switcher ringing noise

The theoretical loss is even larger. It's very difficult to make advanced discrete front ends without them, because single-ended amplifier circuitry doesn't have much in the way of supply rejection.

With a switcher whose output ripple is 100 mV, 140 dB gets you down to

If the suppression were only 85 dB, I'd need two of them in cascade. (I've done that on occasion--four poles, two transistors.)

Cheers

Phil Hobbs

The photo doesn't show much effective ground plane stitching. Loop length from the SBD, through the board, to IC contacts is pretty dicey, where SBD turn-off current is expected to flow.

Check a reactance chart, looking at the 40MHz line for familiar layout and component values. You started with a 400MHz ringing and stepped backwards to see 40MHz in earlier work. You have to ask yourself, as well, just how much it matters in your application. It helps if there's an actual problem that needs solving.

Probing has its own issues. Can the probe produce the coax waveform? What do the ground plane points look like on the scope probe? Do the coax outputs shift when the scope probe shifts or the scope probe (plus ground probe) is removed?

Does the coax output shift as you finger certain components? Choke bodies can be screened and grounded, if they carry a lot of noise for re-radiation.

Traces that connect to the measurement point, without local series impedance, can also act as a pick-up to pump current into local non-ideal decouplers. Sense lines are the most frequently ignored.

I scanned an old bundle of paper that sometimes saves time wasted googling crap. It's a big file just because that's what the scanner pumped out, so it won't be mounted for long. Page 3 (?) is useful for checking ground plane and wide trace L/W ratios for L.

That would be just as bad as letting the thing just grunt.

The laptop-type supply is rated 24v and 65 watts. If it's shorted, it makes a 100 ms 9 amp pulse about once a second. So maybe I can ignore the switcher startup, on the theory that the supply can brute-force the load up to +5, and then the switcher will start to switch.

Laptop type supplies must be designed to pull up nasty loads.

The 40 MHz ringing is in fact the coax up to the scope. If I add a DC block and set the scope to 50 ohms, the ring goes away. What I see then is a pretty nasty impulse with a bit of 170 MHz ring; that's what was shocking the coax.

Of course the DC block wrecks the low frequency response. I should hack a giant blocking cap and a source terminator onto my board, and run the scope hi-z again.

Electronics is fun. You get so many puzzles to solve. Poirot had it easy in comparison.

The 400 MHz ring from the LTM8078 is very real. We can see that everywhere in our box.

You don't have to do it like a wildman. ;)

Something like a 78L05 with a MOSFET on its output would charge it up quickly and then go away.

I've built a fair number of POC systems powered by random old laptop bricks.

Cheers

Phil Hobbs

A ferrite bead in series with the buck inductor might be worth trying. If the SMPS spike is using the inductor's winding capacitance as a conduction path, then the ferrite bead should definitely make a difference in that 10-100MHz range.

Have you ever tried these "amobead" parts? Might try one in series with the freewheeling diode.

I finally figured out why cap multipliers are useful compared to using an op amp and a reference. The op amp frequency response is lower than a single transistor. The cap multiplier has limitations that an op amp and voltage reference don't have, but the transistor can be faster. Both designs require a drop out voltage, so can dissipate significant heat.

When people talk about such dramatic improvements in one aspect of a design, I have to wonder how much difference it makes in performance. The overclockers use all manner of thermal paste when the thermal resistance of the few micron thick layer of paste has a lower thermal resistance than the metal of the heat sink because it is a much longer path. If you are going to optimize, why try to optimize the part that has the tiniest impact on the result?

There's still the load. Disconnecting the load and charging the cap won't help much. As soon as the load is connected, the output cap will collapse, the switcher will go to 100% duty cycle, the input current will equal the load current, and there we are.

Besides that, the 7805 and the mosfet and the timing would be a nuisance.

If the 24 supply can provide 1.5 amps, the max 5v load current, it will just work. I think. The switcher *is* the linear regulator during startup!

Right. My concerns are probably silly, given that things like this usually work.

Don't think so. Simple switchers do everything cycle-by-cycle, so if it doesn't collapse during normal operation, it shouldn't do it on the first cycle either.

Sure. It's just a possible alternative.

Sort of.

Cheers

Phil Hobbs

I think we want the switch node to have minimum activity, so low diode impedance might be best. I tried a fb in series with the switcher output pin and it was horrible. As I discovered today, the 40 MHz ring was actually the coax to the scope. Oops. Dogged persistance is a workable substitute for extreme intelligence.

This might work for startup:

Beads are nonwound components, and have appreciably lower parallel capacitance. The Toshiba link also covers wound "spike killers" which have a couple of turns of wire on a toroid, which typically means about 1pF of parallel capacitance, although their text doesn't seem to mentions this..

LTSpice lists a great many Würth Elektronik GmbH & Co. ferrite chip parts - not all of them by any means.

You do have to provide an alternative (capacitative) path to ground for the high frequency current that the ferrite bead is supposed to block. Without that the ferrite bead just produces a larger voltage spike at the output of the switcher, which isn't helpful.

Somebody smart enough to spell "persistence" correctly might have found the problem faster.

You do need capacitors to ground before the inductors.

Whatever.

There's also a lower frequency concideration where the two coax pick-offs exit the board at opposite ends, then connect to the same (or connected) monitoring instrument(s). The two coax lines form a loop, that will generate antiphase readings, as common-mode shield current is seen as dif mode by the monitor.

So, compare the two coax to see if the 'bump' is visible in both, inverted.

RL