OK past midnight, I gave the board one last try before calling it a day and moving on...
... I increased the load even more, to see if the chip would switch to full frequency and what would become of the ripple. Found a beefy 15W 15R resistor. That increased the load to 600mA or so.
Ripple plummeted !
So much so that the fixed x10 Agilent probe that came with the scope, was attenuating a tad too much to zoom on the ripple.
So I replaced it with my usual Tek x1 / x10 probes. Of course doing that limits the B/W to 6 MHz or something, but still good enough to see all the crap coming out of the regulator... no shortage of that for sure.
Ripple as you can see is now only 1mVpp... if you conveniently omit the bursts that contain huge spikes. If you count the spikes then it's more like 5mVpp, which is still very good for a switching anything, especially a cheap one eh ?
So I tried to measure frequency... saying "try" because the waveform looks so noisy, there is no saw tooth to look at any more, it's just noise.... but well, you gotta measure something so....
If I measure between two big bursts, the packets that feature the huge spikes, then I get 100kHz. If I then zoom in to look at the smaller oscillations in between two bursts, then I get 900kHz.. spot on the theoretical switching frequency that we calculated, so that's quite nice. So assuming that's not just luck / coincidence, then that means that all these spiky bursts are at LOWER frequency that the switching frequency ?! How can that be possible ?! I mean of course I am OK having high frequency parasitic oscillations, but LOWER than the commanded frequency... no I don't quite get it.
So that means instead maybe that the chip at 600mA still is running at 100kHz not 900kHz, and that the 900kHz oscillations we see are just a coincidence... it's nothing to do with the 900kHz we set with that resistor.
Lots of fun.
OK taking everything apart and clearing the bench, more component sorting on the agenda tomorrow...
EDIT : tried to add averaging to clean the signal a bit but jo joy : the deep memory of the Megazoom makes it a no go : adding even a tiny amount of averaging, lie 4 8 or 16, slows down the redrawing tremendously !
With my modest Combiscope and ancient TDS, short memory so I can add LOTS of averaging to clean stuff up really well, with little to no impact on the trace responsiveness. That's great.
So for me at least that's one big problem with the Megazoom, so I am not getting rid of my Combiscope and TDS anytime soon.
Would be nice if the Megazoom allowed you (does it ? Didn't find it in the manual), to disable the deep memory in this particular case.
Also another scenario where the deep memory hurts is refresh rate when trying to find a runt pulse or something. You don't need silly high time resolution to find it, but rather high refresh rates.
I understand the modern scopes allow you to set yourself the trade-off between sampling speed and memory depth which is the best of both worls I guess. The merger between the older scopes, and the Megazoom revolution.
So yeah, I love the Megazoom but I will keep my other scopes as well, I want BOTH !