CFW for KSGER/Quicko STM32 Soldering Stations

[DavidAlfa]

But there's no

Edit: ?? I don't  what happenned to the message... neither where did that come from.
I think I left the phone unlocked lol.. luckly didn't  type my credit number 

[DavidAlfa]

In my schematic it's a simple 1n4148 diode.
The schematic is wrong, it's reversed.
So it limits the amp input to ~0.7V. All ok.
The ksger uses a LED to do the same thing.

[Bozog]

In my schematic it's a simple 1n4148 diode.
The schematic is wrong, it's reversed.
So it limits the amp input to ~0.7V. All ok.
The ksger uses a LED to do the same thing.

Ok, so in this case there is nothing to protect the opamp input from the negative voltage spike, I'll have a look at that later.

[DavidAlfa]

Edit:  Let me check the schematics...
Edit2:  The're input clamp diodes. There's a 4K7 limiting resistor + filtering cap. Might be ok.

[DavidAlfa]

6. hakko tip @ opamp o/p, while switching from 0% pwm to pwm on, you can see there is a significant drop in temperature, which of course over ramps the pwm and causes overshoot and hence vicious cycle.

Does that statement corrrespond to this pic? I see it needs almost 4mS to be accurate! Now please repeat in lower voltage scale to see the variation in uV


If that is produced by the thermal impedance, we have a problem. We can't measure the temperature after each cycle on these, the wait is too long... 5mS is 25% of the pwm cycle!
I'll have to change back to tasked measurement, every x mS stop PWM, wait, measure, and resume PWM. I don't like that too much but if it works...
Or! We can apply a factor without waiting that much.  Measure and removing  a %. Although that may worsen the problem. I don't know.



Edit: I attach 0.1%, 0.5% , 1% and 2% min PWM bins. Please reset the settings holding the button before powering on (For now that settings are hardcoded and apply only on resetting). Please report results
And have a look at the temp! Anyway it's unlikely to burn the tip at 2%, I remember about 10% to hold the tip at 450ºC

[mikes]

Just for reference:

I just received a KSGER "2.1s" unit. It's what's called around here a "ve2.1s 'r2'". Except, it has a STM32F103RBT6 (128K flash). I'm guessing they buy the cheapest which are available when they do a run. #dreamcat4, you may want to note that on your page.

Pulled #DavidAlfa code from here. I think I've got it moved over to the B version (in addition to less flash, B doesn't have timers 6 & 8 ). Why does ST make this so hard??? Have an MCU on order to try (so I can unsolder the original as a backup), but it's not a B, I'll see how it works.

In any case, let me suggest that the STM32F103RBTx be used as the future target. AFAIK, that should be forward compatible with the C and higher versions. As David said:

The actual purpose of this project is to make a compatible firmware -with- existing soldering stations, just as they come from aliexpress, ande require the minimum modding from the user.

Looking through github, I didn't see any obvious versions which support I2C displays, like the #DavidAlfa code mentioned above, including his git. Is the intent to eventually allow that to be #define'd into the mainline code?

[Bozog]

6. hakko tip @ opamp o/p, while switching from 0% pwm to pwm on, you can see there is a significant drop in temperature, which of course over ramps the pwm and causes overshoot and hence vicious cycle.

Does that statement corrrespond to this pic? I see it needs almost 4mS to be accurate! Now please repeat in lower voltage scale to see the variation in uV


If that is produced by the thermal impedance, we have a problem. We can't measure the temperature after each cycle on these, the wait is too long... 5mS is 25% of the pwm cycle!
I'll have to change back to tasked measurement, every x mS stop PWM, wait, measure, and resume PWM. I don't like that too much but if it works...
Or! We can apply a factor without waiting that much.  Measure and removing  a %. Although that may worsen the problem. I don't know.



Edit: I attach 0.1%, 0.5% , 1% and 2% min PWM bins. Please reset the settings holding the button before powering on (For now that settings are hardcoded and apply only on resetting). Please report results
And have a look at the temp! Anyway it's unlikely to burn the tip at 2%, I remember about 10% to hold the tip at 450ºC

This one is a chinese tip waking from sleep 0%, and it works perfectly.

I think you are right the inbuilt opamp diodes are sufficient.

Thanks for the bins I tried the 0.1% one but unfortunately it doesn't help, it seems the temperature reading error is proportionate to the duty cycle, i.e. 1% pwm, error is negligible but 50% pwm introduces a big error, temperature reads lower than actual. This means when the pwm is ramping down as we approach set point we have already overshot and nobody knows lol!

I don't like to admit defeat but I'm running out of ideas

[DavidAlfa]

This one is a chinese tip waking from sleep 0%, and it works perfectly.

I think you are right the inbuilt opamp diodes are sufficient.

Thanks for the bins I tried the 0.1% one but unfortunately it doesn't help, it seems the temperature reading error is proportionate to the duty cycle, i.e. 1% pwm, error is negligible but 50% pwm introduces a big error, temperature reads lower than actual. This means when the pwm is ramping down as we approach set point we have already overshot and nobody knows lol!

I don't like to admit defeat but I'm running out of ideas

Just for testing... This fw allows much bigger delays. Max 5mS, and 5mS by default(Again, reset settings).
I didn't test it... be careful about the tip.
If that doesn't work... Man I'm out of ideas.
Given the low impedance of the TC + heater, how can it hold energy for so long?
Can you post the exact picture where  the TC goes lower with the PWM? I can't see a big  difference in the TC output in your pics.
Don't worry, it's my IQ not shining again  .
The recovery time, sure, it was very noticeable

[DavidAlfa]

Just for reference:

I just received a KSGER "2.1s" unit. It's what's called around here a "ve2.1s 'r2'". Except, it has a STM32F103RBT6 (128K flash). I'm guessing they buy the cheapest which are available when they do a run. #dreamcat4, you may want to note that on your page.

Pulled #DavidAlfa code from here. I think I've got it moved over to the B version (in addition to less flash, B doesn't have timers 6 & 8 ). Why does ST make this so hard??? Have an MCU on order to try (so I can unsolder the original as a backup), but it's not a B, I'll see how it works.

In any case, let me suggest that the STM32F103RBTx be used as the future target. AFAIK, that should be forward compatible with the C and higher versions. As David said:

The actual purpose of this project is to make a compatible firmware -with- existing soldering stations, just as they come from aliexpress, ande require the minimum modding from the user.

Looking through github, I didn't see any obvious versions which support I2C displays, like the #DavidAlfa code mentioned above, including his git. Is the intent to eventually allow that to be #define'd into the mainline code?

Oh, don't be shy. Select the B part even if you have a "8". 99.99% It will work.
The fw is still made for the 8 in mind.
The project properties will say it's a B, I noticed that when creating a new project, It always select the B part on the preprocessor define, no matter if you choose a 8 or a B.
It's the linker script what limits the flash.
Anyway, the code will work the same between R, C versions... the pin coun't doesn't matter, the core is the same.
Only few adjustments are required. Yes, we were figuring out the correct way to unify everything.
But Stm32 Cube Ide gave some trouble, so in the end  I'll probably have a folder for each hw with the base project and few definitions, requiring the user to copy it to the project root. Not a big mess.
The i2c is barely tested yet, there's a secondary branch for it, pretty old, I just made the i2c work in DMA mode and left it there.
I will soon add the i2c support to the main branch.
But it's hardware i2c, these ksgers don't use it. A simple 2-wire mod will fix this, check this:
https://www.eevblog.com/forum/reviews/stm32-oled-digital-soldering-station-for-t12-handle/msg3319738/#msg3319738

As if i2c wasn't already slow enough, so to make matters worse they used software IO! Those chinese must be high... or their water poisoned .
Sorry for the multiple posts!

[mikes]

Oh, don't be shy. Select the B part even if you have a "8". 99.99% It will work.

No, I have a "B", the archive is for a "C". "8" is nowhere to be found.  But, I'll be replacing the B (which will be the backup) with an E, because that's what I can get soon. I don't doubt it will be fine. The point is that future builds should be for the "least common denominator", to support less sophisticated users in the future.

it's hardware i2c, these ksgers don't use it. A simple 2-wire mod will fix this, check this:

I saw that, and that's the plan.

Edit: are all 103s the same die, and they just bin them when screening? There are so many, that doesn't provide much faith in their production processes.

[Bozog]

Can you post the exact picture where  the TC goes lower with the PWM? I can't see a big  difference in the TC output in your pics.

Yeah I'll have a go at that tomorrow. It does seem strange how it performs so badly compared to the cheap chinese one, like I speculated before it could always be a faulty tip, but I don't fancy buying another  and either way it seems predominantly a hardware issue.

[DavidAlfa]

No, I have a "B", the archive is for a "C". "8" is nowhere to be found.  But, I'll be replacing the B (which will be the backup) with an E, because that's what I can get soon. I don't doubt it will be fine. The point is that future builds should be for the "least common denominator", to support less sophisticated users in the future.

Edit: are all 103s the same die, and they just bin them when screening? There are so many, that doesn't provide much faith in their production processes.

It's already done for the c8!
It uses the linker script "STM32F072C8TX_FLASH.ld"
And yes they seem to be all the same die. If your C8 gets programmed  as a CB without errors, your extra flash is ok!
Not sure if it's fault free in the complete voltage and temperature range, but at least for me it worked well both for a 072C8 and a 103C8.

[Bozog]

Some screenshots of the error caused by the pwm, these are all waking from sleep at various temperatures, you can see the error ranges from 0 to -80mv from 0 to ~90% pwm.

The first 4 screenshots are the hakko which produces a negative error, the last one is the chinese tip which produces a similar error but it is positive.

If you think about it the negative error increases the overshoot tendency, whereas the positive error decreases it, and this is the problem.

I have a new theory, the difference between these tips is in one tip the heater is in the positive leg of the thermocouple and in the other tip it is in the negative leg of the thermocouple.

[DavidAlfa]

If that was true, the TC would push negative voltage, the amp would always see ~0°C.

[Bozog]

If that was true, the TC would push negative voltage, the amp would always see ~0°C.

No it wouldn't, it would look like this.

[DavidAlfa]

Oh, you mean that! In that case there's no difference. They're in series in any case...
That negative spike is huge! Needs a diode or RC snubber ASAP.
I think the problem is the spike. The chinese tip stores less energy and doesn't discharge the amp cap completely.
But the hakko spike is so huge that it needs a lot of recovery time?
A snubber would be an option here.
Ex. 10nF in series with a 10ohm resistor. Totally empiric, bigger cap, lower the spike, higher the losses.
This in parallel with the tip. Will absorb the spikes but don't use power once the RC constant is over.
Very used in flyback circuits.

[Bozog]

Yeah after drawing it I couldn't see why it would make any difference either.

That is not the negative spike you can see in the last screenshots (0v is at the bottom), this is the amp output at 0v for around 100us.

I previously measured the negative spike (with 1n4007 flywheel diode across heater), the chinese tip was -1.6v and the hakko -3.2v, removing the diode made the hakko go upto around -4v, funny thing is these spikes only last less than 1us.

I think slowing the mosfet turn-off may help somewhat, I see the turn-off resistor R6 is 10k, perhaps I will try a 20k in its place

Screenshots taken from across heater, 0v at centre:

[shh]

Hello all,

I'm sorry to interrupt just in your eager conversation.
I received my Quicko station, new hw-design (sais 108W, AC and has also a DC24V4A), HW:V3.1 SW:V7.1M, with a 6-pole connector for the soldering tip - grounding of the chassis is okay out of the stock
However, I'm wondering if and how I coud try out your new firmware for this device (because I cannot set different settings for different tips)
The controling device has got a μC I cannot find any infos for:
STM32
F072TBT6 <- for me this really is a B not an 8
GQ28Z 129U
CHNGQ 024

Do you know what STM32 piece this is - maybe it's a fake?
And, if this chip is possibly able to load your firmwares.
Thanks in advance!

Cheers
shh

[DavidAlfa]

And for that the snubbers exist!
If you have the components, try 10ohm/470pF. And increase the capacitor value while taking screenshots.
Also, 1N4007 is not a fast diode.
Put a fast schottky rectifier and you'll see!

[DavidAlfa]

CB is the same as the C8 with a litle more flash.
It should work perfectly with the C8 fw.
However you'd better replace the mcu instead, to preserve the original fw.

[Bozog]

My chip is a CBT6, I'm assuming TBT6 is a typo?

I'll have a look for a faster diode and try that, I'm not convinced the snubber is going to help much but I'll give it a go.

[Bozog]

Well you were right, I found a sb560 schottky diode and that clamps the negative spike much better than the 1n4007, brings it down from -4v to around -1v

Unfortunately it doesn't help with the pwm temperature error, I think ideally this needs compensating for in the software on a per tip basis, although I quite like the idea of a jbc t245 handle

[floobydust]

When the heater is on, the op-amp input is saturated and needs time to recover.
There are at least five different thermocouple op-amps out in the wild, and some are low power, slow parts. Their overload characteristics vary - some cheap op-amps phase-invert when overloaded, especially with any -ve input spikes.

Here are scope traces from my KSGER T12 H/W V2.1S with OPA336N op-amp "A36" and their D12 tip, idling at setpoint 300°C.
It shows the heater voltage along with TC op-amp output to give some clarity.
You can see the mosfet takes about 100 20usec to turn off.
The op-amp takes 250usec to desaturate.
The thermocouple-to-heater thermal time constant is about 1.7msec, so 5 tau to drop to 99.3% of settled value needs 8.5msec.

That gives us a blanking period of around 10msec required.

The PWM frequency is 5Hz (200msec). I think 50/60Hz worked for old triac controllers but we cannot use it due to the long (thermal) time constant of the TC. I don't see an advantage to usec timing, when the accuracy reading the TC is lost due to the high speed.

[Bozog]

When the heater is on, the op-amp input is saturated and needs time to recover.
There are at least five different thermocouple op-amps out in the wild, and some are low power, slow parts. Their overload characteristics vary - some cheap op-amps phase-invert when overloaded, especially with any -ve input spikes.

Here are scope traces from my KSGER T12 H/W V2.1S with OPA336N op-amp "A36" and their D12 tip, idling at setpoint 300°C.
It shows the heater voltage along with TC op-amp output to give some clarity.
You can see the mosfet takes about 100usec to turn off.
The op-amp takes 250usec to desaturate.
The thermocouple-to-heater thermal time constant is about 1.7msec, so 5 tau to drop to 99.3% of settled value needs 8.5msec.

That gives us a blanking period of around 10msec required.

The PWM frequency is 5Hz (200msec). I think 50/60Hz worked for old triac controllers but we cannot use it due to the long (thermal) time constant of the TC. I don't see an advantage to usec timing, when the accuracy reading the TC is lost due to the high speed.

Interesting, so I assume this is custom firmware and you are measuring the TC 10ms after the heater switches off?

[DavidAlfa]

The only possible approach possible I see is to only take 10 measures/second, and wait 10mS for each.
We lose 100mS time, 10% power. We can barely filter with so few values, the filter must have a very low coefficient to avoid delay.

We can't take a lot more of measurements, we'll lose a lot of power.
We can't take less measurements, as the high power heat the tips very fast and risk overshooting.