README.md

SkyLive

A skydiver exits at 4 000 m. Before the canopy opens, the drop zone is already watching through her eyes.

Live from 4 km — 14 ms behind reality.

_{the actual gate-verified CAD — spin it and watch the parts fly home yourself in the 3D Lab}

▶ Play the jump · 🧊 3D Lab · 🪂 Freefall Simulator · 🚧 Gate Simulator · ⚖ vs GoPro · 🎛 Antenna decks · Build it — one page · What the gates caught · The numbers · Legal (DE)

---

What if the whole drop zone could watch — live?

Today the ground sees a dot in the sky. Spectators, the waiting area, your own team — they follow the jump with the naked eye, and the footage arrives only after landing. The moment itself stays invisible.

SkyLive puts the jump on the screen as it happens. A helmet-mounted transmitter the size of an action cam sends a digital HDZero picture from ~4 km up, down its own 5.8 GHz radio link — no internet, ~14 ms — straight onto the big TV in the waiting area. Not a recording. The present tense.

---

Four parts. Two ways to wire them.

That's the entire transmitter: a radio, a camera, a battery, and a button. The simplicity is the design — and you pick your wiring path:

• Soldered (recommended for the flight unit): every lead cut to its true run length — shorter cables, less weight, nothing to rattle, the most robust joints a jump can ask for.
• Wago quick-build (great first build): three 221-412 lever clamps, zero solder, re-openable in seconds — perfect while you're still iterating.

	part	what it does	the real part
📡	Radio (VTX)	turns the picture into a 1 W digital signal, ~14 ms	HDZero Freestyle V2 (30 × 29 × 14 mm, runs directly on 3S — no flight controller, no BEC)
👁	Camera	HD skydive POV, 162°	HDZero Nano90 (ships in the VTX kit, powered over its MIPI cable)
🔋	Battery	~40 min at 1 W (calculated: 850 mAh / ~1.3 A)	Tattu R-Line 3S 850 mAh (XT30)
🔘	Switch	on/off — breaks the battery + line directly	12 mm latching push-button, panel-mount

The VTX side plugs in via its stock JST-GH 6-pin harness either way. Full step-by-step (with the three hardware-killer rules): build/BUILD_GUIDE.md.

The shell: an upright, two-storey GoPro-style case — battery downstairs behind its own tab-locked door, radio + camera upstairs under a screwed roof lid — printed in PETG/ASA (never PLA) with a sacrosanct 3 mm wall, long passive louver vents, and a GoPro mount underneath. Outer dimensions: 71 × 40 × 56 mm — genuinely action-cam-sized. Both short sides carry an identical T-clamp strain-relief interface at the top edge — guide slot, round seat, and a screw-driven nose clamp — so the antenna can anchor left or right; the unused side is closed by the same T-piece. It builds from the parametric script in build/cad/ and passes every geometry gate on each rebuild; this exact file set is what went to the printer.

Three sizes, one architecture. Don't take a photo's word for it — spin them in the 3D Lab, where every dimension tag is a real millimetre from the executed CAD. The 850 (71 × 40 × 56 mm) is the flight unit; the mid sender (69 × 38 × 48 mm) and the Mini 300 (59.5 × 39.5 × 48 mm) are the same design wrapped around the smaller Tattu 300 3S HV pack (measured 45 × 17.5 × 15.3 mm) — same T-clamp antenna anchors (literally the same printed T-piece), same tab door, same camera corner.

---

One antenna, anchored like a tool — not like an afterthought.

There is no electronic antenna switch on the sender — the omni rides outside on its semi-rigid coax, and the clever part lives where it belongs: on the ground.

The mount is a proper screw clamp, measured off a working reference build's STEP files: the Ø 3.1 mm semi-rigid coax drops from above through a guide slot all the way down into a round Ø 3.2 seat that passes horizontally through the case wall — the seat has clearance, on purpose. The clamping is the T-piece's job: its stem ends in a convex nose of R 1.55 mm — exactly the cable's radius — and two vertical M2 cap screws pull that nose 0.4 mm down onto the cable, pinching it into the seat. The screws are the clamp; a yank on the antenna loads the printed wall and the bolted nose, never the connector. Outside, the RHCP omni sits directly against the wall, its axis horizontal, pointing straight through it. That 90° turn is the whole point: an upright omni's donut pattern has its nulls pointing up and down, which in head-up and head-down would aim a null exactly at the receiver. Lying sideways, the donut fires down and up and all around — signal toward the ground in every jump attitude. Both short sides carry the identical interface, so the anchor moves left or right to suit the helmet setup — one T-piece design, printed twice, no drilling, no notching.

honest caveat	status
Body shadow, not the antenna, is the limiter in belly/sit poses (−7…−12 dB literature midpoints)	assumption, to be jumped
Nose clamp holding force and S11 with the coax clamped	MEASURE_ME — fit-print pull test + NanoVNA
CAD asserts a 0.4 mm nose-to-cable engagement (1.15 mm³) on every rebuild	calc, not a torque test

Two earlier antenna integrations — the fully encapsulated side-capsule omni and the down-firing patch shell — are preserved as engineering studies with their full RF derivations in build/ENGINEERING/antenna_capsule.md; the external anchored omni won on serviceability (swap an antenna in seconds, nothing to detune, one part to reprint).

The gain lives on the ground. You're tumbling; the ground isn't. A bigger helmet antenna buys ~2–3 dB; the ground antenna buys 10+ dB. But not by tracking — a mover has a zenith keyhole exactly where the jump happens. The doctrine is several fixed beams: one at the zenith, three around the horizon, and the receiver's 4-way diversity fuses the branches. The beams themselves are self-printed axial-mode helix antennas (5.8 GHz, RHCP, 7 turns, C/λ 1.00, 10.5° pitch, copper-tape cup reflector — the higher-gain cone loses in the minimax over real geometry despite +4 dB peak; HPBW ~37°). Gain, honestly: estimators span 10.7–13.9 dBic — measurement decides. Each helix sits on a ball head so elevation is set in the field, the receiver's stock 2 dBi stubs come off (HDZero fuses by data integrity — one bad branch degrades the whole picture), and a Wi-Fi scan of the drop zone happens before anything transmits (5.8 GHz video channels overlap Wi-Fi). The store-bought patch-plus-omnis set remains as the interim/comparison station.

---

The numbers — calculated, labelled, and published even when they got worse

	value	status
📡 Transmit power	+30 dBm (1 W) — 25 mW SRD for all tests, PMSE assignment for the event	planned path
📡 Free-space loss @ 4 km	119.8 dB (5.8 GHz, Friis)	derived
📡 Link margin @ 4 km	head-down ≈ +9 dB · back ≈ +3 dB · belly rides the threshold (−0.2 dB) · sit ≈ −2 dB — margins improve 2–3 dB per km of descent	calculated, not measured
🧍 Body shadow	−7…−12 dB with the side-mount offset (literature midpoints) — the single biggest uncertainty	assumption, to be jumped
🌡 Heat at 1 W	~13 W of waste heat. On the ground, in still air, no passive case can hold that — so the doctrine is 25 mW on the ground, 1 W only at door-open; in freefall the 200 km/h wind is the heatsink (4–8× surplus)	calculated
⏱ Latency	~14 ms	manufacturer figure

The full model — pattern math, pose-by-pose margin tables, every assumption and its direction of error — is in build/rf/, including an interactive link-budget explorer you can open in any browser. The thermal, structural and print derivations live in build/ENGINEERING/.

No overclaiming. A CAD boolean check is not a test. Nothing in this repo carries a measured badge yet — and when the 2026 recalculations made numbers worse, the worse numbers were published.

---

Build one yourself

Everything a re-builder needs is under build/:

• 📋 BUILD_GUIDE.md — both wiring paths (soldered flight build, Wago quick-build), the wiring map, the three hardware-killer rules, and the power/thermal operating doctrine.
• 🛒 BOM.md — every part with real EU prices (as of 2026-07).
• 📐 MEASURE.md — the dimensions you must caliper yourself (nothing in this project is guessed).
• ✅ VERIFICATION.md — how a CAD model is turned into a trustworthy printable part: a seven-layer defense-in-depth, the honest limits of gates vs. physical tests, and the release checklist.
• ⚖️ LEGAL_DE.md — the German regulatory situation, honestly: what is legal today (25 mW SRD), what the event path is (PMSE), and why 1 W under an amateur licence is locked pending clarification.
• 🧊 cad/ — the parametric build123d scripts (spec.py is the single source of truth for every dimension).

---

Status

Print released (2026-07). The concept, the electronics, the RF doctrine and the engineering derivations are done and published here; the case design is frozen, and the first fit-prints are back from the printer: the printed GoPro fork clicks into a real GoPro mount and the parts fit as drawn (builder fit test, 2026-07-15).

• ✅ Sender electronics bought and specified — four parts, two wiring paths (soldered flight build, Wago quick-build).
• ✅ RF doctrine derived and published (donut orientation, multi-beam ground station, legacy capsule study) — calculated.
• ✅ Thermal, structural and print-factor derivations published — calculated.
• ✅ Final case CAD (71 × 40 × 56 mm) builds watertight and passes every geometry gate — roof lid on 3 corner inserts, tab-locked battery door, twin T-slot antenna anchors, 3 mm wall. Print files released — and the first fit-prints are back: the GoPro interface locks into a real mount, the parts fit. Fit-verified by print (2026-07-15); strength and RF stay open below — a fit is a geometry result, not a pull test.
• ✅ Mid sender (69 × 38 × 48 mm) and Mini 300 (59.5 × 39.5 × 48 mm, Tattu 300 3S HV): full architecture ports of the final build — same T-slot anchors (the T-pieces are literally the same printed part), same tab door; the Mini has no power switch (the electronics storey has no room for one; power = plug the battery). build/cad/mid_sender.py · build/cad/mini_300.py, geometry-verified, not yet printed.
• 🔜 Then: thermal measurement (multimeter protocol is written) → antenna S11 with the coax clamped → clamp pull test → 25 mW range test → test jump.

Modelled → gated → printed — the provenance trail

Everything physical in this project traces end to end; nothing here is a mockup:

• Modelled. Three senders — 850, mid, Mini 300 — each generated by one executed build123d script (v3_min.py · mid_sender.py · mini_300.py), six printed parts per size: the one-piece two-storey body, the tab-locked battery door (M2 brass insert), the roof lid (3× M3 corner inserts), 2× XT30 clamp bar and 2× strain-relief T-piece (proud + flush). The GoPro two-prong fork is printed into the floor of the body — teeth 3.0, gap 3.3, drawn against caliper-measured hardware. All 18 release STLs live in build/stl/.
• Gated. Every STL passed the release chain before it was allowed to exist: watertight boolean, the nominal-envelope interior gates (the same arithmetic the Gate Simulator replays), and printability gate v2 — thin-wall plateau = FAIL, cosmetic cusp = WARN. What the gates caught on the way is documented in WHAT_THE_GATES_CAUGHT.md; the layer-by-layer method in VERIFICATION.md.
• Printed. The first fit-prints are back (PETG, per PRINT_DE.md): the printed GoPro fork clicks into a real GoPro mount, and the printed parts fit as drawn — builder fit test, 2026-07-15. That upgrades the GoPro row in the ledger below from printing to fit-verified. Still open and honestly labelled: thermal, S11 and the pull test — a part that fits is a geometry result, not a strength or RF result.

⭐ Star the repo — releases will carry the first real measurements and, eventually, the first freefall footage from the system itself. Building one, or flying camera and have opinions? Open an issue.

---

The jump, as a number line

4000 m ─┤ ██ EXIT      link margin: head-down +9 dB · belly −0.2 dB   [CALC]
3000 m ─┤ ██ freefall  ~200 km/h — the airstream IS the heatsink      [CALC]
1500 m ─┤ ██ canopy    margins improve 2–3 dB per km of descent       [CALC]
 300 m ─┤ ██ pattern   fixed ground beams — the receiver fuses 4 branches
   0 m ─┴─▓▓─ beer     footage was live the whole way down            [PLAN]

_{margins = the conservative single-beam floor from build/rf; the BoxPro's four fused helix branches ride 10–15 dB higher — fly it yourself in the simulator.}

Truth ledger — what is measured, what is math

The whole repo runs on a two-word doctrine: a CAD boolean is not a test. Current state of every load-bearing number:

number	value	status
case dimensions, all three senders	71×40×56 · 69×38×48 · 59.5×39.5×48	🟢 executed CAD, gate-checked
battery	58×30×22 (850) · 45×17.5×15.3 (mid + Mini)	🟢 measured with calipers
brass inserts	M3 Ø5×6 · M2 Ø3.2×3	🟢 measured
XT30 wire, coax jacket	Ø2.8 · Ø3.1	🟢 measured
GoPro teeth 3.0 / gap 3.3	printed — clicks into a real GoPro mount, fits as drawn	🟢 fit-verified by print (2026-07-15)
antenna clamp holding force	screw-driven nose, 0.4 mm engagement (calc)	🔴 MEASURE_ME — pull test pending
antenna S11, insert strength, snap cycles	—	🔴 MEASURE_ME
link budget @ 4 km, thermal model	full derivations in build/rf	🟡 CALC — to be jumped

🟢 measured · 🟡 calculated/derived and labelled · 🔴 open, honestly. When the 2026 recalculations made numbers worse, the worse numbers were published — that policy stands.

Honest note

A solo-built, prototype-stage project shared in full. Calculated values are marked as such and separated from what still has to be measured. Transmit power is regulated: the plan of record is licence-free 25 mW SRD for all development tests and a PMSE short-term frequency assignment for event operation — see DISCLAIMER.md and build/LEGAL_DE.md.

_{License: CC-BY-4.0 · CAD: build123d (Python) · Made by @SchoenTom}