmerrill]

The display is an optional extra in my book.

Fair enough. It does help to have some kind of indication of whether it thinks there's a tip present or not.

I'm also thinking it might not be the worst idea to put the microcontroller on a separate PCB and make it actually, rather than theoretically, optional. If the signals to and from the microcontroller need to be shielded you could use copper tape, or wire braid, or even coax.

Currently the microcontroller divides its clock in half to produce the 13.56MHz. You would need to provide an alternate generator of that and make sure that you respond to the tip disconnect event by immediately holding Q1 gate low (turned off) and waiting at least 100 ms before turning it back on. I do not know if instead disabling the DC-DC works just as well.

With stacked PCBs in mind, how high do the tallest components stick up off the board? I'm guessing the T130-6's need headroom of 4cm or so?

About that yeah. One other consideration with this board is the one inductor on the bottom that sticks out way further than any of the other bottom-layer components.

[t0m]

I figure that without the microcontroller you need a flip flop to divide the clock by 2, some logic for the tip detect, and something like a bicolor LED for signalling. There should be plenty of space on the bottom board for all the low voltage stuff, I can prototype it without changing the RF circuit, and people can remix it to put their favourite microcontroller on or whatever. Twisted pair should be OK for the connections between boards.

The bulky surface mount inductor will probably have to move to the top of the board. But if necessary it's easy enough to use a deeper box e.g. Red Dot IDH3-1 instead of IH3-1.

[rfmerrill]

I figure that without the microcontroller you need a flip flop to divide the clock by 2, some logic for the tip detect, and something like a bicolor LED for signalling.

You might as well just steal the 13.56M oscillator from the mamalala design. You'd need something to generate a good clock from the crystal. I think Sergey just went with the divide-by-2 because he couldn't source a 13.56MHz crystal in Russia.

You could take the SN74LVC1GX04 and the crystal and associated passives from that design, drive the "in" pins of the MAX1760x with its output, and then connect the tip detect logic to the "ENA" pins--if the enable pins are grounded, the output will be low regardless of whether you have the inverting or non-inverting version.

[t0m]

I have some suggestions for further changes to your BOM, based on what I can see from the pictures of your build

* You have a LM2596 module (as discussed) and NTC 8D-20 thermistor on the board that are not in the BOM.

* AFAICT C13,C27,C37,D11,D12,D14,D9, and L4 are omitted from your build.

* It looks as though you have omitted D7 and R18, so you probably intended to omit R10 as well.

* I'm not sure if you went through with your idea of a ferrite core for L8, but it's not in the BOM.

Some further comments:

* I think R17 should be omitted.

* I'm not entirely sure that you need C36 (but it probably can't hurt).

[rfmerrill]

I have some suggestions for further changes to your BOM, based on what I can see from the pictures of your build

* You have a LM2596 module (as discussed) and NTC 8D-20 thermistor on the board that are not in the BOM.

that's true but I restricted the BOM document to just things which are placed in actual footprints on the board. I'm not including the housing, switches, jack etc either.

* AFAICT C13,C27,C37,D11,D12,D14,D9, and L4 are omitted from your build.

* It looks as though you have omitted D7 and R18, so you probably intended to omit R10 as well.

I left components on the primary side partially filled in because I started buying them before I realized I would not be able to use the AC-DC circuit. I can probably just remove them or make a column for "not fitted". I figure if someone does want to build that part, having a suggested list might help, but maybe it would be good to separate it since I never actually built or tested that.

* I'm not sure if you went through with your idea of a ferrite core for L8, but it's not in the BOM.

nope I used an air-core inductor, it works fine once I got myself the proper coil form and learned how to measure it properly in order to tweak it.

* I think R17 should be omitted.

I kept it since it discharges C6 once power is removed.

* I'm not entirely sure that you need C36 (but it probably can't hurt).

I don't trust the output caps on the cheap DC-DC module, and it's a pretty cheap and small capacitor to add. Linear regulators are great but not perfect, and any noise on the input will get passed through to the output, even if significantly attenuated. LM7805 datasheet does recommend a cap on the input if it is located "far" from the power supply filter. It's not clear from the schematic but C37 is located much much further from the 7805, right on the rectifier by the transformer.

[t0m]

Thanks again for the detailed responses -- very helpful!

[t0m]

Here is the modified schematic I am working from.

[rfmerrill]

As an update, I have an album of pictures that show what the first unit I built looks like today.



Here's the power supply I'm using. Isolated 36V 4A LED supply--Amazon's choice! Spliced into the load end of the IEC cable to get an earth ground to pass along


Here are the connections on the back. I only had metal barrel jacks, so I used some plastic washers and a rubber grommet to isolate DC negative from ground. The screw holding Q1 to the housing doubles as a grounding lug (afaik the ground connection can be anywhere, but the area by the transistors is particularly noisy).


Inside you can see some minor changes I've made.


And here it is right at home

[rfmerrill]

For anyone who is revising this design or iterating on it: I might recommend also making the feedback path to the DC-DC more like the one in the mamalala design (using a buffer/amp and the potentiometer in series with the RF feedback instead of the regular feedback) as it's a bit more obvious how it's supposed to work. Mamalala's design in general uses a lot of op-amps to buffer signals which I think simplifies getting the design to work.

[quadtech]

Here is the modified schematic I am working from.

Nice...I did something similar, pruning Sergey's PCB to remove the Switch mode supply components,
and add a couple of connectors for the 32V and one for the 10V (using an external LM2596 or other DC-DC buck module).
The diptrace file is attached - rename it from .doc to .dip extension
The intent was to learn Diptrace, so did not really check if I ripped out anything which is really needed.

[quadtech]

Tom, if you want to use an external processor, only 4 pins of the STM are really used :
PA0 , PA1 : current detection (I think) (shown in blue, yellow in the attached image)
PA9 : Tip detect input to the cpu (shown in Red in image)
PA10: 13.56MHz freq output from the cpu (shown in Violet in the image)

Pins 7, 8 , 9, 10 of the LCD Header are unused - could be put to use for this.

[t0m]

Tom, if you want to use an external processor, only 4 pins of the STM are really used :
PA0 , PA1 : current detection (I think) (shown in blue, yellow in the attached image)
PA9 : Tip detect input to the cpu (shown in Red in image)
PA10: 13.56MHz freq output from the cpu (shown in Violet in the image)

Pins 7, 8 , 9, 10 of the LCD Header are unused - could be put to use for this.

(Attachment Link)

PA0 measures the current, PA1 measures the voltage, and together they are multiplied to calculate power

I've decided to use a cheapo 1604 LCD display, which is overkill for communication purposes, but fits nicely into a GCFI wallplate.

[t0m]

For anyone who is revising this design or iterating on it: I might recommend also making the feedback path to the DC-DC more like the one in the mamalala design (using a buffer/amp and the potentiometer in series with the RF feedback instead of the regular feedback) as it's a bit more obvious how it's supposed to work. Mamalala's design in general uses a lot of op-amps to buffer signals which I think simplifies getting the design to work.

Certainly the output impedance of the INA138 is way too high for the ADC and should be buffered, per the datasheet.

[rfmerrill]

Certainly the output impedance of the INA138 is way too high for the ADC and should be buffered, per the datasheet.

I don't know what exactly that output impedance number is supposed to mean. I assume it means when the output transistor of the INA138 is not turned on, otherwise the typical application wouldn't work...

Edit: Oh silly me: the INA138 has a high output impedance because it's a current output device.

The only requirement here is that the shunt resistor to ground be much lower impedance than the ADC input, which I think we're doing just fine. The effective ADC input impedance is a function of the sampling time, and according to the STM's datasheet, the 71.5 cycle sampling time that sergey's code is using results in an effective impedance that's basically really high.

But that's not what I was talking about: I meant the signal that comes off of C75/R40 and connects to the feedback node of the DCDC via a 2.2k resistor.

The feedback network of the buck converter is kind of headscratchy to figure out in Sergey's design, since VR1 has an effect that isn't quite obvious--you'd think it changes the nominal voltage of the DC-DC, but it also changes how much that voltage is influenced by the "outer" feedback loop.

I've attached the RF board schematic for mamalala's design to this post. You can see it achieves roughly the same thing through an arguably more complex but (imo) easier to understand circuit:

• The RF feedback signal comes in as RF_DET from the peak detector. The source of this signal is very different but you can think of it as equivalent to the node between R40 and C75 in Sergey's design in that when it increases in voltage, the buck converter decreases its output voltage in response
• RF_DET is then buffered and amplified by an op-amp which adds an offset, producing VFB
• VFB feeds into the DC-DC converter's feedback node via a variable resistance.

Now that I type this out and look at the circuit, I think it's not actually quite as simple either way... but the advantage of mamalala's design, I think is that it's much easier to know what voltage range the DC-DC will operate over: Since VFB is produced by an opamp running off of +5 and 0V, its voltage can never be outside the range of 0-5V.

I think in both cases you have some feedback signal Vf (RF_DET in the mamalala case, the voltage between R40-C75 in Sergey's case) and your control is something like:

Vsupply = Vs0 - k(Vf-Vf0)

Ideally if I were designing it I'd want both Vs0 and k to be intuitively adjustable.