H2-1. System boundary & performance targets (ring-specific, measurable)

Intent: Define “good” for smart-ring hardware using measurable acceptance criteria (power, sensor headroom, contact robustness, and link cost) without drifting into phone-app/cloud workflows or medical diagnosis claims.

Why this chapter matters: If the acceptance table is not locked first, later “optimization” becomes guesswork. Power, headroom, and contact reality must be proven with waveforms and counters—before any feature discussions.

H2-2. Power budget & duty-cycling architecture (the real ring moat)

Intent: Achieve multi-day battery life by engineering the ring as a controlled set of bursts (LED pulses, ADC windows, compute, BLE events) separated by true deep-sleep—then proving it with current waveforms.

• Shipping: deepest storage mode. Only a minimal wake source allowed (dock attach or long-press, if present).
• Idle (not worn): ultra-low baseline; periodic wear-detect probe (short, bounded).
• Wear-detect: confirm contact state using a low-energy discriminator (avoid keeping the full sensor chain awake).
• Measure burst: enable only the required rail(s), run LED pulse train + AFE settle + ADC capture inside a fixed window.
• Sync burst: batch data into short BLE events; avoid long “connected” periods that silently dominate the day budget.
• Charge: charge-control + protection. Measurement policy must be explicit (what is allowed/blocked while charging).

• Quiet windows: reserve conversion windows where RF TX is blocked and rails are stable (PPG/EDA capture stays inside them).
• Batching principle: store locally first, then sync in short bursts; avoid keeping the radio “warm” by frequent small transfers.
• Overlap matrix: explicitly prohibit LED+RF and LED+write overlaps; enforce with rail gating and event arbitration.
• Baseline discipline: if baseline is high, measure which wake source dominates (RTC vs GPIO vs dock bounce) before changing anything else.

• Rail current waveform: show µA baseline + burst envelopes; flag unexpected periodic spikes (wake storms).
• LED current pulse: verify pulse width/count and alignment to ADC window; ensure no extra “tail” current.
• BLE event current profile: integrate to µC/event; confirm event cadence matches the intended connection interval / advertising schedule.

Field-proofing tip: If users report “random reboot” or “battery drops fast,” start with the waveform: (1) baseline too high → hidden wake source, (2) periodic tall spikes → retry storms or overlap, (3) peak+droop coincidence → rail gating / UVLO margin problem.

H2-3. PPG AFE: optical front-end, noise, and motion artifact discriminators

Intent: Build a PPG chain that (1) stays out of saturation across skin/contact variability and ambient light, and (2) preserves usable SNR at low LED current. Motion handling is ring-safe: IMU is used only as a quality tag, not as an algorithm deep dive.

• LED wavelength (HW angle): choose for energy-per-useful-SNR, not marketing claims. Lower LED current budget reduces thermal/self-heating and extends duty-cycle margin.
• Photodiode area & capacitance: larger PD increases signal but raises capacitance, impacting TIA stability/noise and bandwidth margin. Balance PD size with TIA feedback values.
• Leakage paths: ring geometry and window reflections can create “false DC” that consumes headroom. Treat leakage as a measurable offset that must not drive ADC toward rails.
• Mechanical fit sensitivity: contact pressure and micro-gaps change optical coupling. Design for acceptable headroom across expected fit variance, not only “ideal press-fit.”

• TIA noise & stability: avoid chasing gain by default. If gain is too high, saturation appears first under strong ambient/leakage; if too low, SNR fails at low LED current.
• Ambient cancellation margin: keep cancellation output away from its limit; otherwise the system becomes fragile to sun/indoor transitions and window contamination.
• ADC dynamic range: lock a headroom rule: typical operation stays well inside rails; rare peaks should not clip during walking or ambient transients.
• Anti-alias filtering: protect the HR band by blocking out-of-band noise folding; aliasing often looks like “mysterious in-band noise” in PSD.

• LED pulse → blanking → ADC window: blanking hides switching and optical transients; the ADC window must sit after settle and before unnecessary drift.
• RF quiet window: block BLE TX during capture; otherwise RF burst coupling can raise the in-band noise floor and mimic motion artifacts.

Ring-specific takeaway: If headroom is not protected, the system becomes user- and light-dependent; if SNR is not preserved at low LED current, power budget collapses. The fastest debug path is always: ADC distribution → timing alignment → PSD in HR band.

H2-4. Temperature sensing on skin: placement, self-heating, and time constants

Intent: Make temperature meaningful on a tiny wearable attached to a warm, moving heat source by engineering the thermal path, sensor placement, and self-heating policy—then validating with step response and activity correlation.

• NTC (thermistor): flexible placement and fast local response, but reads strongly through its mounting path (pad, adhesive, metal contact).
• IC temperature sensor: stable transfer function and easier manufacturing control, but package and PCB copper can bias readings toward device temperature.
• Placement rule: closer to skin interface → more “skin-like” but more sensitive to moisture/sealing; closer to PMIC/battery → more “device-like” and more burst-coupled.

• LED pulses: short, high-power bursts—bias depends on thermal path from LED/AFE to the sensor node.
• RF bursts: periodic energy injection—can create pseudo-periodic temperature ripple if the sensor is near the RF/PMIC copper.
• Charging: the dominant bias source—battery and charger losses raise local temperature; treat “charging state” as a hard context flag.

• Wear/remove step response: characterize settling time and hysteresis; this is the fastest way to reveal whether placement is skin-coupled or device-coupled.
• Burst correlation: time-align temperature with LED/RF/charge activity; quantify whether bursts create measurable spikes or ripple.

Ring-specific takeaway: Temperature validity is defined by thermal coupling + time constant + context flag. Without step response and burst correlation, “calibration” is usually compensating for uncontrolled self-heating or poor placement.

H2-5. EDA/GSR on a ring: electrodes, biasing, and impedance reality

Intent: Make EDA usable on a ring by engineering the electrode–skin interface, bias/protection strategy, and a reliable “contact / no-contact” discriminator. The goal is to separate motion, dry skin, and poor fit from real impedance behavior, using measurable evidence rather than assumptions.

• Effective contact area: the usable area is set by pressure and micro-gaps; the nominal electrode size is not the measurement area.
• Pressure sensitivity: micro-slips during motion modulate contact impedance; treat it as an artifact source with recognizable signatures.
• Moisture state: dry skin raises impedance and noise; moist contact lowers impedance but can increase slow drift from interface polarization.
• Layout goal: make both electrodes see similar mechanical conditions; asymmetry often creates common-mode swings and false differential readings.

• Bias scheme: choose a bias level that avoids forcing the interface into non-linear behavior. Keep a margin to absorb slow polarization drift without saturating the front-end.
• Input protection: ESD and dock-contact events can introduce leakage paths; leakage looks like “mysterious long-term drift” and can shrink usable ADC headroom.
• Common-mode management: ring-scale grounds are noisy. Ensure the IA/AFE input common-mode stays inside its valid window across motion and digital bursts.
• ADC resolution: value is limited by drift and leakage. Prioritize stable headroom and known limits over chasing extra bits.

• Impedance vs time during wear: look for settling, drift, and rail behavior across contact changes; identify “state switching” zones.
• Raw ADC stability (dry / moist / motion): compare variance and saturation; confirm whether instability is physical contact or electrical coupling/leakage.

Ring-specific takeaway: EDA becomes usable only when invalid windows are labeled with an independent contact discriminator, and when bias/protection prevent leakage-driven drift from consuming ADC headroom.

H2-6. MCU + BLE SoC partitioning: what runs where (and why it matters for power/noise)

Intent: Avoid the classic ring failures: radio events coupling into analog measurements, and firmware preventing deep sleep. Partitioning and time scheduling must enforce quiet windows for AFE capture, while preserving local data integrity with counters and buffers.

• Ground bounce: RF/digital current pulses modulate analog reference. Mitigation is both layout and scheduling: block RF TX during capture.
• Rail coupling: shared supply droop or ripple shifts AFE bias/ADC reference. Use rail gating/filters and avoid high-current operations during capture.
• I/O activity: bus toggling and interrupts near sampling windows inject jitter and noise. Gate “chatty” interfaces during AFE conversion.

• Timestamps: monotonic time base; mark discontinuities after reset/undervoltage so gaps are visible.
• Sample counters: increment at acquisition; compare expected vs actual to expose missed samples.
• Ring buffer: bounded storage with explicit overrun signature; avoid silent overwrite without flags.

• Analog noise vs RF event timing: overlay ADC noise/PSD with RF event markers; confirm capture window is RF-quiet.
• Missed-sample / overrun signature: verify counters change exactly when buffer pressure rises or when scheduling violates windows.

Ring-specific takeaway: The system becomes predictable only when capture is protected by scheduling gates, and when local counters/buffer signatures expose violations (missed samples, overruns, or RF overlap).

H2-7. Energy management & rails: PMIC topology under tiny battery constraints

Intent: Keep analog sensors stable while the micro-battery droops and system load is bursty. The power tree must separate always-on (AO) from switched (SW) domains, gate bursts to avoid worst-case concurrency, and convert brownouts into predictable, logged degradations rather than reboot loops.

• Retention LDO: keeps critical state alive with predictable noise and leakage while the rest is gated off.
• Buck vs LDO rails: efficiency helps battery life, but AFE stability needs ripple/PSRR discipline and clean references.
• Load switches: define inrush, sequencing, and “what wakes what.” Gate SW domains so bursts can be serialized.
• PG (power-good): treat PG as a “rail truth signal” used by firmware to avoid sampling on unstable rails.

• Voltage-only estimation: easy but misleading under burst load; droop is not equal to true state-of-charge.
• Coulomb counting / fuel gauge: more reliable for decisions, but still needs temperature awareness and offset control.
• Practical goal: use battery data to trigger degrade policies (reduced LED current, delayed sync, block concurrency).

• UVLO + hysteresis: avoid “threshold ping-pong” that causes reboot loops at low battery.
• Graceful degrade ladder: reduce LED pulse current/length → reduce BLE sync rate → block overlap → stop high-sensitivity capture if needed.
• Event logs: on every reset/brownout, record reset reason + last known burst context (LED/RF/charge state) to pinpoint root cause.

• Rail droop during LED + BLE concurrency: capture VBAT + key rails; look for droop depth and recovery time, and whether PG toggles.
• Brownout/reboot event log: verify reset reason and context flags; confirm the device degrades instead of entering repeated resets.

Ring-specific takeaway: a stable ring needs a domain plan (AO vs SW), gated concurrency, and brownout policies that leave evidence (rails + logs) instead of reboot loops.

H2-8. Tiny dock charging: pogo-pin/magnet alignment, safety, and charge noise isolation

Intent: Treat the dock as a system-defining interface. Dock mechanics determine contact resistance and bounce, which in turn drive charging reliability, ESD risk, and noise coupling while charging. The charge path must protect against intermittent contact and enforce a clear “measure while charging” policy.

• Misalignment: reduces effective pin contact and increases resistance, creating voltage steps that disturb the charge controller.
• Contact bounce: intermittent open/close produces current “sawtooth” pulses and repeated charger state transitions.
• Contamination: sweat residue/oxidation shifts resistance over time, making charging behavior drift across weeks.

• CC/CV control: stable current requires stable contact; bounce forces retries and increases ripple on system rails.
• Thermal limits: ring enclosure traps heat; NTC placement and limits prevent local hot spots during charge.
• Termination robustness: termination criteria must tolerate bounce without false “charge done” events.

• Charge current vs contact bounce waveform: observe I_CHG while inducing small dock movement; confirm bounce causes state cycling.
• Charger ripple coupling into AFE: overlay VANA ripple with ADC noise/PSD; confirm artifacts match charge ripple timing.

Ring-specific takeaway: charging reliability is shaped by alignment and bounce, while measurement integrity requires a strict “disable or invalidate high-sensitivity capture during charge” policy plus clear test points for ripple evidence.

H2-9. EMC/ESD/waterproofing: protecting a high-impedance mixed-signal device on the body

Intent: Survive sweat, ESD, and RF while keeping high-impedance analog inputs clean. The design must force disruptive currents into controlled return paths (TVS/structure return) and away from AFE inputs. Waterproofing must be treated as an electrical constraint: sealing can change electrode behavior and create trapped-moisture leakage.

• Sealing changes electrode behavior: the electrode perimeter and dielectric environment shift effective contact and polarization signatures.
• Trapped moisture: sealed cavities can retain humidity and ions, creating slow leakage that biases high-Z inputs and drifts baselines.
• Surface leakage: residue (sweat/salts) lowers surface resistance; high-Z nodes need physical separation and controlled surfaces.

• Short, separated high-Z routing: keep EDA high-impedance nodes short and isolated from charge/RF/clock edges.
• “Closest clamp wins” placement: TVS/ESD diodes must be physically close to entry points to shape the current path.
• Return-path clarity: provide a defined return route (structure/ground path) so ESD does not seek a path through AFE inputs.

• ESD gun outcomes + rail glitches: record reset reason and capture VBAT/VSYS/VANA behavior near the event; verify analog rails stay within safe bounds.
• Humidity exposure drift: measure leakage-driven drift (sleep current shift, high-Z baseline drift, EDA raw stability change) after humidity soak.

Ring-specific takeaway: waterproofing and ESD are electrical constraints. The only stable outcome is a controlled ESD return path plus leakage-aware sealing that keeps high-impedance inputs isolated from moisture-driven drift.

H2-10. Calibration & production test: what you can actually test at scale

Intent: Convert “lab-good” into “mass-good” using scalable hardware checks. The production plan focuses on go/no-go windows and binning from measurable signals: optical sanity, temperature self-consistency, electrode continuity/leakage bins, and power signatures (sleep + bursts + charge acceptance).

• Test pads: expose VBAT/VSYS/VANA and at least one charge current sense point (or proxy).
• Deterministic modes: fixed LED pulse train, fixed BLE burst window, fixed “quiet mode” for analog capture.
• Counters/flags: sample counter, missed-sample counter, brownout flag, reset reason, charge-state flag.

• Go/No-Go thresholds: record min/max windows used for each test to catch drift across lots.
• Counters: sample counters + reset/brownout flags during burst tests to detect silent instability.
• Minimal jig description: dock jig (power/charge), optical dummy (遮光/反射块), electrode fixture (open/short/resistor bins).

Ring-specific takeaway: scalable production quality comes from deterministic test modes, pads, and thresholds. If the line cannot measure it quickly (nodes + jigs + counters), the field will measure it painfully.