Solar-powered soil-moisture probes, a receiver and a cloud dashboard, built to answer one question: does this plant need water?
Project Treebeard watches how the soil in a houseplant's pot dries out and gets watered again. A probe in the soil wakes about once a minute, takes a measurement, broadcasts it over Bluetooth LE and goes back to sleep, running on a small battery charged by a solar panel. A receiver on a shelf hears every probe, shows each plant on a screen and uploads the history to a phone dashboard. Four probe generations, MSV1 to MSV4 ("moisture sensor version"), have been built since April 2026.
Sam's role: hardware architecture, Altium schematic and PCB design, hand assembly and rework,
test planning, bench and field measurement, firmware requirements, and hardware validation.
Software implementation was AI-assisted and verified against physical hardware and recorded data;
the exact contribution boundary is documented in CONTRIBUTIONS.md. This is an
active engineering project, and the repository keeps the failures alongside the things that worked.
Four assembled MSV4 boards, an MSV3 probe in its pot, five probes sharing one pot for a side-by-side test, and the DisplayCenter receiver.
Interactive dashboard demo · 60-second engineering tour · Latest PCB and bring-up evidence
- How it works
- Start here for engineering roles
- The hardware
- Four generations
- Engineering highlights
- Status and next steps
- Repository guide
- Building and flashing
- About
The short engineering evidence map links each hardware claim to
photos, schematics, bring-up notes and measured results. The fastest paths are:
- PCB design and bring-up: four board generations, the ESP32-S3 receiver and fine-pitch fault isolation on MSV4.
- Validation and reliability: four-board mux screening, sleep-current measurement and explicit separation of measured, inferred and modeled results.
- Analog and power: the lock-in front end, electrode experiments and the 6 V panel / 4.49 V battery overvoltage root cause.
- System integration: BLE probes, receiver, cloud history and the clearly labeled synthetic-data dashboard demo.
- The probe. A narrow circuit board pushed into the soil. Sleeping almost all the time is what lets it run from a small rechargeable battery and a solar panel about 3 cm across.
- The receiver (DisplayCenter). A wall-powered box with a screen. It listens all the time, filters the readings, shows every plant, and uploads the history when Wi-Fi is available. The plants stay on its screen even when the internet is down.
- The dashboard. A web page for a phone with each plant's status and history.
flowchart LR
P[Battery and solar<br>soil probes] -->|Bluetooth LE| D[DisplayCenter<br>ESP32-S3]
D --> T[Tabletop display]
F[Phone on local Wi-Fi] <-->|Setup and status| D
D -->|HTTPS| S[Supabase<br>history and calibration]
S --> W[Hosted phone dashboard]
L[Laptop] -.->|Signed firmware<br>over Bluetooth| P
Plant names, artwork, probe assignments and Wi-Fi are set up from a phone. The receiver keeps raw and filtered history, flags probes that have gone quiet, and marks a "water now" region on each plant's meter, learned from where that plant was actually watered. Probes take signed firmware over Bluetooth without leaving their pots.
Why it is harder than it sounds. Two electrodes in soil give a number easily. A number that means "wet" or "dry" is another matter: the reading also moves with temperature, sunlight on the pot, fertilizer salts, and how tightly the soil grips the probe. Most of this project has been about separating those effects from the water.
Every board was designed in Altium. Each board page has the schematic, layout, 3D views and the Gerbers the board was built from.
A two-layer probe that measures the soil's impedance with a lock-in front end built from two analog switches, at eight frequencies from 250 kHz to 2.67 MHz. The blade carries two matching electrodes inside a guard ring on both faces, and a temperature sensor at the tip.
Main parts: nRF52840 (E73 module) · 2 × FSUSB42 analog switch · BQ25504 solar harvester ·
TPS7A0233 LDO · 0402 NTC at the tip
Board page ·
Schematic ·
Gerbers
A four-layer probe built around a capacitance-to-digital converter, with an analog multiplexer that swaps the two electrode nodes, and solar and USB charging.
Main parts: nRF52840 (E73 module) · FDC2214 capacitance-to-digital converter · TMUX1109
multiplexer · BQ25504 solar harvester · BQ25170 charger · TPS7A0233 LDO
Board page ·
Schematic (PDF) ·
Gerbers
A two-layer controller board that hears every probe over Bluetooth LE, drives a 480 × 320 touch display and uploads over Wi-Fi.
Main parts: ESP32-S3-WROOM-1U-N16R8 (external antenna) · AP2112K-3.3 LDO · 2N7002 switches ·
USB-C · buzzer
Board page ·
Schematic ·
Gerbers
A two-layer probe carrying a XIAO nRF52840 Sense module, which times how fast the electrode charges through its own GPIO pins. Solar charging through a USB charger IC.
Main parts: Seeed XIAO nRF52840 Sense · BQ25100 charger
Board page ·
Schematic ·
Gerbers
The two-layer analog prototype: an interdigitated comb electrode read by 555 timers, an RC low-pass filter and a comparator.
Main parts: 555 timers (astable and monostable) · RC low-pass filter · comparator
Board page ·
Schematic ·
Gerbers ·
Course report
| Approach | What it taught | |
|---|---|---|
| MSV1 | A comb electrode read by 555 timers and analog filtering | Soil moisture can be measured this way, but the analog output squeezed most of the useful range together |
| MSV2 | Time the electrode digitally and broadcast from a sleeping radio chip | A useful monitor is a whole system: sensing, power, radio, display, calibration and history all matter |
| MSV3 | A custom board with a capacitance-to-digital converter and solar charging | Resolution was never the bottleneck. Electrode geometry, soil temperature and the solar panel's voltage were |
| MSV4 | A new electrode, an impedance measurement instead of capacitance, and a panel that suits the charger | Still being learned |
MSV2 proved the system. The receiver, the phone setup, the cloud backend and the dashboard were all built around it, and it monitored real plants for months. Field use turned up the problems that make sensor products hard: sleep-current leaks that drained batteries, lost radio packets, calibration drift and stale web caches.
MSV3 got the platform right and the sensing wrong. It updates over Bluetooth in 30 to 50 seconds, with checks that stop an image built for one probe from landing on another. It draws about 3 mAh a day, roughly 50 days on a full battery with no sun, and every packet carries the reading, board temperature, battery voltage and status flags. But its reading moved with temperature and sunlight when the water in the soil had not. Its fine comb sensed only the half-millimeter of soil touching the board, so a protective coating took up a large share of what it could see, and the daily swing came in through the electrode rather than the electronics.
MSV4 changes three things at once. The electrodes are spaced to sense several times deeper. Instead of capacitance, the probe measures the soil's impedance and fits a small circuit model, and the tip temperature is sent alongside so the receiver can correct for it. And a 2 V solar panel stays well inside the charger's input limit, where MSV3's 6 V panel did not. So far: a window-light test produced a 108 mV battery rise over a morning and showed cutoff near 3.82-3.83 V; its first watering took soil resistance from 15.3 kΩ to 3.0 kΩ; and a probe now checks that it is in clean air before calibrating itself, after one calibrated against a hand. Its power draw is still modeled, not measured.
Problems that were measured to the bottom rather than patched around.
- Sleep current measured, then reduced. Disconnecting divider and status GPIOs during sleep reduced measured MSV3 current from about 65-71 µA to 10-15 µA. Measurement record
- Four boards screened with one discriminating test. A mux sweep initially suggested a dead part; repeated measurements after reseating showed the expected 8-15% switching step on all four boards and prevented unnecessary bypass rework. Triage record
- Fine-pitch assembly faults separated from design faults. MSV4 board 1 had three hidden solder faults from excess paste, including a 0 Ω short under R15. Resistance checks and controlled reflow recovered the board without a schematic change. Bring-up record
- A solar charger cannot protect a battery from the wrong panel. MSV3's 6 V panel drove current through the charger chip into the cell, which reached 4.49 V against a 3.89 V limit. Root cause
- Electrode or electronics? Four sensor combinations shared one pot so the electrode and the readout could be swapped independently. The MSV3 blade swung about ten times more over a day than an older electrode on the same kind of circuit, the circuit accounted for at most a fifth of it, and a blade on the shaded side of the pot was quiet. Write-up
- A lock-in impedance measurement on a microcontroller. MSV4 drives its electrodes at eight frequencies, demodulates in firmware and fits a circuit model. A simulation using parameters fitted from board measurements provides a consistency check within a few degrees of phase; it is not an independent predictive validation. Raw sweeps from the second board are stored in the cloud so other electrode models can be compared; a constant-phase-element model had the lowest residual on the documented soil capture. Notes · Electrode models
- Compared with a pot-mass reference. In one pot over eight days and one 460 g watering, daily mass measurements were used to compare seven candidate moisture metrics and temperature effects. The result is exploratory—not a calibrated accuracy specification or a general winner—and the write-up records where the estimators did not match the reference closely enough. Why the readings miss
- A receiver that stopped uploading for 6.7 hours with Wi-Fi up. Nothing was lost: 1,812 rows were queued and delivered after a reset. The receiver now records the real TLS error and restarts itself after a stall. Incident report
As of 2026-10-18. MSV3/fleet.json is the live record of where every board is.
| State | |
|---|---|
| MSV4 | Two boards running. Board 1 has been in a weighed Alocasia pot since 2026-10-08; board 2 is in a pot of its own and sends raw measurement sweeps to the cloud. More boards are being assembled from a batch of ten PCBs. |
| MSV3 | Five boards built. One is back in soil; the rest are on the desk or out of soil since a two-week comparison experiment ended on 2026-09-30. |
| MSV2 | One probe is still in a pot outdoors. |
| DisplayCenter | Running. It hears all three probe generations and uploads to the cloud. |
| Dashboard | Live. |
Next:
- A second weighed watering. Weights just before and after it will show whether MSV4's offset is the probe settling in or something each watering adds.
- Choose MSV4's moisture reading by the scale. Derived soil capacitance leads so far; its temperature correction rests on a few days and needs more dry days to confirm.
- Measure MSV4's real power draw. Every energy figure for it is still modeled.
- Bring up the next MSV4 boards from the batch being assembled.
| Folder | What lives there |
|---|---|
MSV1/ |
The analog prototype: board files and course report |
MSV2/ |
The XIAO nRF52840 probe: board files, firmware, experiments and field notes |
MSV3/ |
The FDC2214 probe: board files, enclosure, firmware, bootloader, test write-ups and the fleet registry |
MSV4/ |
The impedance probe: board files, enclosures, firmware, parts lists and test write-ups |
DisplayCenter/ |
The receiver: board files, enclosure, firmware, phone setup interface and test scripts |
cloud/ |
The database, its edge functions and the hosted dashboard |
docs/ |
Cross-project notes: known issues, roadmap and the engineering paper |
hardware/ |
Shared Altium libraries and parts ordering |
scripts/ |
Shared tools: the Bluetooth updater, the fleet scanner and the wired-programming utilities |
data/, dumps/ |
Captured measurements and probe diagnostic dumps |
Note
Dated documents are history and are never edited afterwards. Hardware moves between work
sessions, so MSV3/fleet.json is the only place that describes the present:
where each probe is, what is in its soil, what powers it and what has been reworked. Throughout
the repository, "the code is written", "it compiles", "it was flashed" and "it was seen working"
are four different claims.
The build helpers expect Windows PowerShell and the arduino-cli.exe that ships with the Arduino
IDE. From the repository root:
# MSV2 probe
powershell -ExecutionPolicy Bypass -File MSV2\scripts\arduino_probe.ps1 -Action compile
# DisplayCenter receiver
powershell -ExecutionPolicy Bypass -File DisplayCenter\scripts\arduino_display.ps1 -Action compile
# MSV3 probe
powershell -ExecutionPolicy Bypass -File MSV3\scripts\arduino_msv3_probe.ps1 -Action doctor
powershell -ExecutionPolicy Bypass -File MSV3\scripts\arduino_msv3_probe.ps1 -Action compile -Sketch arduino/MSV3_Moisture_BLE_Sleep
# MSV4 probe
powershell -ExecutionPolicy Bypass -File MSV4\scripts\arduino_msv4_probe.ps1 -Action doctor
powershell -ExecutionPolicy Bypass -File MSV4\scripts\arduino_msv4_probe.ps1 -Sketch arduino/MSV4_Moisture_BLE_Sleep -SensorId 10Warning
Always use the script for that generation, never the Arduino IDE Upload button or a raw
arduino-cli command. MSV3 and MSV4 need opposite clock settings: the MSV3 board has no
32.768 kHz crystal, and MSV4 has a crystal and no USB. Each script enforces its own board's rules
and refuses unsafe states. Run doctor first, and read Known Issues
before changing clock, sleep, watchdog or Bluetooth behavior.
- Firmware over Bluetooth: MSV3 uses
scripts/ble_ota_update.py --probe N; MSV4 usesMSV4/scripts/msv4_ble_ota.py <package> --board N. Run either with--preflight-onlyfirst. - Wired programming goes through the scripts in
scripts/, which wait for steady contact before touching a board because the programming pins are held by hand. The rules are inCLAUDE.md. - Receiver setup, wiring, libraries and Wi-Fi provisioning: the DisplayCenter guide.
- Cloud:
netlify.tomlpublishescloud/web/plant-dashboard/; database setup, security and receiver sign-up are in the Supabase setup guide. The receiver is local-first: it makes outbound requests only, should never be exposed to the internet, and keeps Wi-Fi passwords to itself.
Project Treebeard is a personal hardware project by Sam Amezcua. It includes working deployments alongside unfinished research, failed approaches, diagnostic firmware and manufacturing iterations, on purpose: that history is often the most useful part for anyone building a low-power environmental sensor.
© 2026 Sam Amezcua. All rights reserved. The repository is public so the work can be read and reviewed; no license is granted to copy, modify or redistribute the hardware designs, firmware or software.