Skip to the content
4 layers24 × 115.7 mmPCB carrier, the printed circuit board in 3D

Carrier PCB and flight firmware for the js-rocket electronics sled

Size
24 × 115.7 mm
Copper layers
4
Thickness
1.6 mm
Parts
21
Nets
14
Pads
80
Tracks, vias
268, 6
Pours
2
Routed
Yes
KiCad
KiCad 10.0 (file 20260206)
Created
6 Aug 2026
Updated
7 Nov 2026

Overview

PCB carrier is an open-source printed circuit board by jwilleke, from the jwilleke/js-rocket-avionics repository on GitHub. It is a 4-layer board of 24 × 115.7 mm with 21 parts on 14 nets, routed with 268 tracks and 6 vias.

README

README.md

js-rocket-avionics

Carrier PCB and flight firmware for the js-rocket electronics sled.

This is one candidate payload for the rocket, not the payload. The rocket flies on ballast alone. The interface it has to satisfy — 40.0 mm bore × 150 mm, an M3 × 55 retainer at nose z 15, a ~50 g mass ceiling, an optional camera port — is js-rocket/docs/payload-bay.md, and that is the only page in that repo this project depends on.

Flying the rocket, not building it? Read docs/using-avionics.md — what the electronics do, and what to do before, during and after a flight. Everything below is for builders.

Separate from the rocket repo on purpose: js-rocket is geometry and documentation with no code, no package manager and no dependencies, and its CLAUDE.md says so explicitly. Copper, firmware and the design record — every decision and why, in docs/design.md — live here instead.

Why it is built this way is in docs/design.md. Every part this project needs is in docs/BOM.md, which is the single source of truth for part numbers and weights — plus what each part is, which board it serves, and what was rejected. What to change next time, and what can be bought ready-made, is in docs/next-electronics.md. What was actually bought is in docs/shopping-list.md — orders, costs, arrival status. The printed sled that carries these boards, and the PayloadAdapter it loads through, are rocket parts and stay in js-rocket.

Status: one two-sided board (PCB-carrier-design.md), placed, netted and routed. DRC: 0 violations, 0 unconnected. The generator routes the signal nets itself (carrier_router.py, #15); ground and 3.3 V reach every pad through the two inner planes. What the board is, both sides with a drawing, is hardware/PCB-carrier/PCB-carrier.md.

Do not order it yet. Left before ordering: the reprinted fit mock, fitted, and the pre-order review — regenerate the fabrication files from this board and read its connections back (#16).

What a flight records

Planned, not yet written. The recorder firmware on XIAO-ESP32S3-cam is not started — today a flight would record nothing. The GPS is not recorded at all; its only copy is what the ground receiver hears.

DataFromRecorded?Where it ends up
Acceleration and rotationLSM6DSO32, XIAO-ESP32S3-camplanned — 500 Hz, read from the sensor's FIFOPSRAM through the flight (~900 KB a minute), written to microSD in slices as it goes
Pressure, so altitudeBMP388, XIAO-ESP32S3-camplanned — in the same logthe same
VideoOV3660 on the Sense camera board, XIAO-ESP32S3-camplanned — from launch, with a few seconds before it, until landing is detected; idle on the padmicroSD, as it records
Position and GPS altitudeL76K-GNSS, XIAO-ESP32S3-lorano — broadcast over LoRa by stock Meshtastic every so often, stored nowhere on the rocketthe ground receiver and the phone on it — #23, ordered 2026-09-11

Two things that follow:

  • A reset of XIAO-ESP32S3-cam in flight loses the slice of the sensor log not yet on the card — the rest is already written. Whether the shared battery browns it out is #8
  • The sensor log has no GPS in it. The two XIAOs share power and nothing else, so position and motion are never in one record

Why the log sits in PSRAM rather than going straight to the card: design.md.

The interface to the sled

One board on the PayloadSled's centre line, hanging off the web on two standoffs: the tall side (both XIAO stacks) faces 270°, the camera port; the low side (sensors, standoffs) faces 90°, the web. The sled side is js-rocket#99, which moves the web toward 90° and drills it at the standoffs. Change these and the sled changes.

Value
Board24.0 × 115.7 mm, aft end at nose z 25.4, forward end 141.1
Thickness1.6 mm, 4 copper layers — OSH Park's only 4-layer thickness
Corner radius2.0 mm
Standoffs2 plain M3 holes (3.2 mm) at board (x, y) (12.0, 4.0) and (16.5, 110.5) — nose z 29.4 and 135.9, neither under a module. A female-female M3 hex standoff, 10 mm, between the board's low face and the web, a screw into it from each side, as on the fit mock — nothing soldered
XIAO-ESP32S3-cam centreboard y 18.0 (nose z 43.4), under the camera pad
XIAO-ESP32S3-lora centreboard y 41.4 (nose z 66.8), turned 90°

Board x runs across the board, y from the aft end, forward positive, in KiCad's top view (the tall side). Everything above is read out of the generated board by gen_carrier.py; the gerbers in fab/ predate it (#16).

Why 24 mm wide. It matches the web. The sensors fit across it or hang just past its edges, below the web — PCB-carrier.md.

Why 115.7 mm long. The camera fixes the cam XIAO at the aft end, and the tall side then carries the lora XIAO, the L76K-GNSS, the data-link jumpers and the buzzer end to end. The low side carries the battery as well as the sensors (layout B2, 2026-09-15): LSM6DSO32 aft, the battery behind the lora XIAO and the L76K, the BMP388 forward of it, turned across the board under the buzzer (#36). 90.4 mm until the battery moved onto the board.

Why 1.6 mm. It is the only thickness OSH Park makes a 4-layer board in — "63mil (1.6mm) nominal". The 1.0 mm this page carried until 2026-09-15 was never checked against the fab; the operator caught it on the printed fit mock ("looks too thin"). It is also the better board: the standoffs are now 106.5 mm apart (81 before 2026-09-15), and 1.6 mm is about 4× as stiff as 1.0 over that span (stiffness goes as thickness cubed). The low face — the standoffs' face — stays at r 0.5, so the sled does not change; the tall side moves 0.6 mm toward 270°. What that costs across the bore is in PCB-carrier-design.md

The XIAO stack

The expansion board — Sense or Wio-SX1262 — sits above the XIAO, on the B2B connector on its front face. The XIAO stands on two standard 7-pin headers, ~2.50 mm, onto the board. No cutout and no tall stacking headers are needed. Stack heights and the fit across the bore are in docs/design.md.

Connections

There is no schematic file. Nets are assigned to pads directly in the generator, which is more robust than hand-authoring schematic s-expressions for 11 nets and keeps one generator as the single source of truth. The cost is no ERC and no drawn diagram — so this table is the wiring diagram. Keep it in step with gen_carrier.py.

NetFromAlso reaches
GNDboth XIAOs, pin 13In1 plane, every module
+3V3XIAO-ESP32S3-cam pin 12 onlyIn2 plane; LSM6DSO32 VIN; BMP388 VIN and CS (I2C mode, #19)
+3V3_LORAXIAO-ESP32S3-lora pin 12the L76K-GNSS's 3V3 only — kept off the plane, so the two regulators are never paralleled
SDAcam D4 (pin 5, GPIO5)LSM6DSO32, BMP388
SCLcam D5 (pin 6, GPIO6)LSM6DSO32, BMP388
BUZZERcam D0 (pin 1, GPIO1)one buzzer lead (TP_BUZ); the PS1240 sits on the tall side's forward end
BUZZER_Bcam D2 (GPIO3)the other buzzer lead (TP_BUZ_G), driven in opposite phase to BUZZER for about +6 dB (#25)
GPS_TXlora D6 (pin 7, GPIO43)the L76K-GNSS's pad at the D6 position, which is the module's RX
GPS_RXlora D7 (pin 8, GPIO44)the L76K-GNSS's pad at the D7 position, which is the module's TX
LINK_GPSthe GPS link's 2-pin header, pin 2 (pin 1 is on GPS_RX)its 1 kΩ, whose far end is GPS_CAM
LINK_EVTthe event link's 2-pin header, pin 1its 1 kΩ, whose far end is EVT_CAM
GPS_CAMXIAO-ESP32S3-cam D1 (GPIO2)the GPS link's 1 kΩ: the cam XIAO hears the GPS (#26)
EVT_CAMXIAO-ESP32S3-cam D3 (GPIO4)the event link's 1 kΩ: event text out
EVT_LORAXIAO-ESP32S3-lora D1 (GPIO2)the event link's 2-pin header, pin 2: Meshtastic's Serial module receives here. Stock Meshtastic leaves D1 free (#26)

GPS_TX and GPS_RX are named from the XIAO's side. Wire the L76K-GNSS by position, never by its silkscreen word — L76K-GNSS.md.

The pin map, and an error worth not repeating

The KiCad footprint numbers its pads 1–14 with no signal names, DIP-style — 1 at top-left, down to 7 at bottom-left, 8 at bottom-right, up to 14 at top-right. Seeed's pinout gives the meaning:

text
pin  1  2  3  4  5  6  7   8  9  10  11   12   13   14
sig  D0 D1 D2 D3 D4 D5 D6  D7 D8 D9  D10  3V3  GND  5V

An earlier revision of the generator put GPS on pins 6/7 and I2C on 4/5 — which are D5/D6 and D3/D4, every one off by one. Caught by reading the mapping back out of the saved board rather than trusting the script's own output. The generator now names pins (XIAO_PIN["D6"]) so the numbers never appear by hand.

Each 1×7 header numbers its own pads 1–7. The second row's pad 1 is XIAO pin 8, at the far end, so xiao_pin = pad + 7 running back toward the USB-C. This file once said 15 − pad, which is #18's mirrored row, written down as the rule.

The GPS is an L76K

The MAX-M10S was rejected on 2026-08-06: 44.2 × 30.5 mm — wider than this 24 mm board — and ~$60. The chosen part is Seeed's L76K GNSS for XIAO (109100021) — 18 × 21 mm, UART on D6/D7. It takes its own footprint on the tall side, forward of the lora XIAO (where), and its antenna is a separate part on the sled. Pinout, power and orientation: L76K-GNSS.md.

The same staleness once reached the barometer: BMP390 → BMP388, settled on 2026-08-06 when the 390 went out of stock. Same BMP3xx driver, same 25.5 × 17.8 mm STEMMA QT outline, address 0x77, and its pinout confirmed off the physical part.

Battery — read before assembling

BAT+/BAT− are underside pads on the XIAO, not brought out to the castellated edge, so the battery cannot reach a XIAO through the headers. The battery reaches each XIAO through a Y-cable, off the board, and a short soldered pigtail on its BAT pads.

  • Solder the pigtails before the XIAO goes onto its headers. The board, not the expansion board, is what covers the pads.
  • On battery power there is no voltage on the 5V pin — nothing can be fed from a XIAO's 5V rail.
  • Both XIAO chargers sit in parallel on one battery. Charge through one USB port at a time, and read why that is not sufficient on its own.
  • The battery is not on the board. A Y-cable joins it to both XIAOs' pigtails (LiPo-500mAh.md). Check every new cable's polarity with a meter; reversing a LiPo into a XIAO destroys it.

No RF on this board

The L76K and the Wio-SX1262 each carry their own U.FL connector, and both antennas mount on the sled, so no RF ever crosses the carrier. It is a purely digital and power board of 9 nets, which is what makes it tractable to generate and verify headlessly.

Building

Requires KiCad 10. The generator uses KiCad's bundled Python so pcbnew matches the CLI exactly.

bash
KPY=/Applications/KiCad/KiCad.app/Contents/Frameworks/Python.framework/Versions/3.9/bin/python3
CLI=/Applications/KiCad/KiCad.app/Contents/MacOS/kicad-cli

$KPY hardware/scripts/gen_carrier.py
$CLI pcb drc  --format json -o /tmp/drc.json hardware/PCB-carrier/PCB-carrier.kicad_pcb
$CLI pcb export gerbers -o fab/gerbers/ hardware/PCB-carrier/PCB-carrier.kicad_pcb
$CLI pcb export drill   -o fab/gerbers/ hardware/PCB-carrier/PCB-carrier.kicad_pcb

The board is generated, not hand-edited. .kicad_pcb format shifts between KiCad releases; building it through pcbnew means the file is written by the same code that reads it. Edit hardware/scripts/gen_carrier.py and re-run — do not edit the board file directly, or the next run overwrites you.

gen_outline.py is the superseded 2a-only version, kept for reference.

DRC must report 0 violations before anything is ordered.

Remaining stages

StageState
2a — outline, standoffs, stackupdone, one two-sided board
2b — nets + footprintsdone — every footprint placed and netted
2b — GND/+3V3 planesdone, In1 and In2 filled
2c — placementdone in the generator, pending review of the drawing — #14
2d — signal routingnot started — #15
2e — Gerber + drill for fabchain proven; needs a finished board

Do not order copper before breadboarding. A layout error costs ~$43 and two weeks; a wiring error costs minutes.

Fab target is OSH Park. The 24 × 70 mm version quoted $26.00 for 3 copies — OSH Park's 4-layer price is $10/in² for three — so 24 × 115.7 mm (4.30 in²) should land near $43. Re-upload to confirm; three copies — one to fly, two spares.

OSH Park accepts the .kicad_pcb directly — confirmed by upload, not assumed. It read the layer count and outline unaided, so fab/gerbers/ is not in the ordering path and exists only as a check that the export chain works.

Parts

21
QtyValueFootprintReferences
6PinHeader_1x07_P2.54mm_VerticalPinHeader_1x07_P2.54mm_VerticalJ_CAM_A, J_CAM_B, J_L76K_A, J_L76K_B, J_LORA_A, J_LORA_B
6TestPoint_THTPad_D2.0mm_Drill1.0mmTestPoint_THTPad_D2.0mm_Drill1.0mmTP_BUZ, TP_BUZ_G, TP_CAM_BAT_N, TP_CAM_BAT_P, TP_LORA_BAT_N, TP_LORA_BAT_P
2MountingHole_3.2mm_M3MountingHole_3.2mm_M3SO1, SO2
2PinHeader_1x02_P2.54mm_VerticalPinHeader_1x02_P2.54mm_VerticalJ_LINK_EVT, J_LINK_GPS
2R_Axial_DIN0204_L3.6mm_D1.6mm_P5.08mm_HorizontalR_Axial_DIN0204_L3.6mm_D1.6mm_P5.08mm_HorizontalR_LINK_EVT, R_LINK_GPS
1PinHeader_1x05_P2.54mm_VerticalPinHeader_1x05_P2.54mm_VerticalJ_LSM_A
1PinHeader_1x08_P2.54mm_VerticalPinHeader_1x08_P2.54mm_VerticalJ_BMP
1PinHeader_1x09_P2.54mm_VerticalPinHeader_1x09_P2.54mm_VerticalJ_LSM_P

Similar boards

  • 4 layers26.7 × 21.9 mmBuck Converter, the printed circuit board in 3D

    Buck Converter

    murexrobotics/electrical

    Highly integrated electronics design and printed circuit boards (PCBs) for the MATE ROV Competition. Attempt the impossible.

    21 parts192MIT
  • 4 layers26 × 30 mmDay 01 Wi Fi Halow, the printed circuit board in 3D

    Day 01 Wi Fi Halow

    i2cjak/Framework_Expansion_Cards

    In which i2cjak creates a new expansion card design every day until he is given a NICE Framework laptop

    21 parts95CERN-OHL-P-2.0
  • 4 layers20 × 48.5 mmMidea WiFi Dongle, the printed circuit board in 3D

    Midea WiFi Dongle

    reneklootwijk/mideahvac-dongle

    This repository contains the hardware design of a WiFi dongle for Midea-like airconditioners and humidifiers. Midea provides a WiFi dongle (SK102/SK103), either as optional feature or built-in, to control the appliances…

    21 parts75
  • 4 layers80 × 80 mmZoxnoxiouscard, the printed circuit board in 3D

    Zoxnoxiouscard

    brer-rabbit/zoxnoxious

    Analog synth / software interface

    21 parts68GPL-3.0
  • 4 layers90 × 75 mmCM4 2 Backup, the printed circuit board in 3D

    CM4 2 Backup

    olvint/CM4-NAS-MiniPCIE

    Raspberry Pi CM4 based NAS solution board

    21 parts51
  • 4 layersTiav3, the printed circuit board in 3D

    Tiav3

    OpenSiPM/sipm-bias-control

    Single channel silicon photomultipliers for high-speed fluorescence imaging

    21 parts44
  • 4 layers88.9 × 20.3 mmGW4194-SOJ DC, the printed circuit board in 3D

    GW4194-SOJ DC

    garrettsworkshop/RAMSIMM-30

    4 MB and 16 MB 30-pin DRAM SIMMs

    21 parts34

Questions

How big is the PCB carrier PCB?

It measures 24 × 115.7 mm and has 4 copper layers, 1.6 mm thick.

What parts does the PCB carrier use?

21 parts. The full list of its parts is on this page, with links to each part's symbol and footprint where PCBKit's library has them.

Where can I download the PCB carrier KiCad files?

From the jwilleke/js-rocket-avionics repository on GitHub, at the commit this page was made from: the board is hardware/PCB-carrier/PCB-carrier.kicad_pcb. PCBKit opens the board in the browser, with nothing to install.

Can I use the PCB carrier design in my own project?

Its repository states no licence, so its author keeps all rights: ask jwilleke before reusing the design.