zBRETT
DE·EN
To the architecture →
Language: DE · EN
HARDWARE · FIELD-TEST BOM

Procurement,
with reasons.

For the MVP field test: 2 hubs (one of them the gateway) + 3 endnodes. Every line item has a reason. Two rules set the whole list — which radio chip, and how many watts at the solar hub.

Status: draft — not procuredAs of 2026-07-17
GET INVOLVED

Got compatible hardware? Become a field tester.

zBRETT lives from everyday use. If you already own a suitable LoRa device, help us test the end-user experience — as an endnode in a real city network. We're looking for volunteers with nRF52840 + SX1262 on the 868 MHz band (EU868).

What your device needs
MCU: nRF52840 (Cortex-M4F) Radio chip: Semtech SX1262 — not SX1276 Band: 863–870 MHz (EU868) RNode-flashable — or willing to become so
Typical candidates
RAK4631LilyGO T-EchoHeltec Mesh Node T114XIAO nRF52840 + SX1262ThinkNode M1MeshPocket Qi2 ⚑

Not sure your device qualifies? Just send us the model — we'll tell you whether it's RNode-capable. ⚑ = interesting, but the firmware port is still in progress.

Sign up as a tester ↵
01

The rule everything follows from

An RNode is a modem, not a computer. The Reticulum firmware on the board only does the radio. The real stack — RNS, LXMF, brettd and above all the DutyCycleInterface — runs on a host that talks to the modem over USB-serial (KISS).

RoleNeeds a host?Why
HubYes, mandatoryThe DutyCycleInterface is an RNS interface. RNS doesn't run on the board.
EndnodeYes — but the phone countsSideband on Android talks to the RNode over BLE/USB.
GatewayYes — plus IPFSKubo needs a Linux host anyway.

Every hub is board + SBC.

This is the most important sentence on the page. Where an overview lists “RAK4631 / T-Beam,” it implies the board is the hub. It isn't — and that doubles the parts list per hub, which is why solar becomes a topic at all.

02

SX1262 only. No SX1276.

Two entries in older documents are wrong and get replaced. The SX1262 correction is strategic, not cosmetic.

It saysThe truth isWhy it matters
Endnode: “ESP32 + SX1276”SX1262 (T-Deck, RAK4631)SX1276 is the old generation. SX1262 has +22 dBm, lower draw — and Channel Activity Detection in hardware.
Hub: “RAK4631 / ESP32”RAK4631 = nRF52840Cortex-M4F, 256 KB RAM. Not an ESP32. The B3 rationale (“no BLS crypto on ESP32”) doesn't apply here at all.

Decision A1 (duty cycle instead of LBT) was made partly because LBT needs hardware that can do it. SX1262 can do CAD. The upgrade path to LBT/AFA — 10× the budget — stays open, provided nothing with SX1276 is bought.

03

Bill of materials

3.1 · Hub (2×)
ItemPartQtyNote
RadioRAK4631 (WisBlock Core, 868 MHz)2nRF52840 + SX1262, 1 MB flash / 256 KB RAM
 RAK5005-O baseboard2Carrier board for the RAK4631
HostRaspberry Pi 4 (2 GB) or Pi Zero 2 W2Mandatory — carries RNS + DutyCycleInterface + brettd
 microSD 32 GB, industrial224/7 operation; consumer SD dies on the write cycle
Antenna868 MHz omni, ≥3 dBi, outdoor2The biggest range lever there is
 Lightning / surge protection2Not optional for roof mounts
 Coax LMR-200/240, ≤3 m2Keep it short — cable loss eats transmit power
PowerSolar 30–50 Wp · LiFePO4 12 V/20 Ah · MPPT · DC-DC 5 V2No Li-Ion. Sizing → §5
UplinkLTE stick or Wi-Fi range2The gateway role needs internet
EnclosureIP65, UV-stable, cable glands2 
3.2 · Gateway (1×) — on top of hub #1
SSD 128 GB via USB — Kubo pins all board CIDs. Not on SD: IPFS write load kills cards. Host possibly to Pi 4 / 4 GB — Kubo + brettd in parallel.

The gateway isn't a separate device — it's hub #1 with an SSD and Kubo. The diagram draws it apart only for legibility.

3.3 · Endnode (3×)
Variant A — integrated (2×)

LilyGO T-Deck Plus (868 MHz): ESP32-S3, SX1262, keyboard, display, 1400 mAh battery, GPS. For the MVP in RNode mode on a host — tests “will anyone carry it?”

Variant B — modem + phone (1×)

RAK4631 + baseboard flashed as an RNode, plus Android with Sideband. Tests the standard LXMF path with no special firmware.

3.4 · Bench / measurement rig (M1, M2)

This item isn't an add-on — it's the first order.

2× RAK4631 + baseboard · 1–2× Raspberry Pi 4 or Linux laptop (rnsd with packet logging) · 2× 50 Ω/SMA dummy load · 2× 20–30 dB attenuators. M1 decides whether the architecture holds.

04

Why these parts

RAK4631 over T-Beam for hubs

The deciding factor is the T-Beam's batch uncertainty: “SX1276/8 or SX1262” means you only know on unboxing. For devices meant to sit on roofs for years that's an unacceptable risk — it could wreck the LBT path. On top of that the nRF52840 draws ~2 µA asleep; on solar in a Berlin December every µA counts.

No off-the-shelf Meshtastic board

zBRETT uses Reticulum, not Meshtastic. The hardware overlaps, the firmware doesn't. A board is flashed to an RNode with rnodeconf — Meshtastic firmware is overwritten in the process.

LiFePO4 over Li-Ion

A 24/7 solar node in Berlin does ~300 charge cycles/year and sees −10 °C. Li-Ion takes permanent damage when charged below 0 °C. LiFePO4 tolerates it, lasts 2000+ cycles and doesn't burn.

Antenna before transmit power

At a 1% duty cycle transmit power isn't the bottleneck — time is. A better antenna improves the link without costing a fraction of a second of airtime. The only range gain that takes nothing from the budget. Money into antenna and height, not PA modules. And: never transmit without an antenna — it destroys the PA permanently.

05

What's still to be calculated

Solar sizing, Berlin, December.

The critical month averages ~1 sun-hour/day. From the Pi 4 base load (~3 W) plus the RNode (~0.5 W) follows panel and battery need — plus reserve for a week of fog. The result could be that the Pi 4 is too hungry and a Pi Zero 2 W or an nRF-native approach is needed. A real architecture question, not a procurement one.

Site selection.

Two hubs with line of sight or ≤3 km. Berlin is flat enough, but roof access, power and lightning protection decide more than the straight-line distance.

06

Procurement order

1

Bench

2× RAK4631 + baseboard, 1× Pi/laptop, dummy loads. Now — M1 blocks everything.

2

Hub hardware

Rest of the hubs, antennas, solar. After M1.

3

Endnodes

T-Deck ×2. Before phase 4.

4

Gateway SSD

Before phase 5.

Stage 1 is the only one that should be ordered now. If M1 turns out badly, the whole governor approach may change — and with it whether hubs even need a Pi or something bigger.

07

Open items

Calculate solar sizing for Berlin December

Decides Pi 4 vs. Pi Zero 2 W.

Confirm two sites with roof access and ≤3 km

Roof access, power, lightning protection.

Lightning-protection concept for roof mounts

An insurance question — not optional.

Is an nRF-native brettd feasible?

Would halve power draw and parts list, but it's a substantial porting effort.