Adjusting an LC filter while circuit is running: bad idea ?

[ratatax]

Just wanted to share with you my magic smoke experience.

I have a dc/dc converter i designed around a switching IC, it has input capacitors but despite that it creates a lot of noise on the input side, going to other circuits.

So why not add an inductor in series before the capacitors, to make an LC filter and hope reducing the noise.

I cut the VCC trace, put some solder pads on it, powered the circuit and tried different inductors while the circuit was running.

Sometimes when you directly touch component's legs to pcb pads, connection is intermittent. I think that's what happened here, I heared a quick "buzz" sound then the switching IC literally exploded. I think it was a 33uH inductor and 47uF capacitor.


I guess what happened here is that an inductor, when quickly switched on/off (by my unreliable contact) can create peak voltages much higher than the nominal current, and that's what killed the IC. Am i right ?

Now I'm bothered by what would happen if the user plugs the power to my device, and put in only half-way creating a bad contact. Will the same thing happen ? I'm not sure, since it can only switch on/off before the inductor, not after, but i'd like your analysis on that

[T3sl4co1l]

Probably inrush surge.  What capacitors are around said inductor (input and output sides)?

Tim

[ratatax]

Sorry for the poor-looking schematic.

Input caps are 4x22uF MLCC in parallel (two directly connected, two with long zigzag PCB traces in series to increase ESR (about +0.5ohm) and damp the input spike created when plugging the power supply)

Output cap is just a 1uF MLCC

[T3sl4co1l]

Yeh what's before the inductor?

What regulator BTW?

Tim

[ratatax]

Before the inductor there is a DC wall adapter providing 12V

The regulator is an AN_SY8205 : https://datasheet.lcsc.com/szlcsc/Silergy-Corp-SY8205FCC_C111875.pdf
It can tolerate about +30V at its input and has configurable output voltage (R30 / R59 divider on the schematic).

Could that means the inductor created a spike higher than 30V, or negative voltage perhaps ?

[T3sl4co1l]

Ah, adapter probably has a few hundred uF capacitance in it, making a pretty low impedance.

During inrush, your inductor will saturate pretty quickly, bringing the total inductance down to a few uH (the cable will be a significant part of the total here, in fact).  We can write the resonant impedance Zo = sqrt(L/C), taking C as the series total, which, eh, might as well assume just the ceramic caps for simplicity, but the total will be slightly less due to the adapter not having infinite capacitance.

That gives Zo somewhat less than an ohm.  But the capacitors have very low ESR (maybe 10mohm), so the Q can be large, and the peak resonant voltage will nearly equal the supply, i.e., 12V on top of 12V, for a voltage peak of 24V.

Which should still be fine compared to the regulator's 30V rating, but there's one more catch.

The ceramic caps are probably dropping in value under bias.  Check their datasheet, if they have a C(V) curve.  As C drops, the voltage rises that much faster, and the peak can be several times the supply voltage.

Simplest cure: probably just put a SMAJ15A (or any from 12 to 20V should do) from GND to regulator VIN.  Or P6KE or any other TVS of comparable rating.  This will clamp the overshoot.

No need to use so many input caps, and don't do goofy stuff like using PCB traces for resistance -- copper has much more inductance than resistance (in most geometries), and the resistance has a terrible tempco besides.  Use a real resistor, or a capacitor with internal resistance (tantalum or electrolytic; preferably NOT tantalum as they are wont to ignite in this position ).

Or, simply using a bulkier, lossier electrolytic in parallel with the ceramic.  When ESR ~= sqrt(L/C), the inrush is dampened, reducing overshoot.  You can still use the TVS for added protection, but you can likely get away with just one or the other.

Did you see that the regulator recommends 47uF output, by the way?  1uF isn't nearly enough.  But if there's more stuff attached, I have no idea.

Tim

[ratatax]

I lowered the output capacitor for a strange reason. When bigger, I get a lots of audible whine (high pitched noise) from the load i'm powering (an ARM board with its regulators). With smaller value, there is no noise or at least it's not audible, and ripple it's much worse according to my scope. I feel it's not a very good solution, but I haven't found another way to remove this noise.

[T3sl4co1l]

Got a schematic for the ARM board?

Tim

[ratatax]

Yes, the current one I use is simply a Raspberry Pi 3B, full schematics here : https://www.raspberrypi.org/documentation/hardware/raspberrypi/schematics/rpi_SCH_3b_1p2_reduced.pdf

My regulator outputs 5V, that is directly fed into RPi's 5V input on the GPIO expansion (pins 2 & 4, labelled '5V')

Basically, the higher the capacitance the higher the noise. Putting an electrolytic cap and it will "squeak" a lot. Ripple amplitude on the scope is about the same, but waveform is way cleaner with high capacitance despite the increased whine sound. Probably just high inaudible harmonics that, once filtered, let the lower ones appear

[T3sl4co1l]

Oh, then it's probably using onboard capacitance.

The regulator doesn't have external compensation: you need to use the L and C that match its internal compensation, which are as recommended in the datasheet.  Too much of one or the other and it's probably going to squeal, which is likely your observation.

Looks like there's about 80uF on the 5V rail.  May even be a bit much for the regulator, in which case you may opt to replace it with one that has external compensation, so you can tweak it properly.  Can also add an R||L in series, so the regulator is somewhat isolated from the rPi's capacitance, and can function with its own (normal sized) capacitor.

Tim

[ratatax]

By R||L in series you mean like this ?



I don't know the name of this type of filter, could you help me guessing the (approximate) values of those components ?

Btw, thanks for the time you took to answer my posts

[T3sl4co1l]

Unfortunately, calculating it would require a model of the regulator.

I would guess maybe 10uH and R for critical damping, and C on the reg side as recommended.

Tim

[OwO]

The inductor will resonate with the input bypassing caps and give you a peak in the power supply impedance vs frequency. That combined with the negative resistance characteristics of the DC-DC converter on the input side (caused by the fact that higher input voltage => less current draw to maintain the same power and vice versa) can cause instability. All DC-DC converters have this negative resistance behavior so you need to design the input circuit carefully always. I would either add a snubber (series RC across the input but please simulate it) or just oversize the input caps so that the Q is limited (depends on the inductor esr).

[Al3579]

it has soft start and an input enable, have you tried playing with the enable voltage it has a threshold. Also I didn't see a cap on the soft start.
also the data sheet is vague how is a PWM power supply pseudo constant frequency of 500khz. Must mean the frequency changes with load but their was no graph for that haha.
What is the amperage rating of the wall wart and the load you are suppling. Might be better just getting a common mode choke bridge rectifier caps that will handle the load you are suppling. Looks like it meant to supply 125-150 watts of 5v power for displays and such.

Does anybody know what pseudo constant frequency is?

[T3sl4co1l]

It's probably a constant-on-time controller, or something like that, and "pseudo" just means, they measured it at that particular condition and make no claims about it otherwise.  Probably, it's similar at higher loads, but much lower at light load, or something like that.

Tim