P0 · Contact
Wired, no RF. Establish placement, mechanical transfer, comfort, and wearer rejection.
Build path
Prove the riskiest thing first.
The shortest responsible route is mechanical signal → passive modulation → body-loaded RF → integrated reader → product. Each gate prevents months of beautiful work around a broken assumption.
Week one: wired throat signal
The first prototype is deliberately not wireless. Obtain a safe, isolated, battery-powered audio interface; a small piezo/PVDF contact element; compliant pads of several stiffnesses; a soft breakaway textile band; and an ordinary reference microphone. Do not connect a body-worn sensor to non-isolated mains-referenced bench equipment.
- Print the sensor pod and cap as loose fit-checks. Deburr and cover all skin-facing plastic with a removable compliant pad.
- Mount the passive contact element so the pad transmits vibration without bending the sensor leads.
- Record paired contact and room channels while the wearer reads the same short script at normal level.
- Repeat at several lateral neck positions, band tensions, head angles, and walking conditions.
- Score intelligibility, wearer/other-speaker rejection, motion artifact, comfort, and repeatability. Preserve anonymized test metadata—not incidental conversations.
P0 passes only if: the wearer channel is consistently intelligible enough for downstream restoration, other nearby voices are materially attenuated relative to the wearer, and the contact arrangement remains comfortable and removable. Numeric thresholds must be frozen after an initial exploratory set and before the confirmatory set.
Minimum responsible bench
| Capability | Minimum | Why it exists |
|---|---|---|
| Mechanical/acoustic | Battery-powered isolated recorder, reference mic, force gauge, calipers, interchangeable compliant pads | Separates contact failure from RF failure. |
| RF measurement | VNA covering intended resonances, spectrum analyzer or calibrated SDR, attenuators, directional couplers, 50 Ω terminations | Measures resonance, Q, match, spurs, and receiver headroom. |
| Containment | Shielded enclosure or access to an RF lab | Keeps unapproved cross-band experiments from radiating. |
| Safety | Breakaway tensile fixture, temperature measurement, documented stop protocol | Neck-wear and RF exposure are product requirements, not paperwork. |
| Evidence | Versioned test scripts, calibration log, anonymized IDs, preregistered pass/fail | Prevents anecdote from becoming a requirement. |
Prototype ladder
P1 · Modulation
Bench fixture, no person. Prove sensor voltage changes the measurable passive response in a shielded setup.
P2 · Phantom
Body phantom and enclosure. Map detuning, orientation, range, and repeatability.
P3 · Wearer
Supervised, conservative, reviewed. Combine contact mechanics and RF; no unsupervised take-home.
P4 · Reader
Replace research instruments with a constrained puck, room channel, indicator, local storage, and physical mute verification.
P5 · Pilot
Certified hardware, FTO cleared, jurisdictional consent UX, monitored small cohort, support and deletion operations.
The engineer’s order of operations
Read SAFETY.md, copy the relevant test from DVP.md, assign a unique build ID, photograph the setup, record instrument and calibration IDs, run the test, attach raw results, and update claims.csv. If the architecture changes, write an ADR in decisions/.
Do not do these yet
- Do not order injection molds.
- Do not tune an antenna in free space and call it wearable-ready.
- Do not train a large enhancement model before the raw channel is stable.
- Do not choose global launch bands from a US-only assumption.
- Do not publish range, accuracy, batteryless privacy, or “speaker identification” claims without a reproducible dataset.
- Do not use a normal Wi-Fi router as if it were automatically a coherent full-duplex backscatter reader.