SkyLive – static exploded view

3D-Ansicht benötigt WebGL (Grafikbeschleunigung). Statische Ansicht — auf der Projektseite dreht sich das echte CAD-Modell.

● Live · SkyLive

Live from 4 km above the drop zone — 14 ms behind reality.

The reveal

No flight controller. No BEC. One printed body. Four parts.

Every FPV builder expects a flight controller, a voltage regulator, a rat's nest of wiring. There is none of it here — a radio, a camera, a battery and an antenna, inside one printed body. The few connections that remain are minimal, and soldered once, properly, for flight. The restraint is the engineering.

The SkyLive sender exploded into its four parts — radio, camera, battery, antenna — plus clamp bars, T-pieces and screws.
The hardware

The sender, in your hands.

A GoPro-form-factor transmitter that rides on the helmet. Drag to inspect it, or pull it apart to see what's inside.

SkyLive sender — interactive 3D model (drag to rotate)
SkyLive – static exploded view

3D-Ansicht benötigt WebGL (Grafikbeschleunigung). Statische Ansicht — auf der Projektseite dreht sich das echte CAD-Modell.

TAKE-APART0%

drag to rotate · toggle assembled ↔ exploded · take-apart = scrub the explosion yourself

The physics

Five numbers the whole design hangs on.

One exact figure each — the small, deliberate numbers a build quietly depends on. Sourced where sourced, modelled where marked, honest either way.

R 1.55
The nose that matches the cable exactly.
The antenna cable drops all the way down into a round seat with clearance — insertion never fights the print. Then the T-piece's stem lands on it: a convex nose of radius 1.55 mm, precisely the cable's own radius, so it cradles instead of cutting. Two vertical M2 screws pull that nose 0.4 mm onto the cable — the screws are the clamp, measured 1:1 off a working reference build. A yank on the antenna loads the printed wall and the bolted nose, never the connector.
−0.2 mm
The squeeze standard (where squeezing is right).
The battery-lead saddles are under-sized by a tiny, deliberate amount: 0.2 mm less than the wire they hold. It's the line between gripping and crushing. The antenna clamp deliberately does NOT use it — coax hates being squeezed by tolerances (it detunes), so there the seat has clearance and two screws deliver a controlled, serviceable clamping force instead. Same goal, two honest mechanisms.
90°
The turn that dodges the null.
The omni doesn't stand upright like on an FPV quad — its coax runs horizontally through the case wall and the bell sits directly against it, axis pointing straight through the wall. An upright omni's donut pattern has nulls straight up and down: in head-up or head-down freefall that aims a blind spot exactly at the receiver. Laid sideways, the donut fires down, up and all around — signal toward the ground in every jump attitude. And the strain relief becomes trivially simple: the cable never turns, so the T-clamp just drops onto it and screws shut.
14 ms
Glass-to-glass latency.
From the lens in freefall to the picture on the ground TV is about 14 milliseconds — faster than a blink, faster than you can perceive as delay (manufacturer figure). It's not the internet and it's not a stream; it's a private digital radio link with nothing in the middle to buffer. The jump you watch on the screen is the jump happening in the sky, in the present tense.
4 parts
No flight controller. No BEC. One printed body.
The entire transmitter is four things — a radio, a camera, a battery, and an antenna — inside one printed body. Every FPV builder expects a flight controller, a voltage regulator, a rat's nest of wiring; there is none of it here. The few joints that remain are soldered once, properly, for flight — every joint earns its place. The restraint is the engineering: fewer parts, fewer failure points.
The depth

Show above, build below.

Everything above is the pitch. Everything below is the whole build — the chronological plan, the step-by-step assembly, the full bill of materials with dimensioned alternatives, the RF link-budget model, and the German-law reality. All install-free, all in the open.

More demos

Play the jump — a body is a shadow.

Radio doesn't go through a person well (−2…−12 dB by pose). Spin the jumper and watch each antenna's coverage move with the pose: the down patch — a legacy engineering study — aims a cone at the ground until the body swings it away; the donut omni fires down + up past the body. The omni is captive in the sender's wall, nothing to swap; on the ground, multiple fixed helix beams cover zenith and horizon, and the receiver fuses the branches.

BELLY
4000 m · FREEFALL
LIVEBELLY · FACE DOWN00:12:04
NO SIGNALantenna blocked by body
LINK
100%
● LIVE — image holdsdrop < 25 %
The omni is captive in the sender's wall — nothing to swap, no electronic switch.
On the ground: multiple fixed helix beams, zenith + horizon — the receiver rides the strongest branch.
Drag the jumper · or pick a pose · then try the down patch
Interactive

Spin the hardware. Read the millimetres.

The real CAD, live in your browser — assembled, exploded, and all three sizes (300: 59.5 × 39.5 × 48 · mid: 72 × 39 × 57 · 850: 71 × 40 × 56), with dimension tags on the geometry itself. Plus the antenna decks that argue the RF doctrine.

🧊
3D Lab
850 · mid · Mini 300 · exploded — orbit + mm-hotspots
🪂
Freefall Simulator
four helix beams, one fused link — drag her through the sky, watch the handover
🚧
Gate Simulator
drag the parts, break the rules — watch the export get refused
📡
Dual-antenna deck
the interactive RF argument · archive
EN · DE
🪂
Pitch deck
the earlier story, kept for the record · archive
EN · DE
SkyLive pitch — 2026 edition
the new one-piece sender, the lying-antenna doughnut, a self-exploding 3D model
DE · EN
Explainers

The idea, the journey, the budget, the heat.

The supporting figures — from "a dot in the sky" to the full 4 km link budget and the passive-thermal doctrine.

Left: what the ground sees today — a dot in the sky. Right: what SkyLive shows — the freefall POV live on a monitor, same moment.
A dot — or the whole jump. Today the ground sees a speck; SkyLive puts the jumper's own view on the screen ~14 ms later.
The signal's journey: camera, 1 W radio, antenna, 4 km of air, ground antennas, receiver, HDMI, to the big TV — about 14 ms.
The whole path the picture takes — helmet to the waiting-room TV, about 14 ms end to end.
Link budget closing 4 km: +30 dBm transmit, −119.8 dB free-space loss, honest RX gain, −90 dBm conservative threshold — head-down closes with about +9 dB, belly rides the threshold; a model, not a measurement.
Closing 4 km of sky — a link-budget model, not a measurement. When the numbers got worse, the worse numbers got published. Full model, pose tables & every assumption in build/rf →
119.8 dB
FSPL @ 4 km
≈ +9 dB
head-down margin · calc
~13 W
waste heat @ 1 W
~14 ms
latency
1 W
TX power
Passive thermal with an operating doctrine: 25 mW on the ground, 1 W only at door-open — in freefall the ram-air carries the 13 W of heat; off within 60 s after landing.
Passive thermal with an operating doctrine — vents, a thermal pad, the 200 km/h wind, instead of a fan.
Ground station: multiple fixed helix beams — one at the zenith, three fanned toward the horizon; the receiver takes the strongest branch.
Ground station: multiple fixed helix beams — zenith plus horizon; the receiver takes the strongest branch.
Latency: ~14 ms, before you blink; the old way is footage after landing.
~14 ms — before you blink. The old way is footage after landing.
Engineer feedback · round 2

A mechanical engineer took the red pen. The parts got tougher.

The printed prototypes came back from a professional review with one recurring verdict: everything that broke, broke while removing support — always at a layer seam. The lesson is topology, not thickness: a feature must grow out of the printed body inside the layer plane, like the GoPro teeth do — not sit glued onto a seam. Since 07-26 every fix below is ported to all three sizes — one principle, three senders, each re-proven by its own gates. Try it yourself:

CROSS-SECTION · DOOR PLATE + TOOTH Round 1 — tooth hangs off a 90° step CRACK — layer seam support removal ≈ a pull test Layers are drawn as printed — bottom to top. The seam between two layers is the weakest direction of every FDM part.

a geometry argument, drawn to the layer — not a strength simulation · the same principle moved the camera flange onto the shelf and put 45° fans under the door teeth

F0 · breathe

Mid grows back to 71 mm

A 2 mm "optimisation" had silently eaten the power-switch bay: 11.8 mm left for a Ø12.5 body. Reverted — and the switch check is now a hard gate, probed against the whole body.

gate: 13.8 mm clear behind the VTX · envelope ∩ body = 0.026 mm³
F1 · root it

Camera flange stands on the shelf

The clamp flange used to hang off a wall seam and tore off with the support. Now a foot welds it to the mid-storey shelf — with 45° fillets, in the layer plane.

point-cloud diff: change only in the target zone · 6/6 printability
F2 · bridge it

XT30 strain relief, rebuilt

Free-standing fingers with "holes that were barely there" became one wall-to-wall clamp bridge — both cables in grooves, a solid latch bar screwed over them. One latch part fits both sides.

screw cores now fully jacketed · latch v3: printed flat
F3 · straighten

Door teeth, straight and rooted

The 90° hooks and blind floor channels are gone. Straight teeth sit in blind notches — and since the flap has a rounded corner right at the tooth root, 45° fans tie each tooth into the full plate.

door ∩ body = 0.000 mm³ · swing-in gates unchanged
F4 + F5 · subtract

Fewer features, freer corners

The omni-cap side holders left without replacement — the rounded camera-side look wins. The roof lid dropped to two diagonal M3 screws; the freed corner belongs to the switch.

removal proven by diff — the old "gate" had probed inside the wall
One material

Seven aborted prints, one root cause

Colour metadata in the STEP exports kept flipping lab printers into multi-material mode. All three models now export colour-free — slice the STLs, nothing else.

16× COLOUR stripped · rule: print from Druck/*.stl only
The doughnut lab

Spin her. The doughnut doesn't care.

An omni antenna radiates a doughnut — and a doughnut has a hole. Point the antenna up (the default) and the hole points at the drop zone. SkyLive lays the antenna sideways: tilt it yourself, then hit spin — in freefall she turns constantly, and the signal toward the DZ must not care.

DROP ZONE ⌀ dead zone signal toward the DZ 0 %
TILT

idealized omni doughnut (sin² gain) — the geometry argument, not a range promise · tilt 90° and spin: the DZ never leaves the maximum

The ground station

It grows its own antennas.

The ground station listens through self-printed 5.8 GHz helix antennas — axial mode, RHCP: seven turns above a cup reflector lined with copper tape. Not a flat plate (the ground plane needs its 0.75 λ of copper and walls), and not the big cone either — the cone wins +4 dB at its peak and loses across the real elevation window. Minimax, again. Each helix rides a ball head, so elevation is set in the field, not baked into the print.

How many? One prototype first — bench blocks A/B and measurement D1 come before the second winding is ever wound. Then the receiver fuses four branches (one at the zenith, three fanned to the horizon, stock stubs off), and the self-built station scales the same pattern to seven. Honest gain: three estimators span 10.7–13.9 dBic — nothing here is measured yet, and the plan says so out loud.

The self-printed 5.8 GHz axial-mode helix antenna — seven turns on a skeletonized former, cup reflector lined with copper tape, ball-head mount with 1/4-inch tripod thread.

printed in ASA — a black PLA antenna sags on a summer drop zone (Tg 55 °C)