Most hardware startups hide their schematics and inflate their costs. We publish both. Our defensibility was never the device — it's the network, the data, and your trust. Safecast opened its radiation monitors after Fukushima and became the world standard. We intend to follow that path.
No exotic parts, no magic. Consumer-grade MEMS barometers, a Wi-Fi microcontroller, GPS time sync — and the discipline to use them correctly.
HERD One, Proto-0 architecture. The device captures and compresses; detection happens in the cloud, where hundreds of stations are correlated.
Three design decisions carry the whole device. Three barometers instead of one — averaging uncorrelated sensor noise buys us a √3 improvement in the noise floor, the cheapest sensitivity upgrade in existence. GPS PPS time sync — for network correlation, microsecond-accurate timestamps matter more than any single station's sensitivity. A 72-hour ring buffer — home Wi-Fi fails, routers reboot, passwords change; the station keeps recording and re-sends when it's back online.
Why no on-device detection? Because one station can't tell a tsunami from a slammed door or a passing truck. The entire point of HERD is that a slammed door can't be phase-coherent across 200 km of coastline. The network is the instrument; the device is one honest microphone in a choir.
Full KiCad schematics and PCB files go to GitHub once Proto-0 passes validation against the reference instrument. Until then — the exact module-level wiring we are building from off-the-shelf parts. Enough to replicate our stand on a breadboard.
Proto-0 DIY stand, module-level wiring. Three barometers need two I²C buses: the BMP390 only has two addresses (0x76/0x77), so sensor #3 gets its own bus. GPS PPS goes to a dedicated interrupt pin — timestamps are pinned to the pulse, not to Wi-Fi latency.
The firmware for this stand goes public together with the first side-by-side recordings against the reference instrument. The GitHub link will appear right here — and yes, the repo will include the wind-inlet hose geometry too.
The same radical transparency as our $25 breakdown — at component level. First-batch costs vs. what becomes possible at 10,000+ units.
| Component | First batch | At 10k+ units |
|---|---|---|
| 3× MEMS barometer Bosch BMP390 class; Proto-1 A/B test: BMP581 vs DPS368 | $2.70 | $1.80 |
| ESP32-S3 module Wi-Fi, dual core: capture on one, uplink on the other | $2.20 | $1.60 |
| GPS module u-blox NEO-8M, position + PPS timing | $1.80 | $1.20 |
| microSD + card 72 h ring buffer | $1.20 | $0.80 |
| PCB, power, passives | $1.40 | $0.90 |
| Enclosure + wind inlet injection mold amortized at scale | $1.70 | $1.10 |
| Assembly, test, calibration | $1.50 | $1.10 |
| Packaging | $0.50 | $0.50 |
| Hardware cost per unit | $13 | $9 |
Estimates from our sourcing research (LCSC/JLCPCB pricing, Seeed/Alibaba ODM quotes), not supplier contracts — we'll publish actuals per batch, audited. Certification (CE/FCC) is a one-time ~$3–5k, not in the per-unit number. On top of hardware cost sit delivery to your door, support and returns — that's how $13 hardware becomes the $13 "first-batch cost" line in our $25 breakdown.
Infrasound is not audio. The signals we hunt live between 0.001 and 10 Hz, so 25 pressure samples per second is plenty — and slowly-changing pressure compresses 4–8× with delta encoding.
| Mode | What's transmitted | Traffic / month |
|---|---|---|
| Full default, home Wi-Fi | continuous compressed 25 Hz stream, batched into 1-minute packets | 30–80 MB |
| Eco metered / LTE connections | per-second band summaries; raw windows only on trigger or cloud request | 3–10 MB |
| Research Proto-0 stations | everything raw, no economy | 0.3–1 GB |
The eco-mode trick: raw data is always recorded locally. If the cloud sees a suspicious correlation among your neighbours, it requests your station's window retroactively. Full network science at messenger-app traffic.
Claiming a $25 device hears infrasound is easy; proving it is the whole game. Proto-0 runs side by side with an Infiltec INFRA20 — a research-grade infrasound monitor (0.05–20 Hz, 0.001 Pa resolution) used by amateur and academic networks worldwide. Every claim we make about our noise floor will come with the reference trace next to it.
The validation sequence: two DIY stands from off-the-shelf modules → a 10-board small batch on a proper PCB → side-by-side recordings against the reference, published raw. Recent peer-reviewed work (see our science page) has already shown consumer MEMS barometers detecting volcanic explosions at hundreds of kilometres — we're standing on published ground, not hope.
Full design files on GitHub once Proto-0 passes validation. Build your own, improve ours, point out our mistakes — every clone that feeds data into the network is a free station, not a competitor.
Sampling, filtering, compression, the trigger logic — all readable. If your data goes through our pipeline, you deserve to see the pipeline.
This page, plus audited per-batch actuals. The same rule as everywhere on this site: not a single number you can't verify.
Each station signs its data with a unique key provisioned at the factory. That stays closed — it's what stops anyone from flooding the network with fake stations and fake tsunamis. Open hardware, authenticated data.
Publishing everything means waiving patents on this design — consciously. We're not Bosch; a patent was never going to protect a $25 device. What protects HERD is a network of thousands of verified stations and the years of correlated coastal data nobody can clone retroactively. We'd rather own the dataset than a paper certificate.
Engineer? Scientist? Want to tear this design apart? We'd genuinely love that: herd.network.team@gmail.com