Low-frequency PWM-controlled Mosfet heater circuit
The circuit above is used to control a 240W ceramic heater by means of a PWM signal at 30Hz, generated by a microcontroller, with 5V logic. The input is a standard 280W AC-DC external power adapter, the Meanwell GST280A24.The C1 is a large electrolytic capacitor, 2200 uF, with 35V maximum voltage, placed close to the load. This capacitor has a maximum ripple current of 2,06A @ 100 KHz. The voltage signal measured in the MEAS node is very clean, there are no spikes or other strange behaviour. On the other hand the 24V line is showing a 880 mV voltage drop everytime the mosfet is energized: there are no spikes but the 24V line just goes down, in a square wave manner, by 0,8V. The overall system works fine, but I have doubts in terms of reliability over time. First of all, turn on behaviour when SW1 is closed: the microcontroller is programmed for a soft start of the Mosfet M1 and does not immediately powers the mosfet for the first 2-3 seconds. I measured the inrush current to be sure of not trepassing the power adapter limits (120A). It looks that I have a voltage rise from 0 to 24V in circa 5ms. That, giving the 2200 uF of input capacitance, gives 18A of inrush current. The second question is the switching behaviour of the mosfet. I voluntary put a huge gate resistor to slow down the mosfet turn on. I want at all cost to avoid ringing and possible EMI issues (this circuit needs to meet the EN 55011). Do you see any problems in switching the mosfet so slowly? When I measure at MEAS, I get a nice square wave, a bit rounded in the top part. Do you see any flaws in this circuit at all?
Dear Olin, thank you, you are a reference for all of us.
The C1 was put there for to reasons. One is dampening the resonance of input capacitance with wire inductance - I got 1.5 meters wires from the power adapter to the DC connector - which could lead to potentially destructive outcome. The other is to help the hungry heater keeping is current requirements. I did try to eliminate C1 and substitute it with a ceramic capacitor of 20 uF with low ESR and I ended up blewing up the Tracopower, the micro and a bunch of other stuff. I simulated it and I got a 42V resonance peak, which almost completely disappears using a large C1 with high ESR instead. R5 is there to empty the big fat C1 cap when power is off. I verified and a small voltage of some volts keeps staying there for long time after power is removed.
Regarding Mosfet switching, here they come the scope printouts for the MEAS probe node. With an R4 of 10 kOhm:
With an R4 of 10 Ohm
There you see the mosfet ringing. With R4 at 10 kOhm the mosfet is running cool and nice. Switching takes approx. 100us, is this really a problem?
Regarding the 800 mV drop, it becomes 900 mV without C1. This was measured directly on the DC input. There are 10cm wires from the DC input connector to the control card.
I will try to add the decoupling caps as you suggested. What should I add in total?
- A bunch of small ceramic capacitors in parallel, between 24V and GND, close to the load: 10uF, 1uF and 100pF
- A small 100pF capacitor between the Mosfet drain and GND
Your main complaint seems to be that the 24 V power rail sags 800 mV when the heater is on. Three responses pop to mind:
- So what?
- It probably isn't anyway.
- What did you expect?
#1: So what? I don't see any harm in the 24 V supply actually being 23.2 V when the heater is on. Unless there is stuff you're not telling us, this supply is only used to power the heater, and to make the 5 V to run the microcontroller from. Both those purposes should be met fine with some ripple.
Since you are PWMing the heater, probably inside some feedback loop, the slightly lower input voltage will cause a slightly higher PWM duty cycle with no harm. The only real drawback is that the maximum heating power is reduced to (23.2 / 24.0)2 = 93% of what it would be if the 24 V supply stayed solidly at 24 V. Does that really matter? If it does, you're probably operating too close to the edge anyway.
#2: It probably isn't anyway. 800 mV does sound like a lot for a real commercial regulated power supply, although I didn't dig out the datasheet to check. However, with 10 A flowing around, there can be ground offsets in various places. Where exactly did you put the scope probe ground clip to measure this 800 mV dip? Try putting the scope probe (both the tip and the ground clip) immediately across C1. I suspect you will see a significantly smaller bounce. Also put the scope probe across the supply right at the power supply output. Again, I suspect significantly less than 800 mV change.
#3: What did you expect? You say the supply is rated for 24 V at 280 W, and you are applying a 2.4 Ω load to it. That will draw 10 A. It only takes (800 mV)/(10 A) = 80 mΩ somewhere to cause 800 mV of drop. Consider not only the resistance of the wire from the supply to the heater and the FET RDSON, but also the resistance of the ground current return path. The latter is too often overlooked. Remember that the whatever the FET source is soldered too has to be able to handle the full current too.
You say your main worry is radiation. That's a valid concern, but I don't like how you're addressing it. Trying to slow down the gate transitions is usually a bad idea. As you noticed, that doesn't seem to have slowed down the switching much anyway. If it had, though, you might have overheated and blown up the FET.
Reduce R4 significantly, or eliminate it altogether. The on-resistance of the output driver FETs in the microcontroller probably already are a few 10s of Ohms.
The snubber at right might be doing some good, but I'd look at the waveform on the resistor carefully to make sure.
C1 isn't doing anything useful, and I can't even guess what you think the purpose of R5 is. A big fat cap like C1 can be useful to hold up the supply after a sudden current demand, like when the heater is switched on. However, if the wires to the supply are thick enough, there shouldn't be much of a voltage drop to counter in the first place.
RF radiation is about high frequencies, which C1 isn't going to help much with. Put a few small ceramic caps right at the top side of the heater to ground. Maybe something like a 10 µF, 1 µF, and 100 nF in parallel to keep the impedance low over a range of frequencies.
However, the most noise will come from the FET drain connected to the bottom end of the heater. This is the node that will see 24 V transitions 60 times per second. Try a small cap at the FET drain to ground, like 100 pF to 1 nF. 1 nF has an impedance of 160 Ω at 1 MHz, and it goes down from there proportionally at higher frequencies. Keep the cap small to not cause excessive surge current thru the FET whenever it turns on.
Keep the node from FET drain to heater physically as small as possible, and maybe consider shielding it.
0 comment threads
This answer is regarding the resonance issue that C2 is fixing, and how it could be done with a smaller cap.
There are two aspects of it. To start, consider the 5.6uH vs the 10uF capacitors. It can be stopped from resonating with the addition of a series resistor. With the modest DC current (powering only the micro) and plenty of voltage to spare, it shouldn't cause trouble.
Once the resonance from the 5.6uH is out of the picture, it becomes easier to examine the 24V node. This node vs the inductance of the incoming +24V supply (and return) wires will also be very underdamped. Even if the resonance is not excited by the circuit's operation, it will still produce an overshoot upon turn-on. It can be damped with a parallel (R+C) shunting the low-ESR capacitance on the +24V. This is what the 2200uF cap with its larger ESR is doing, but we can get the same damping with less capacitance. It is the R in the parallel damping network that really matters, the C in the parallel damping network only has to be big enough so that the parallel (R+C) has significantly lower impedance (at the resonant frequency) compared to the low-ESR C we are "fixing". I found that 5x the low-ESR C seems to work well.
Here is a crude (qualitative purposes) simulation without the damping. V1 is the 24V node. I assumed the minimal 10uF capacitance on the 24V node. 1uH "wire inductance" arbitrarily chosen to illustrate the concept as I think is relevant here, not really a properly modeled wire.
(graphs: 10V/div and 10us/div, click for link)
And next, here is with some damping (again 10V/div, 10us/div, clickable). Note the series damping added to the 5.6uH and parallel damping (R+C) added to the 10uF on the "V1" (+24V node).
Since we shouldn't rely on a particular +24V/return wire being connected, it's not possible to pick the ideal parallel damping R. So I think a suppression diode (not shown) is really a good idea, to limit any overvoltages that remain. An avalanche diode with Vrwm=28V (Vbr=32V typ) should be about right. Another note is that the caps may be derated significantly at the +24V operating voltage, depending on the exact type.
Finally, to be clear, all the values (apart from suppression diode) above are illustrative, and don't take into account the needs of the "business end" of the circuit, although the resistive heater should be relatively benign.
1 comment thread