A smart pendant / chest wearable is a sternum / upper-chest worn module (pendant or clip-on) that measures heart rate (HR), optionally HRV, and respiration (via BioZ/biopotential or motion proxy), plus posture states (via IMU), and delivers event alerts, local logs, and BLE sync. This page focuses on hardware evidence: sensor feasibility, placement constraints, reliability counters, and system boundaries.

Not this page: sports chest-strap coaching, smart rings, EMG rehab patches, sleep EEG headbands, cloud/backend tutorials, algorithm deep-dives.

Form factors (define measurement conditions, not just shape)

• Pendant (hanging): higher risk of swing/rotation → posture instability and contact-dependent sensing becomes fragile. Expect more “quality gating.”
• Clip-on (clothing clip): placement can shift with fabric thickness → “contact” may alternate between skin-coupled and cloth-coupled.
• Sternum module (close chest coupling): best signal potential, but higher exposure to sweat, touch ESD, thermal comfort, and antenna detuning by body.

Outputs (each one has a “validity condition”)

• HR: valid when contact quality is stable and motion tier is within spec; must expose dropout rate and quality flags.
• HRV (if supported): requires stricter quality gating (contact + timing stability). Treat as best-effort output, not always-on.
• Respiration rate: either BioZ/biopotential-derived or IMU proxy; must carry a confidence signal and degrade under high motion.
• Posture states: depends on mounting stability; must include hysteresis and re-calibration triggers (placement changes).
• Event alerts: should be traceable to evidence (motion + quality + battery/thermal state), not “magic decisions.”
• Local logs: timestamped samples/events with sequence counters and gap detection to prove integrity without cloud dependence.

Evidence chain (sensor → required physical condition → common failure evidence)

• ECG / biopotential → stable skin coupling + controlled impedance → evidence: lead-off/contact flags, baseline wander, input saturation, intermittent dropout.
• BioZ respiration → intact injection/measurement loop + stable coupling → evidence: injection amplitude anomalies, phase jitter, respiration “doubling” under motion.
• PPG (only if true skin contact exists) → light blocking + stable pressure → evidence: ambient leakage, pulse disappearance, accel-correlated noise.
• IMU posture → consistent mount orientation + limited swing → evidence: posture flipping during stillness, temperature-correlated drift, placement-dependent bias.

The purpose of this chapter is to ensure every later design choice can be audited by measured states (contact quality, motion tier, rail health, log integrity).

Convert “health + posture + breathing” into pass/fail KPIs that can be audited by waveforms and log counters. In wearable reality, user trust is driven by availability (few dropouts), stability (no random flips), and integrity (no silent data gaps) more than “perfect numbers” under ideal lab conditions.

HR (heart rate) — engineer-grade KPIs

• Update rate: how often HR is refreshed under each motion tier (rest / light walk / daily activities).
• Dropout rate: count of “no valid HR” intervals per hour/day; publish it as a reliability metric.
• Recovery time: how quickly HR becomes valid after contact returns (contact bounce is common in pendants/clip-ons).
• Quality flag coverage: every HR point should have a signal-quality / contact-quality indicator.

Why this matters: HR failures often trace back to contact instability, AFE saturation, or rail droop during radio/haptics bursts.

Respiration — reliability & false-peak control

• Latency: respiration rate update delay (window length) must match the intended use-case (alerts vs trend logging).
• False peaks / doubling: track and minimize “rate jumps” correlated with motion or haptics events.
• Confidence gating: under high motion, degrade output gracefully (hold-last, lower confidence, or pause output).
• Activity dependency: define “where it must work” (rest + light activity) and “where it may degrade” (high motion).

Practical rule: respiration quality must be judged against motion index and contact state, not just filtered waveforms.

Posture states — stability, hysteresis, drift tolerance

• State stability: no rapid flipping during stillness (a hallmark of pendant swing or mounting offset).
• Hysteresis: thresholds must include hysteresis to avoid chattering at boundary angles.
• Drift tolerance: posture baseline should not slowly drift with temperature or time-on-body without detection.
• Placement robustness: define when re-calibration is needed (e.g., clip moved to a different shirt position).

Posture issues are often mechanical + IMU bias problems, not “algorithm bugs.”

Wear-time & comfort — power budget + charging experience

• Average current budget: break down always-on vs measurement vs BLE sync vs haptics bursts.
• Recharge cadence: how often users must recharge in real wear patterns (not just lab duty cycles).
• Thermal comfort: charging and high-burst scenarios must respect skin-contact comfort (monitor coil + enclosure temperature).
• Data integrity across charge events: charging interruptions must not cause silent data gaps (log stop reasons + sequence gaps).

Evidence checklist (KPI → what to measure/log)

• HR dropouts → lead-off/contact flags, AFE saturation indicator, UVLO/brownout flags, sample gap counter.
• Resp false peaks → motion index, contact state, haptics event timestamps, (BioZ) injection amplitude/phase sanity counters.
• Posture flips → IMU bias/temperature correlation, mounting offset change detection, “stillness” classifier counters.
• Wear-time complaints → burst peak current (radio/haptics), BLE retry/disconnect reasons, charge start/stop reason, coil temperature.

These evidence points become the backbone for later chapters: power rails, haptics coupling, wireless charging, and EMC/ESD validation.

Chest-worn sensing is constrained by physics: coupling to skin/clothing, motion vectors, and parasitic paths that inject artifacts. This chapter compares what is feasible for a pendant/clip-on chest module and what breaks first, using field-auditable evidence (contact state and motion tier) instead of algorithm discussions.

ECG / biopotential (sternum coupling; dry-electrode reality)

• Why it fits: sternum/upper-chest placement can provide usable biopotential coupling if skin contact is stable.
• What breaks first: dry-electrode contact changes (impedance jumps) and intermittent coupling in pendant/clip-on wear.
• Must-have instrumentation: lead-off/contact detect and a signal-quality flag; without it, the system cannot separate “bad contact” from “bad physiology.”

Bioimpedance respiration (BioZ) — loop integrity & sanity checks

• Why it fits: respiration changes chest geometry/coupling; BioZ can track low-frequency impedance modulation under controlled excitation.
• What breaks first: injection/measurement loop instability from contact changes; output can “look plausible” while being wrong if loop health is not monitored.
• Safety constraints (high-level): excitation must be energy-limited; the system must degrade/stop when sanity checks fail.

PPG on chest (only if true skin contact is maintained)

• Feasibility gate: PPG is only appropriate when the design ensures stable skin contact + light blocking + near-constant pressure.
• What breaks first: ambient leakage (edge light through clothing gaps) and motion-induced pressure variation.
• Evidence mindset: correlate PPG instability with motion index and ambient changes; desk tests are not representative of on-body leakage.

IMU (posture primary + respiration proxy)

• Posture: IMU is the primary hardware path; success depends on mounting stability (pendant swing is a top failure source).
• Resp proxy: chest expansion/motion can be observed, but high motion tiers require confidence gating; treat it as a robustness layer, not a universal solution.
• What breaks first: orientation bias shifts from placement changes; drift with temperature/time; false flips during stillness due to swing.

Turn the HR analog front-end into a selection + debug checklist. In chest-worn pendants, the dominant failures are often contact instability, motion-driven baseline wander, or overload during bursts. The AFE must fail gracefully: expose quality flags, avoid silent “plausible but wrong” outputs, and support fast diagnosis with minimal captures.

Checklist 1 — input-referred noise vs usable chest amplitude

• Noise is only valuable inside a valid window: if motion/contact overload dominates, priority shifts to dynamic range and overload recovery.
• Usable amplitude is contact-dependent: define a minimum “contact quality” level at which HR is allowed to be reported.
• Auditability: pair every HR output with a signal-quality indicator; track “invalid time” explicitly.

Checklist 2 — CMR/RLD (high-level), protection, biasing, and lead-off

• Common-mode disturbance: body/environment coupling creates large common-mode swings; prioritize robust CMR behavior under real contact conditions.
• Input protection: include ESD/abuse handling appropriate to touch, charging proximity, and electrode exposure points.
• Biasing / return paths: dry electrodes need stable bias paths to prevent input wandering or latch-like behavior.
• Lead-off + contact-quality: distinguish true open/lead-off from “poor but connected” contact; both must map to explicit flags.

Checklist 3 — ADC choice: resolution, bandwidth, power

• Resolution: enough to preserve useful waveform detail when contact is valid; excessive resolution does not fix contact artifacts.
• Bandwidth: cover the targeted signal range while avoiding unnecessary wideband noise pickup.
• Power: define duty-cycling strategy (always-on vs burst sampling) in coordination with BLE/haptics events.

Checklist 4 — motion artifact behaviors (how the front-end should fail)

• Saturation/clipping: motion/contact events drive input beyond range → clipped waveform, slow recovery; must set HR invalid with a clear reason.
• Baseline wander: low-frequency drift correlated with motion tier → quality should degrade smoothly; avoid exposing HRV under poor conditions.
• Contact bounce: intermittent coupling → discontinuities + lead-off/quality toggling; log gaps explicitly and avoid “filling” without marking.

Evidence SOP — first 2 waveforms to capture

• Capture #1: AFE output (or ADC input) → classify: saturation vs baseline wander vs discontinuities.
• Capture #2: rail waveform (AFE rail or SoC rail during bursts) → correlate droop/UVLO flags with signal failure timing.

Robust respiration rate reporting is a systems problem: sensing coupling, motion mixing, power/timing events, and explicit trust gating. The output should be auditable as rate + confidence + dropout reason, rather than a single unqualified number.

Bioimpedance respiration vs IMU-derived respiration proxy — when each is stable

• BioZ stable when: contact is stable, loop health is valid, and motion tier is low enough that chest expansion remains observable.
• IMU proxy stable when: mounting is stable (clip-on more predictable than a swinging pendant) and motion tier stays within a controllable band.
• Practical rule: choose the path based on contact/loop health and motion tier; downgrade confidence instead of forcing rate output.

Front-end pitfalls (what breaks first)

• Injection interference (BioZ): excitation/measurement coupling can create synchronized spikes or apparent doubling, especially near timed system events.
• Electrode impedance changes: dry contact, sweat, clothing pressure shifts cause loop gain and baseline to change; rate becomes unstable without quality flags.
• Motion mixing: walking/swing motion energy overwhelms respiration micro-motion; rate becomes motion-driven unless gated.

Practical trust gating (conceptual)

• Motion gate: only trust respiration when motion tier is low (e.g., below an internal threshold) for a minimum stable time.
• Contact/loop gate: require stable contact (and BioZ loop health if used) before enabling BioZ-based rate reporting.
• Downgrade strategy: when gates fail, output low confidence or an explicit dropout reason instead of forcing a number.

Evidence logs (counters that make field issues diagnosable)

• motion_index / motion_tier (and stable-time counter)
• resp_confidence (quantized or scored) + resp_dropout_reason (motion_high / contact_bad / loop_fail / rail_event)
• dropout_count + seq_gap for missing-rate segments
• event_marker timestamps (BLE TX, haptics, charge-state change) for correlation

Keep posture vertical and auditable: sensor selection, mounting realities, and evidence for drift/misalignment. Posture quality is dominated by placement stability (pendant swing vs clip stability) and bias/temperature behavior, not deep math.

IMU specs that matter (field-visible outcomes)

• Bias stability: determines long-wear posture drift and “slow flipping” while still.
• Noise density: affects stillness detection and micro-movement sensitivity without false toggles.
• ODR: too low risks aliasing; too high increases power and can amplify noise-driven jitter.
• Power modes: low-power mode must preserve consistent bias/noise; mode transitions should be observable in logs.

Mechanical placement (rotation offsets & pendant swing)

• Rotation offsets: different clip positions and clothing layers change the module frame relative to the torso frame.
• Pendant swing: introduces a dominant motion component that can mimic posture transitions; posture should be gated during swing.
• Clip stability: more predictable than swing, but still requires offset calibration markers and periodic sanity checks.

Simple posture states with hysteresis (avoid false toggles)

• State set: upright / leaning / slouch (or similar) with explicit boundary definitions.
• Hysteresis: separate enter/exit thresholds + minimum stable time to prevent boundary oscillation.
• Stillness gate: posture transitions should require a low motion tier window; high motion should reduce confidence rather than switching states.

Evidence for drift and misalignment

• Calibration markers: log calibration events, estimated offsets, and validity windows.
• Temperature correlation: correlate posture bias or offset drift with temperature to reveal thermal effects or mechanical stress.
• Motion-class confusion (conceptual): “stillness misread as leaning” or “walking misread as slouch” indicates swing/placement coupling.

Make data integrity a core differentiator without any cloud talk. Reliability comes from an auditable chain: buffer → timestamp → sequence → replay → loss accounting. The goal is not “never lose data,” but to always know how much was lost, where, and why.

BLE roles: periodic sync vs continuous streaming

• Periodic sync: store locally first, then sync in short windows. Best for low-to-mid rate summaries, posture states, respiration/HR outputs, and event logs.
• Continuous streaming: send while sampling. It raises sensitivity to connection interval, retries, and power bursts; use only when a live view is required.
• Practical rule: define the data products per mode. Periodic sync should be the default path for “reliability first” behavior.

Connection interval tradeoffs (reliability-first framing)

• Long interval: lower average radio cost, but increases burstiness during sync and can push buffers toward high-water marks.
• Short interval: reduces per-sync latency but raises average current and can increase retry exposure under interference.
• Integrity coupling: interval, buffer depth, and power burst headroom must be consistent; if one changes, the other two need re-validation.

Local storage: circular buffer + event log + wear-aware writes (high-level)

• Circular buffer: keeps the most recent data window; supports delayed sync after disconnects. Track high-water mark to reveal near-overflow conditions.
• Event log: records state transitions and anomalies (disconnect reasons, resets, charging state changes, haptics events) to make integrity failures explainable.
• Wear-aware writes: keep high-frequency writes bounded; use tiered records (critical event vs routine stats) to avoid hidden write amplification.

Time: RTC timestamping + sequence counters + missing sample detection

• RTC timestamp: anchors “when it happened” across reconnects, reboots, and multi-session wear.
• Sequence counters: anchor “how many samples exist” independent of time drift; they enable deterministic gap counting.
• Gap detection: compare expected vs received sequence IDs during sync; log gaps with start/end IDs and gap length.

Evidence counters to log (make failures measurable)

• ble_retry_count (per session and lifetime) + disconnect_reason (enum)
• buffer_high_water_mark + optional buffer_overflow_count
• seq_gap_count + seq_gap_total (total missing samples)
• sync_flush_time_ms (distribution) + sync_fail_count
• reset_reason (watchdog / brownout / software) to correlate integrity failures with power events

Tie common symptoms back to rails and current spikes. Tiny wearables fail at the boundaries: burst overlap, UVLO thresholds, battery impedance rise, and handover events. The fastest path to truth is two measurements: battery droop + reset/PMIC flags.

Power domains: always-on vs burst rails

• Always-on (AO): RTC, wake logic, minimal retention. Leakage dominates long-wear performance if AO is oversized.
• Measurement burst: AFE/ADC (and BioZ injection if present). Sensitivity to rail ripple directly appears as signal artifacts.
• Radio burst: BLE TX/RX spikes. A common cause of VBAT droop and brownout-driven resets.
• Haptics burst: LRA/ERM drivers and buzzer events. Kickback and ripple can upset sensitive rails unless contained.

PMIC choices: buck vs LDO mix, load switches, UVLO thresholds

• Buck vs LDO mix: use buck for efficiency on higher-power rails; use LDO where noise isolation matters, while respecting dropout and thermal headroom.
• Load switch / rail gating: isolate burst domains so AO does not ride through unnecessary spikes; reduce burst overlap by sequencing rails.
• UVLO thresholds: too high triggers resets under cold/aged battery impedance; too low risks “running sick” and corrupting data without an obvious reboot.

Battery + fuel gauge: impedance rise, cold behavior, state-of-charge illusions

• Impedance rise: aged cells sag more under burst loads; reported SoC can look healthy while VBAT droops into UVLO during TX spikes.
• Cold behavior: low temperature increases internal resistance, turning “normal bursts” into brownout events.
• SoC illusions: fuel gauge learning depends on load history; mid-SoC sudden drops often mean “load step meets impedance,” not true capacity loss.

Evidence: first 2 measurements (fastest path to root cause)

• Measurement #1 — VBAT droop: capture VBAT minimum during known bursts (BLE TX, haptics, measurement window). Compare droop across temperature and SoC.
• Measurement #2 — reset cause / PMIC flags: read reset_reason from the SoC and PMIC status (UVLO, current limit, thermal) to identify the mechanism.

Vibration and buzzer events are noise sources. They corrupt sensing through two parallel paths: mechanical coupling (motion artifact into chest contact / IMU) and electrical coupling (kickback, rail ripple, ground bounce into AFE/ADC). The design goal is to make every feedback event predictable, isolated, and time-bounded so HR/resp data remain trustworthy.

LRA/ERM driver considerations (what breaks first)

• Kickback events: motor/actuator energy return at turn-off can create spikes on the driver rail and ground reference unless a controlled return path exists.
• Supply ripple injection: burst drive currents can modulate rails that feed AFE/ADC or disturb analog ground, showing up as baseline wander or clipping.
• Ground return mistakes: the haptics current loop sharing sensitive return paths is a common cause of “HR spikes during vibration.”
• Driver behaviors that help (high-level): current limiting/soft-start, controlled turn-off, fault/diagnostic flags, and predictable waveform control.

Containment: electrical isolation + mechanical isolation

• Electrical containment: keep the haptics current loop local; provide near-driver decoupling; use clamp/snub paths when turn-off transients create rail excursions.
• Rail isolation mindset: avoid letting haptics bursts share the same effective impedance path as sensitive AFE/ADC supplies.
• Mechanical containment: reduce direct vibration transmission to sensor contact points and the IMU mounting reference; prevent pendant swing from amplifying vibration artifacts.

Scheduling: blanking windows during haptic events

• Blanking windows: mark a short window during and around haptics events as “low confidence” for HR/resp extraction or sampling.
• Avoid burst overlap: do not align haptics bursts with BLE TX bursts or sensitive measurement windows when avoidable.
• Confidence-aware outputs: when feedback events occur, report a confidence downgrade rather than emitting a hard value based on contaminated segments.

Audible buzzer vs vibration: coupling tradeoffs

• Vibration (LRA/ERM): stronger mechanical coupling risk into chest contact + IMU reference; requires careful blanking and mounting control.
• Buzzer: reduced mechanical artifact but can introduce switching ripple/EMI-like coupling if the drive waveform injects noise onto shared rails.
• Selection lens: choose the feedback channel based on which modality is most sensitive for the product’s primary KPI (HR/resp vs posture stability).

Evidence: correlate feedback events with sensing failures

• Event markers: log haptics_event_id, start timestamp, duration, and amplitude/level.
• AFE integrity: count afe_saturation_count (clip) or equivalent “front-end overload” markers during and outside events.
• Respiration integrity: log a resp_confidence trend and count false peaks correlated with events.
• Power correlation (optional): record vbat_min_mv_during_haptics to distinguish mechanical artifact vs rail droop mechanisms.

Wireless charging is where thermal, power-path handover, and interruption behaviors hide. The engineering goal is stable charge sessions with measurable causes for every start/stop, while protecting skin comfort through predictable thermal derating.

Qi RX blocks (functional chain)

• Coil → RX rectifier/regulation → battery charger → battery + system load sharing.
• What to observe: RX input power behavior, charging state transitions, and whether the system load is sharing smoothly or causing handover stress.
• Reliability boundary: this chapter focuses on Qi session behavior and thermal constraints, not full rail budgeting (H2-8).

Alignment sensitivity: interruptions → complaints → data effects

• Misalignment reduces transferred power and increases variability, leading to repeated session restarts or slow-charge behavior.
• User-visible outcomes: slow charging, warm enclosure, frequent stop/start, and post-charge sync gaps.
• Data integrity coupling: interruptions change sync timing and can trigger buffer pressure or resets; log the cause so integrity issues remain explainable.

System load sharing and handover risk (without full power-tree details)

• Load sharing: the system may continue sampling and broadcasting while charging. Handover events occur when RX input power changes rapidly.
• Handover failure modes: transient VBAT droop, brownout-like resets, or silent data loss if write/sync operations are interrupted.
• Practical rule: avoid scheduling peak bursts during known unstable charge states and record a marker when charging state transitions occur.

Thermal derating: safety constraints and NTC placement

• Skin safety: derating is not optional; comfort and surface temperature constraints define safe charge power.
• Coil heating: misalignment increases losses and heat; stable alignment reduces both interruptions and peak temperature.
• NTC placement: place sensors where they represent what must be protected (coil hot spot and/or user-facing surface path). Incorrect placement can hide unsafe heating.

Evidence logs (make Qi failures diagnosable)

• charge_start_reason / charge_stop_reason (enum)
• coil_temp (and optional surface/skin proxy temp) + derating_level + derating_cause
• rx_input_power (bucketed or averaged over windows)
• recharge_time_distribution (session duration statistics; interruptions inflate tails)
• post-charge integrity: correlate stop events with disconnect_reason, seq gaps, and reset_reason (links to earlier chapters, without duplicating them)