fpv: Pavo20 Pro II follow-up, the flight controller swap that fixed the GPS

[foxis]

Single post, EN + LT, 11 files. Stacked on #27; merge that first and I'll retarget.

Root cause: a leaky inductor and a missing layout

The C1/C4 finding is the strongest evidence in either article, because it is specific and checkable.

• Both rails run the same chip: TPS63070 buck-boost, verified at 2-16 V input, 3.6 A switch current (your "3.5 A at 16 V" was close, but that figure is switch current, not output). Wildly oversized for this load, and it degrades GPS at essentially zero load.
• The ~2.5 mm part beside it is the inductor, found with an in-circuit LCR meter. It has no properly enclosed ferromagnetic core, unlike the larger ones on your other boards.
• The datasheet's recommended layout puts C1 and C4 immediately beside the inductor and the IC, in the tightest part of the switching loop. On this board they are absent altogether. Other caps may exist but sit further out.

So the components placed specifically to keep the hot loop small are exactly the ones that were dropped.

Two independent things now point the same way: the H-field probe beating the E-field probe (backwards, if trace E-field coupling dominated), and the missing loop caps. Together they support switching energy being inductively coupled into surrounding circuitry rather than conducted or radiated, which is why nothing you shielded helped, why shielded wire capped at 4-5 satellites, and why USB power changed nothing.

The diagram is mine, not TI's

Rather than reproduce TI's copyrighted Figure 51, I drew an original Graphviz comparison of the recommended hot loop versus the actual one, in your brand palette. Verified viz.js renders it in both languages. No copyright question, and it fits your diagram rules.

The caveat that cuts against you

I put your extrapolation point in, and then the counter-argument, because it belongs there:

Extrapolating the envelope upward, it plainly does not stop at 1 GHz, and there may well be further peaks up at L1, particularly if anything on the board is resonant there. But resonant peaks are precisely the thing you cannot extrapolate, so that remains a guess and not a finding. I still blame the inductor for leaking.

That keeps your conclusion while being honest that the app note data stops below your frequency.

Length: 10.8 min, kept whole

2374 prose words. Over the 10-minute guide, and deliberate: the root-cause section is 3.1 min and the build half is 4.0 min, so neither side of a split reaches the 5-minute floor you set. I trimmed ~350 words of flab getting here; further cuts would have removed substance.

Everything else

• Motor: not attributed to 4S. Wires tore, glue was not holding at the base, that motor had been wiggling more than the others. Strain relief.
• Satellite table argued both ways: 8-in-2-min vs 0-in-15 is a huge win, and proof the FC was dominant but not the only fault.
• Antenna placement with the rotate-to-face-me reasoning, motors soldered to pads, 115.8 g, closing on your transplanted-brains question.
• Previous article untouched - verified in the build: still says 5V BEC, no banner, same title and URL.
• LT updated to match, 9.6 min. Reviewer-note block now lists ten terms plus a note that the Graphviz labels are intentionally diacritic-free for font reasons.

Verified

Zero em-dashes, one Mermaid + one Graphviz diagram, no ASCII art, draft: false, 9 images resolving in both languages, hreflang pairing EN-LT, valid BlogPosting in both, post in the sitemap, reviewer notes stripped from rendered HTML, all images EXIF-stripped, blog/public/CNAME untouched.

Per your call: the old-SHA photo and the publish timing are both closed. Dated 2026-08-18 and it stays invisible until you merge and rebuild, which is what you want.

---

Lithuanian terminology queries

Moved out of the file and into this PR deliberately. An ANDRIUI PERŽIŪRĖTI comment inside a committed article reads as "someone else wrote this and I did not check it", and the repo is public. It does not belong in the artifact.

Terms I could not confirm against real Lithuanian FPV usage. Correct any and I'll update both versions:

Used	English
persodinti smegenis	transplanted brains (figurative)
šuntavimas / šuntuoti	decoupling / bypassing
artimasis laukas	near field
įtampos atotampa	strain relief
magnetiškai nesandarus	magnetically leaky
jungtuvo srovė	switch current
H zondas / E zondas	H-field / E-field probe
perjungimo mazgas	switch node
ankšti HF kilpa	tight high-frequency loop

Also: Graphviz node labels in the LT version are intentionally written without Lithuanian diacritics, for font rendering in viz.js.

Verified the LT source and both minified and unminified output contain zero HTML comments.

Pre-existing, not fixed here

The same pattern is already merged and live in your Meteor75 post, in public source:

• blog/content/fpv/meteor75-pro-ii-fighting-resonances/index.lt.md — ANDRIUI PERŽIŪRĖTI, plus "nesprendžiau nė vieno, palikta tau"
• blog/content/fpv/meteor75-pro-ii-fighting-resonances/index.md — DRAFT NOTES block

Not touched here, since that is published content and your call. Happy to strip both in a separate PR.

[cursor]

Bugbot is not enabled for your account, so this pull request was not reviewed.

Enable Bugbot in the Cursor dashboard to get automatic reviews on future PRs.