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

[t0m]

I found a useful post about load voltage and current in inductive heating circuits --> http://www.richieburnett.co.uk/indheat.html

In the plots below, the X axis shows the frequency of the tank circuit. The white vertical line represents the resonant frequency of the heating coil. The upper plot shows the amplitudes of the voltage across the load (red lines) and the current through the load (green lines) for a range of values of the Q factor. The sharply peaked lines represent high Q, the smoother lines low Q. The lower plot shows the phase relationships relative to the inverter voltage for the same values of Q.

Quoting from the site: "With no workpiece installed, losses are low and Q factor is high. This gives rise to the sharply peaking currents and voltages in the top graph, and the abruptly changing phase shifts in the bottom graph. As a lossy workpiece is introduced the overall Q factor of the LCLR network falls. This causes less resonant rise in the inverter load current and the voltage across the work coil. The resonant peaks become less tall, and broader as the Q factor falls. Likewise the phase of the inverter current waveform and the work coil voltage slew less rapidly for lower Q factors."

If I understand correctly, supplying a frequency slightly above the resonant frequency of the heating coil (shaded blue) ensures that it presents as an inductive and not a capacitative load.  Under these conditions the load current will lag the inverter voltage by up to 90 degrees, and the load voltage will lag the inverter voltage by between 90 and 180 degrees. Keeping the phase relationships in this stable region avoids stress on the semiconductors in the inverter. Tuning the circuit for maximum power can be achieved by gradually reducing the frequency of the tank circuit towards that of the heating coil.

When the iron is heating up, the heating losses to present as a heavy load that damps the resonance in the network. When it hits its Curie temperature, the load will fall and the network Q will rise. The amplitude of the voltage across the load and the current through it will both increase. Also, the phase of the voltage across the load will lag further and approach antiphase with the inverter voltage as resonance increases. The phase of the load current may change in either direction, and slews in a smaller range.

Now, if you rectify the voltage across the load and put it through a low pass filter, you get a measurement of amplitude that can be used as negative feedback to regulate the power to the heating coil. As I understand it, this is the basis of the Metcal patent and what is implemented in both the MX500 and mamalala's circuit (in a fancier way). The SP200 uses a measure of load current, suitably buffered, in a similar manner.

According to your simulation, the SergeyMax sensor is a negative function of power (I*V) which should provide a more sharply peaked (and presumably more sensitive) feedback function than I or V alone.

[rfmerrill]

*snip*

According to your simulation, the SergeyMax sensor is a negative function of power (I*V) which should provide a more sharply peaked (and presumably more sensitive) feedback function than I or V alone.

Apologies I could only skim this right now, at work currently but can read it more in depth at a later time.

One thing that's kind of buried here is that the Metcal handpiece includes a series L and parallel C hidden near the plug. I was really confused at what I was measuring before I learned that.

richard.cs 's page contains a table of impedance measurements of the iron tip + handpiece that gibes with what I've seen:

	R	X		Equiv X	SWR	S11
Cold:	42.3	+13j		153 nH	1.4	-16dB
Warm (but below Curie temp):	55	-16j		730pF	1.1	-23 dB
Hot (above Curie temp):	12	+24j		280 nH	5.1	-3.4 dB

So when I read this, it seems like if you are using constant current control, then if you are putting 40W into the handpiece in row 2, then you are putting (12/55)*40W = 8.72W into the handpiece when it is above curie temp. This kind of illustrates why constant current is not feasible for more power output, because if you doubled the 40W, you'd also be doubling the 8.72W. My first sergeymax unit drops down to 9W on the display (which is probably more like 5-7W real output) if I hold the tip still and cup my hand around it.

The feedback circuit in the SergeyMax design very roughly outputs a voltage proportional to the phase angle between voltage and current: approximately 0 if they're in-phase, negative if current is leading the voltage, positive if current is lagging the voltage. Messing with the LTSpice model, that voltage pegs at its maximum of 1.2V when the load is 12 ohm + 280nH (the "hot" value), and is slightly below 0 at the "warm".

A nice advantage of this feedback mechanism is that with the iron in the workstand (which contains a magnet that saturates the heating element) the power drops immediately to sub-5W, which I think is a lot better than what it drops to with the older design (and probably why the older design needs that timeout circuit).

[t0m]

It would be interesting to have equivalent measurements for the SergeyMax circuit.

The graphs I posted assume an LCLR system, with the heater having a capacitor in parallel with the induction coil. AFAIK that's consistent with what we know about the handpiece.

I wonder if Richard's circuit was optimally tuned, because if I understand the blog post, it is saying that the capacitative reactance observed in row 2 would have been putting thermal stress on the power transistor.

In the inductive region that Richie recommends, current always lags voltage and the lag tends higher with Q. So if the feedback voltage is proportional to phase it should act to cut power when the iron hits its Curie temperature and the Q factor jumps higher. That makes sense.

[rfmerrill]

It would be interesting to have equivalent measurements for the SergeyMax circuit.

These are measurements of the handpiece + tip cartridge, I assume he heated it through external means and measured it with a VNA or similar, but maybe not? In any case if they were correctly taken it is independent of the power supply circuit.

Also I built my test load circuit with a capacitance of ~730p in series with a 50 ohm dummy load and it does seem to mimic the iron tip okay.

[t0m]

 Thanks for setting me straight there

Ocela at VRTP.ru recommends putting a noise suppression cap across the DC inputs: https://vrtp.ru/index.php?showtopic=30618&view=findpost&p=837982

This one looks like 220 nF / 320V.

[rfmerrill]

Thanks for setting me straight there

Ocela at VRTP.ru recommends putting a noise suppression cap across the DC inputs: https://vrtp.ru/index.php?showtopic=30618&view=findpost&p=837982

This one looks like 220 nF / 320V.

Haha wow either someone followed my idea or just had the same idea as me XD same converter module.

It looks like that person has tied their dc negative to the housing at the input, which I did not do. I have it separated from the housing via a rubber grommet and plastic washers, and a separate mains earth wire connected to one of the transistor screws. DC negative is tied to he housing closer to the RF side -- via the RF output jack and also via the screw on Q2 for shunting the coupled noise from the tab

[rfmerrill]

	R	X		Equiv X	SWR	S11
Cold:	42.3	+13j		153 nH	1.4	-16dB
Warm (but below Curie temp):	55	-16j		730pF	1.1	-23 dB
Hot (above Curie temp):	12	+24j		280 nH	5.1	-3.4 dB

I was trying to work out how to simulate the "hot" impedance since the obvious series solution would require a non-inductive 12 ohm power resistor: I did some calculations and you can get close by putting 330 nanohenry in parallel with 50 ohm. The impedance at 13.56MHz ends up being (12 + 21.36j) ohm

Edit: adding another 33 nH in series with that brings it even closer, 12.01 + 24.17j

[t0m]

It looks like that person has tied their dc negative to the housing at the input, which I did not do. I have it separated from the housing via a rubber grommet and plastic washers, and a separate mains earth wire connected to one of the transistor screws. DC negative is tied to he housing closer to the RF side -- via the RF output jack and also via the screw on Q2 for shunting the coupled noise from the tab

He agrees with you about that... according to google translate:

One more note. Don't forget to connect the station body to the ground! (Since with such a power supply, the ground is not connected to the case (it does not go to minus) and there is a pickup of about 80V on the case)

[t0m]

Here are the Diptrace files I am using at the moment. I haven't done much except change the format to 100x62 mm and add 100 or so vias. A terminal block on the left hand side accepts (up to) +40V/0V/+10V. The schematic has less than 300 pins so it can be edited with the freeware version. I haven't had much joy exporting Eagle files from Diptrace but it is possible in theory.

[quadtech]

I have the L2 routed a bit differently in mine.

I was also thinking to move the crystal and the 22pf caps to the top of
the STM (parallel to P1), to make the routing
similar to the workaround Sergey had in the blog, but didn't get round to it.

[rfmerrill]

I've come up on a bit of an annoyance: My load simulation circuits needs to be either directly connected or calibrated for even a short coax lead, because even the 6-inch (15 cm) coax I was using causes significantly different readings on my nanoVNA :/

I highly suspect that the whole system is intended for 50 ohm transmission lines despite using what is traditionally a 75 ohm connector. I can't say that for sure though. Using a 50 ohm coax gives me different readings too however.

[rfmerrill]

In my second unit rather than populating R42 I put a 2512 SMD resistor on the R45 footprint. I am not sure why sergey went with populating the further-away resistor instead of the closer one. I think putting the resistor on the top layer with shorter routing to gate and source might help something, but I don't know.

[t0m]

Some comments on my layout: SergeyMax designed his board to only have SMD parts on one side. I loosened up that restriction for a couple of parts like C4 (to simplify the BOM) and L2 (which is very tall for a surface mount part). I'm surprised he didn't use any vias at all, but I went overboard on them and I'm sure I could have used way fewer.

[t0m]

I'm thinking of getting rid of the LED and using the spare pin on the microcontroller to measure the feedback voltage. The idea would be: U2 pin 5 -> inverting op amp as voltage attenuator -> ADC. While I'm at it I could put op amp buffers on the inputs to pins PA0 and PA1. Mamalala used a TLV2374 and AFAICT that should be a reasonable choice here too.

[rfmerrill]

Ugh, I really wish I could find a way to limit inrush current that doesn't cause issues. I've tried three different spec NTCs and they all cause the display to glitch out under load.

The power supply I'm using doesn't care, I just hate the rather prominent arc when making the power connection on the bench.

[t0m]

Could you simply put a fixed resistance in series with C6? Or try a smaller cap?

[rfmerrill]

Could you simply put a fixed resistance in series with C6? Or try a smaller cap?

I think a fixed resistance would have the same problem as the NTC if not worse. I think the issue is that the NTC isn't staying hot enough to keep its low resistance. Maybe if I moved R17 and thermally bonded it to the NTC, but that seems like an ugly hack.

I think a better solution would be an instantaneous current limiter, or a resistance that is switched out once the cap is partially charged. If I could perfectly limit the current to 4A the cap would charge in a few milliseconds.


Coming from the complex digital world I always find myself wanting for robust power management where I could do things like switch out the inrush limiting at a certain point in the sequence.

[t0m]

Interesting. I would have thought that a bigger problem was that the NTC stayed hot during momentary power glitches and failed to limit the subsequent rush of current as the cap recharged.

8 ohms sounds about right to limit the current to 4A on start up. Perhaps put the resistance in parallel with a relay that opens after a delay of a couple of milliseconds? Or switch the shunt path with a low RDS MOSFET?

[rfmerrill]

Interesting. I would have thought that a bigger problem was that the NTC stayed hot during momentary power glitches and failed to limit the subsequent rush of current as the cap recharged.

That would mean I'd see worse behavior with the NTC bypassed. The problem is that I picked an NTC rated for the maximum draw of the soldering station (~2-3A) but the draw with the iron in the stand is significantly less, so the NTC will cool down and increase in resistance. Then when the current increases sharply when suddenly applying a heavy load, the NTC does not heat up quickly enough for its resistance to drop, and the capacitor is not big enough to maintain the minimum input voltage which could drop down into the 20s.

[t0m]

I  don't understand. What you're describing sounds to me as if the NTC is limiting the current draw. Isn't that what it is supposed to do? 

If the problem is a glitchy display, could it be that the metal bezel of the LCD is grounded to the case?

[Dave8266]

Does anyone know why OLEDs in that format (16x2 character) seem to have mostly vanished? Sparkfun and Adafruit both used to sell them but now they only sell LCDs.

Part of it may be because graphic OLED displays (and the library software to drive them) are so widely available now. 

Even if you never draw shapes or graphics, the graphic display is still a superior text output device, as it gives you so much flexibility for font styles and sizes, single-pixel positioning, variable character widths (non-monospace fonts), etc.  And they are basically at parity for cost at consumer retailers.

[quadtech]

@rfmerril - I am thinking if it helps to have smoothen the output voltage change curve of the
TPS54560 - maybe increase C24?

@t0m - in your PCB, you moved the MOSFET driver to the middle of the board - isn't it preferable to
have it closer to the mosfet as in Sergey's design?

[t0m]

@t0m - in your PCB, you moved the MOSFET driver to the middle of the board - isn't it preferable to
have it closer to the mosfet as in Sergey's design?

Yes -- good catch!

[rfmerrill]

I  don't understand. What you're describing sounds to me as if the NTC is limiting the current draw. Isn't that what it is supposed to do? 

Yes, the NTC is behaving as designed. I am saying it doesn't achieve what I want.


With the NTC I most recently tried, I measure a resistance of ~6-6.5 ohm when it's cold.

The station draws very roughly 200mA at idle. If I connect the NTC across the output of my bench power supply and set it to 200 mA, after a few moments the voltage drop is 0.74V, so it has dropped to 3.7 ohms. It took about 30 seconds to reach that resistance.

If I then suddenly increase the supply current to 2.2A, the voltage drop across the NTC spikes to >3V, drops to 1V within 1 second, and after a few seconds it reaches 0.74V and seems mostly settled there, about 330 mohm.

What I'm demonstrating here is that the NTC is not an instantaneous current limiter--its resistance reacts very slowly to the current averaged over a fairly long period of time (seconds). When you pick the iron up out of the stand and the system's draw jumps from 200mA to 2.0A, the input voltage presented to all of the regulators on the board will drop suddenly by ~6-7V. The regulators then have to draw *even more* current to drive the same output power, and there will be some momentary instability over the fairly long timescale while the NTC warms up and its resistance decreases.

I think NTCs are just ill-suited for what I'm trying.

If the problem is a glitchy display, could it be that the metal bezel of the LCD is grounded to the case?

I was seeing display freezes and blanks very frequently until I bypassed the NTC. Now I am only seeing uC resets much less frequently, so I think I removed at least one component of the problem.

Another possible issue is that the cheap DC-DC converter module might just not have enough input capacitance. I wonder if placing the NTC only in series with the RF supply and connecting the 10V converter's main input directly to power might help? Again though, not a big deal as the inrush limiting seems largely unnecessary.

Interesting. I would have thought that a bigger problem was that the NTC stayed hot during momentary power glitches and failed to limit the subsequent rush of current as the cap recharged.

This isn't actually the goal of the inrush limiting. I only want to limit the initial large current spike when the cap charges up from zero. The current waveform after that should be much smoother due to the capacitor. If the supply voltage suddenly jumps that's a problem with the power supply and I don't intend to protect it from itself.

[t0m]

This isn't actually the goal of the inrush limiting. I only want to limit the initial large current spike when the cap charges up from zero. The current waveform after that should be much smoother due to the capacitor. If the supply voltage suddenly jumps that's a problem with the power supply and I don't intend to protect it from itself.

That makes sense, and is why I initially suggested a resistor in series with the cap.

I found your experiments with the NTC quite informative. It seems their use case is really fairly specific.