switcher ringing noise

In discontinuous current mode, an asynchronous switcher will produce EMI at the free resonance of the inductor. If you don't mind the efficiency hit at low current, a diode + RC snubber would probably fix it.

We've started putting U.FL coax jacks on all out power supply outputs, so we can figure out what's conducted and what's pickup. Helps a lot. (They don't get populated except on first articles, but they're the size of a SOT23, so nobody cares.)

Cheers

Phil Hobbs

On a sunny day (Fri, 11 Mar 2022 11:39:10 -0800) it happened John Larkin <jlarkin@highland_atwork_technology.com> wrote in snipped-for-privacy@4ax.com:

Is the 10 nF 30 Ohm parallel to the diode a damping network? Use a series LC there tuned to 50 kHz to short it?

That said I do not rememebr those oscillations tried a different make inductor?

The LTM is a synchronous switcher, and my 2576 is running continuous.

Looking at the timings on by breadboard, the rings seem to start at the big di/dt current transitions in the schottky. But nothing we can do changes the ring frequency, so what's resonating?

That's a good idea. We should also always include some way to measure currents.

We have some cute little loop-antenna scope probes. We can park one above an LTM module and get a scope trigger from its internal inductor, then signal average our noisy signals and see which noise correlates to which switcher.

Yes. It reduces the 40 MHz ring amplitude a bit, but not 2:1.

The problem isn't at 50 KHz, it's the fast ringing on both switching edges.

This wouldn't normally be noticed. It's tens of mV rings at 40 or 400 MHz. It's beyond the frequency ranges of the visible components.

I guess we'll dump the LTM things and go with old, slow switchers, and then try to physically segregate them as much as possible, and add a lot of secondary filtering. Create clean and dirty zones on the board, draw a boundary line, and filter the power sigs that cross the line. That might work better for small 40 MHz nasties than for big 400s.

But what's resonating? It doesn't seem to be the pcb itself.

I thought we might have a guard-ring-SRD snap in the schottky diode, but any diode does it, and it rings on both switching edges.

I hear you.

Awhile back we did a small power supply board, in an effort to factor out the noisy stuff and put it inside a shield, so that we could concentrate on what we care about.

It used a TI LMR23630AFDDAR (clocked at 2.15 MHz) to make +13 from +24, which was then inverted by an AOZ1282 to make -16. The other rails were made using linears off those ones or off the +24 directly. (Making -16 from +24 is a bit of a strain for most integrated buck regulator chips that can go faster than 2 MHz.)

It worked fine until we turned on the AOZ1282, at which point the whole board became a mass of VHF uglies. The thing was, everything was some high harmonic of the 2.15 MHz clock synchronizing the TI chip, selected by microstrip stub resonances in the traces. We had 118 MHz ringing here, 183 MHz there, all initially very mysterious. Never did work right.

We've had good success with the 150 kHz Simple Switchers, e.g. the LM2594, using powdered-iron toroids and B340A Schottky catch diodes. Our QL01 nanowatt photoreceiver has one of those within a couple of inches of a very sensitive 10 megohm TIA with a 1 MHz BW, and the switching junk is invisible on the output even using a spectrum analyzer with a 10-Hz resolution bandwidth. But even that one has issues with ground integrity--if the board doesn't make good contact with the box ground, low-level harmonics of 150 kHz start showing up.

At this point we've decided we don't want to be power supply designers, so we use the 2W Murata gizmos with the embedded toroids, inside a board-level steel shield, with the whole works inside a brass or aluminum box with a laser-cut lid. (Laser cutting has recently become monstrous cheap--we pay about $2 per lid in quantity 10, with four-day turnaound.)

Those U.FL connectors are super useful in distinguishing between stuff that our boards are doing and stuff that comes in over the air. The amount of tail-chasing they save is astronomical.

Cheers

Phil Hobbs

We might make provision for one of those Laird shield boxes, just in case.

The next challange is to soft-start the +24 to +5 switcher. Those old parts just grunt at startup. The +24 supply has to deliver 1 amp to pull up a 1 amp load. The LM2576 has an enable pin, but it's not a soft start.

A time delay and huge amount of bulk capacitance on +24 is one way to do it. I have, I think, seven various goofy ideas for sorta or actually soft-starting this beast. Without a Spice model, I'll just have to try them.

On a sunny day (Fri, 11 Mar 2022 13:29:42 -0800) it happened John Larkin <jlarkin@highland_atwork_technology.com> wrote in snipped-for-privacy@4ax.com:

Yes of course I did mean that 40 MHz

It is likely the coil !

They will be immediately after the discontinuity aka Gibb's phenomena on a truncated Fourier expansion for a square wave. It may not be a resonance as such but a side effect of the slew rate limit of the device. It doesn't die away quickly enough to be just that though.

There is a hard high frequency cutoff in gain and some ringing is pretty much what you would expect on a square wave with a truncated Fourier expansion. It may be being exaggerated in time and amplitude by some unfortunate choice of component values providing Q > 1 in addition.

As Phil said some sort of snubber would be the most likely amelioration. There will be an efficiency hit though so you have to choose how quiet you need it vs what losses you can live with.

It is likely the coil !

Look at the coil construction, it is much like a 30 MHz or there about air coil I often use. Probably at that high MHz frequency the far away core material is ignored by them electrons. Leaves turns and circuit capacitance for tuning. I use these kind of coils, less spacing between winding and core material:

It is likely that the parasitic inductance of the diode/high-side switch (L_par) resonates with the switching node capacitance (C_par). In order that your dampening network works the resistors must be about R=sqrt(L_par/C_par). My gut feeling is, that your dampening resistor is too high to work sufficiently. At typical low inductance layouts one lands in the range of 2...10R for the damping resistors these days. If the resistor is chosen too high or too low the effect on the ringing is very small.

Best regards Stefan

There is an RC snubber to ground... see my schematic. The R value is about optimized, and the overall effect is a very modest reduction in the ringing amplitude, no visible effect on the ring frequency or Q.

I can find only one thing that has any effect on the ringing frequency: the +24 input voltage. Higher voltage results in a very slight increase in ring frequency.

It's Saturday, but I might go in and play with it for a couple more hours. I need to be in that part of town anyhow. It's better commute on Saturday.

It's probably good enough, with layout improvements and secondary filtering, but it's interesting and annoying.

Next issue is soft-starting this old beast, so the system always comes up. The 24v supply will be a wart type thing. We'll have a Cuk converter to make +24 into -5, and that chip soft starts. My part, +24 to +5, doesn't.

I could let the Cuk start up, sense its output, and then start up my LM2576... somehow. The "enable" pin is just on/off, so any soft start would probably involve the fb pin. Nuisance.

Check noise effect when scope probe/ground lead is removed/replaced/manipulated. Above 20mhZ, it's going to be radiated.

More ground bonds to PC ground backing near IC, on both ground plane edges, where cut by power train.

Move your ceramic decoupling caps closer to the IC body tab. Shuffle the polymer/ceramic positions, so both work in tandem.

Same with schottky and it's snubber. Take output gound out of switching current loop.

SchottKy RC R too big? small? Cap on flying node - R to ground plane.

Move Noise monitors closer to filtered nodes, or filtered nodes closer to noise monitors. Bare leads feeding sheilded coax? I arsk yer!

RL

The input and output monitors are coax. I'm monitoring the switch node with a 10x scope probe. Removing the probe has no effect on the 40 MHz ring on the output.

The bottom is all ground. Various jumpers/plier grabs/kluges to the ground have zero effect on the ring.

Different schottkies, or parallel schottkies, have no effect.

I don't think swapping the RC in the damper would affect 40 MHz.

An inch of wire flat on a ground plane won't have any effect at 40 MHz. It's only a 500 MHz scope.

The scope is hi-Z. It won't allow 50r and AC coupling. I might go to

If the oscillations are in the coil self-resonance wouldn't it make sense to put the snubber across the coil? Or put one on each end of the coil? Snubbing one end of the coil to ground isn't going to stop the current or voltage of the ringing across the coil.

You don't get Gibbs' ears on just any square wave--you have to use the wrong filter. ;)

There isn't, though. It's just a MOSFET, two poles at most.

Cheers

Phil Hobbs

Another approach is to precharge the output cap before enabling the switcher.

Cheers

Phil Hobbs

Not all wall warts start up nicely into BFCs, though. We ship a beautiful one with our gizmos, but 50 000 uF will make it misbehave.

Cheers

Phil Hobbs

It can be dicey to feed one switcher directly from another. The power conversion folk do know how to do this, but it requires using a spice model encompassing both switchers and the cabling and filter stuff between, as well as the loads. LTspice is what they generally use.

Nor would I be surprised if the switchers were interacting with one another such that their switching frequencies adjusted (by injection locking) to be in some small-integer rational ratio to one another.

If I recall, powered iron toroids have some internal damping, which will control ringing. As others have said, I'm thinking that what is bedeviling Larkin may be coil self-resonance.

In my experience, what is mostly done these days in power supplies for low phase noise electronics is a pair of regulators before the sensitive electronics. The first regulator (a switcher) drops the voltage to almost the final output voltage (and inverts the polarity if needed). The second regulator (analog) brings the voltage down to the voltage needed by the sensitive electronics. There are low-pass and EMI filters as needed before and after the switcher, and after the analog regulator. And, the design is verified by LTspice before prototyping.

I believe it. I've had the same experience with people trying to estimate the temperature of a transistor junction from six inches away. (Insert standard joke about drunk looking for car keys under the light.) The fix was to insist on a thermocouple glued to the AlN spacer between transistor casa and heat sink. Not perfect, but orders of magnitude better, cutting tail-chasing by a like ratio.

Joe Gwinn

Yup. They get pretty toasty at 2 MHz, for sure.

We generally use cap multipliers right on the switcher outputs. With two poles in the base circuit and one in the collector, you can get ~140 dB suppression in one stage at SMPS frequencies. Regulators won't get into that territory.

Yup. For testing I've been known to fuse the thermocouple into a heatsink using one of those big crude $150 transformer-based spot welders. Dramatically better thermal contact than using epoxy!

Cheers

Phil Hobbs