Wearable ECG Patch/Band: ULP AFE, BLE, Power & Security

← Back to: Medical Imaging & Patient Monitoring

A wearable ECG patch/band is a skin-worn, ultra-low-power ECG capture system that combines a clean analog front end, BLE data transport, and reliable power/charging control. This page explains how to achieve usable ECG quality and predictable battery life by making the right architecture choices, enforcing dropout-safe data continuity, and closing the loop with device-side diagnostics, security, and production-ready validation.

H2-1 · What this page answers

One-sentence definition: A wearable ECG patch/band is a skin-coupled, ultra-low-power ECG acquisition chain (electrodes → ECG AFE → ADC) paired with a BLE data pipeline and an end-to-end security loop (identity, encryption, and signed updates).

Two engineering goals must hold together

Usable signal quality: stable R-peak detect, no frequent saturation, and clear quality flags under motion.
Acceptable wearability: realistic runtime within size/thermal limits, with predictable charging behavior.

What readers should take away

Architecture partitioning and interface boundaries.
Power budget logic (average vs peak vs duty-cycle).
Motion artifact containment (hardware + firmware layering).
Security closure: secure boot, keys, encrypted storage/link, signed OTA/DFU.
Validation & production test gates (fast PASS/FAIL criteria).

Coverage intent

what is an ECG patch · how the signal/data/power chain works · design considerations for signal quality, runtime, charging, and security

Why this matters: Wearable ECG reliability is defined by skin interface behavior, motion-driven saturation, radio duty-cycle, and a complete security loop (not by “higher resolution” alone).

H2-2 · Use cases & requirements that drive architecture

Use-case tiers (what changes electrically)

1) Continuous monitoring (Holter-like)

Long runtime emphasis → average current dominates.
Upload can be batch + buffered, not always real-time.
Sleep stability matters → baseline wander and electrode drift are frequent.

2) Event-triggered capture

Most of the time is deep sleep → wake cost becomes critical.
Needs pre/post ring buffer to preserve context.
False triggers waste battery and storage → quality gates required.

3) Motion-heavy (rehab / sports)

Motion artifact dominates → saturation recovery is a top spec.
IMU-assisted quality tagging improves interpretability.
Radio duty-cycle often rises (more events, more metadata).

Specs that actually matter (measurable)

Signal usability gates

Bandwidth & sampling: set by target waveform fidelity and detector robustness.
Input-referred noise: must be stated with bandwidth and gain conditions.
CMRR + mains rejection: wearable environments frequently carry 50/60 Hz stress.
Dynamic range + recovery: determines how often motion pushes the chain into clipping.

System delivery gates

Wear time: driven by average current, but limited by peaks (radio, flash writes).
Wireless robustness: buffering + retransmission policy + quality flags.
Charging predictability: thermal throttling, termination criteria, and user behavior tolerance.

Design knobs (what architecture can change)

Duty-cycling

sleep states · event windows · radio intervals

Analog margins

PGA range · input bias paths · recovery behavior

Lead-off strategy

DC/AC detect · thresholds · false alarm control

Buffering

ring buffer · burst uploads · flash wear policy

Timebase

local ticks · timestamps · resync rules

Security closure

identity · encryption · signed OTA/DFU

Practical rule

When use case changes, do not only adjust software filters—re-check dynamic range + recovery, radio duty-cycle, and buffering policy first.

Reading tip: start from the use-case tier, then lock down DR/recovery and runtime, and only then finalize filters and feature extraction. This prevents “good lab ECG, bad wearable ECG” outcomes.

H2-3 · Wearable ECG signal chain (electrodes → clean analog → trustworthy samples)

A wearable ECG front end must survive three realities at the same time: skin-electrode impedance drift, motion-driven offset jumps, and mains/common-mode stress. A “clean chain” is not only low noise—it is hard to saturate, quick to recover, and always outputs quality flags so downstream logic can trust (or gate) the waveform.

1) Electrode interface (where most problems start)

Input impedance & bias: high effective input impedance reduces “contact drift → waveform drift,” but it must still provide a controlled bias return path so the input does not float into saturation.
Protection is not free: ESD/RFI parts add capacitance and leakage; both can worsen mains pickup and low-frequency wander.
Recovery matters: after a contact slap or offset step, the front end should return to normal operation quickly (avoid long “dead time”).
Defib residue note (brief): wearable designs should tolerate fast transients at the input and recover, but energy-delivery details are out of scope here.

2) AFE chain blocks (organized by control points)

INA/PGA: sets noise floor and motion robustness. Gain strategy must protect against clipping when motion artifact spikes appear.
Baseline control (HPF / servo): removes slow drift but must not create long recovery tails that mask R peaks.
Anti-alias + LPF: defined together with sampling strategy to avoid “filter by accident.”
ADC (ΣΔ common, not mandatory): pick based on low-frequency noise, input range, power, and latency—not just resolution.

3) Why “good lab noise” becomes “bad wearable ECG”

CMR→DM conversion: skin-electrode impedance imbalance converts common-mode mains stress into differential ripple.
Offset steps: motion and sweat change contact impedance, causing sudden input offset jumps that push the chain toward saturation.
Recovery distortion: once clipped, the return path and servo time constants decide how long the waveform is “untrustworthy.”

Practical design stance

Reserve dynamic range + fast recovery first, then optimize noise. A chain that clips often will lose more usable ECG than a slightly higher noise floor.

Lead-off & contact quality (turn “wearability” into measurable states)

Lead-off detection

Injection + sense: apply a tiny stimulus (AC or DC) and observe response to infer electrode continuity.
Engineering goal: detect open contact without injecting visible artifacts into the ECG band of interest.
False alarm control: combine threshold + time qualification so motion does not look like lead-off.

Contact-quality estimation

Impedance trend: track relative impedance changes (not absolute “perfect numbers”).
Saturation markers: count clipping or long recovery windows as “low quality.”
Output: generate quality flags (OK / marginal / bad) so buffering, upload, and event logic can gate actions.

Engineering checklist (quick self-review)

Does the bias return path prevent “floating input → long saturation” under sweat/motion?
Do protection parts introduce leakage/capacitance that noticeably worsens mains pickup or baseline wander?
Is there a defined saturation recovery behavior (not just “filters”)?
Are lead-off and contact-quality separated into distinct states with debounced thresholds?
Do downstream stages receive quality flags to gate buffering, upload, and event triggers?

H2-4 · ULP power budget & duty-cycling (runtime is strategy, not a bigger battery)

Wearable runtime is won by budgeting and duty-cycling. ECG sampling can be continuous, but the radio does not have to be. The most reliable approach is: sample steadily → buffer locally → transmit in short bursts, while tracking peak events that can cause brownouts.

1) Build a power ledger (so numbers can be audited)

P_total(avg) = P_AFE + P_MCU(avg) + P_BLE(avg) + P_Sensors(avg) + P_Memory(avg) + P_Leakage

Watch peak events too:
I_peak ≈ I_BLE_TX(peak) + I_Flash_write(peak) + I_CPU_burst(peak)

Average current sets runtime; peaks decide resets, dropouts, and “mystery” packet loss.
If measured runtime is far below the estimate, investigate leakage and unexpected wakeups before blaming the battery.

2) Duty-cycle core idea (continuous sample, non-continuous radio)

Sample: keep AFE stable and avoid frequent mode switches that create artifacts.
Buffer: use a ring buffer; write to flash in planned bursts to control write peaks and wear.
Burst TX: transmit in short windows; tolerate short link outages without losing the waveform context.

Key runtime lever

Reduce radio duty-cycle first (intervals, bursts, retries), then optimize CPU cycles. RF time usually dominates the “surprise drain.”

3) Knobs that matter + wearable reality traps

Radio & pipeline knobs

Advertising/connection interval, payload sizing, TX power.
Adaptive retransmission (avoid “retry storms” when link is poor).
Event-triggered uploads (send waveforms only when quality gates allow).

Leakage & humidity traps (common in bands/patches)

Moisture paths: sweat and condensation can create parasitic conduction near electrode/protection parts.
“Quiet drain” sources: charger/fuel-gauge quiescent current and sensor standby can dominate deep-sleep budgets.
Self-discharge & protection leakage: small cells magnify these effects.

Debug playbook (when runtime is worse than expected)

Measure deep-sleep current in dry vs humid conditions to expose moisture/leakage paths.
Log wake reasons; frequent short wakeups can dwarf steady sampling budgets.
Capture TX and flash-write peaks; confirm the supply does not dip during burst windows.
Cap retries when the link is bad; “retry storms” silently drain batteries.

H2-5 · Motion artifact & baseline control (hardware + sampling strategy, no heavy math)

In wearable ECG, “noise” is often a mix of contact-driven impedance changes, common-mode stress, and temporary saturation. Robust designs treat artifact handling as a closed loop: detect states, raise quality flags, and apply bounded mitigation policies.

1) Artifact sources (physical → electrical)

Electrode micro-motion: step-like offsets and spikes; can push the chain into clipping.
Skin stretch: slow impedance drift; shows up as baseline wander and low-frequency wobble.
Sweat / humidity: leakage paths and impedance imbalance; increases CM→DM conversion.
Common-mode disturbance: mains ripple becomes visible when impedance mismatch exists.

2) Hardware-side controls (what can be tuned)

HPF / baseline servo: remove slow drift without creating a long recovery tail that masks QRS.
Adjustable PGA: protect against motion spikes while keeping quiet signals usable at rest.
Saturation recovery: define a fast, predictable settle-back behavior after clipping events.
Protection pitfalls: avoid protection networks that add large bias/leakage and worsen drift.

Practical stance

Prioritize not clipping and fast recovery. A slightly higher noise floor is often less harmful than frequent saturation windows.

3) System-side hooks (interfaces & criteria only)

IMU/ACC assist

Expose a simple motion level (low / mid / high) and impact markers.
Use it as a policy input (gating and mode choice), not as a “magic waveform fix.”

Quality flags → mitigation policies

quality=bad: gate event triggers; buffer with tags; avoid aggressive retransmit loops.
saturation=1: protect decisions during recovery; optionally lower gain temporarily.
contact=marginal: mark data as low confidence; reduce “false event” uploads.

Symptoms that look like noise but are often saturation recovery

Symptom	Typical cause	What to check
Flat-top / rail-hugging segments	Input or PGA clipping	saturation flag, gain state, motion level
Slow return “ramp” after a big step	Baseline servo / HPF recovery tail	recovery-in-progress flag, time constant mode
R peaks disappear while “noise” looks small	Gated/biased during recovery window	quality flags around the missing peaks
Random low-frequency wobble in motion	Impedance drift + CM→DM conversion	contact-quality trend, humidity/leakage clues

Quick validation checklist

Do saturation events always raise a visible flag (no silent clipping)?
Is the recovery window bounded and gated from event triggers?
Do baseline-control settings avoid long tails that hide QRS during motion?
Do policies reduce unnecessary retransmits during low-quality periods?

H2-6 · BLE SoC & data transport (device-side only: stability + time consistency)

Reliable ECG streaming over BLE is less about “faster radio” and more about reconstructable packets, bounded retries, and device-side buffering. The device should survive loss, reorder, and disconnect without breaking time continuity.

1) Data organization (minimum fields for robust rebuild)

seq: packet sequence number to detect loss and reorder.
sample_counter: a monotonic counter to preserve waveform continuity even across reconnect gaps.
timestamp (optional): helps align segments and merge buffered uploads.
flags: quality/contact/saturation markers carried with the data (enables gating policies).

Why sample_counter matters

A counter reveals missing samples directly; a timestamp alone can drift or jump after reconnect.

2) Connection strategy (device state machine)

Normal: burst upload on schedule while sampling stays steady.
Congestion: reduce upload frequency, cap retries, and avoid “retry storms.”
Disconnect: enter buffer-first mode; tag segments with counters for later reconstruction.
Reconnect: transmit buffered segments in bounded windows (prioritize event-tagged chunks if used).

3) Device buffering (RAM vs Flash) + lightweight power-loss protection

RAM ring buffer

Lowest energy per operation; ideal for short disconnects and steady streaming.
Pairs naturally with burst TX and bounded retry policies.

Flash log (optional)

Used when long disconnects or “event preservation” is required.
Write in planned bursts; protect the last segment with a simple commit marker to avoid partial records.

OTA/DFU vs transport (avoid “half-brick”)

Atomicity (concept level)

Keep an install flow that can roll back if validation fails (no deep boot-chain details here).

Resume + integrity

Transfer in chunks with checks; resume after disconnect instead of restarting and draining battery.

Gates

Do not start updates when battery is low or link quality is poor; enforce a bounded update window.

Device-side acceptance checklist

Packets carry seq and sample_counter so gaps are detectable and rebuildable.
Retries are capped; congestion triggers lower upload frequency instead of endless retransmits.
Disconnects switch to buffer-first mode; reconnect drains buffered data in bounded windows.
Flash logging (if used) writes in bursts and uses a simple commit marker to avoid partial records.
OTA/DFU starts only when battery and link quality gates pass; chunked transfer can resume safely.

H2-7 · Power/charging management (rechargeable vs primary: architecture boundary)

Battery choice is an architecture choice. Rechargeable thin LiPo pushes designs toward power-path control, thermal derating, and robust low-battery gates. Primary coin-cell designs push toward ultra-low always-on and carefully managed burst domains to avoid brownouts.

1) Battery type sets the boundary conditions

Thin LiPo (rechargeable)

Needs a safe charging path and thermal limits near skin.
“Charge + run” behavior must not corrupt data continuity.
Protection + fuel gating must be predictable under load spikes.

Coin cell (primary)

Weak pulse capability; TX + Flash peaks can cause resets.
Requires an always-on rail with very low quiescent draw.
Data upload must be burst-managed and gate-controlled.

2) Device-side power-path (wired / wireless as sources)

Charge-only: simplest, but “plug-in events” can glitch rails and break continuity.
Load-sharing power-path: preferred for continuous monitoring; input powers the load first, excess charges the battery.
Wireless input source: treat as a fluctuating source; prioritize input stability and heat, not protocol details.

Wearable-specific pitfall

Charging transitions can introduce ripple and ground shifts that look like baseline wander. A practical approach is to tag “charging state” and gate event triggers during transients.

3) Power rails and domains (keep peaks controlled)

Always-on domain

RTC/timer, minimal RAM retention, gate logic, reset-cause capture.

Sample domain

ECG AFE + sampling timebase; keep stable to avoid self-made artifacts.

Burst domain

BLE TX, CPU bursts, Flash writes; schedule and gate peaks to prevent brownouts.

4) Fuel gauge and low-battery gates (make brownout a controlled event)

SoC estimation: trend-based estimation is low-cost but drifts; coulomb-style tracking can be tighter but needs calibration discipline.
Temperature awareness: available power drops at low temperature; gates should consider temp, not only voltage.
UVLO policy: when energy is marginal, allow sampling+buffering but block Flash writes and burst TX that can collapse the rail.
Continuity protection: keep sample_counter and reset-cause logs consistent across low-voltage conditions.

5) Thermal safety near skin (derating strategy)

Measure: NTC near the battery/charger hot spot, not only the PCB average.
Derate: step down charge current and burst activity as temperature rises; avoid oscillation by adding hysteresis.
Behavior: tag “thermal-derate state” so data quality and event triggers are interpreted correctly.

Practical checkpoint

A safe system keeps monitoring predictable during charging and never lets a transient “plug-in” event corrupt buffers or counters.

Acceptance checklist (power)

Plug-in or charging transitions do not reset counters or corrupt buffered segments.
BLE TX and Flash writes are scheduled or gated to avoid peak stacking.
UVLO policy blocks burst domains first, keeping sampling and timebase stable.
Thermal derating is monotonic and does not chatter around thresholds.

H2-8 · Data security & privacy on-device (minimum closed loop on the device)

A practical wearable security baseline prevents easy cloning and firmware tampering, keeps the BLE link from falling back to plaintext, and ensures any local retention is encrypted and erasable. The goal is a small, continuous chain: boot trust → update trust → key trust → link trust.

1) Device-side threat model (what is realistic here)

Cloning: copying identity and pretending to be a genuine device.
Firmware tampering: replacing firmware to change behavior or leak data.
Sniffing/replay: capturing BLE traffic to read or replay segments.
Key exposure: extracting long-term keys from non-secure storage.

2) Minimum viable chain (continuous, not optional)

Verified boot: refuse to run untrusted firmware images.
Signed DFU: update images must verify before install; avoid “install first, regret later.”
Key storage: keep long-term identity keys inside an SE or SoC secure area.
Link encryption: enforce encrypted BLE sessions and prevent downgrade to plaintext.

Practical rule

If the chain can be broken at any step (boot/update/key/link), everything after it becomes cosmetic.

3) Provisioning basics (strategy, not factory IT)

Per-device identity: each unit gets unique identity material (keypair/cert).
Injection posture: avoid leaving long-term secrets readable in general Flash.
Lockdown: disable or restrict debug readout after provisioning.
Rollback barrier: keep a monotonic version barrier concept to prevent easy downgrade attacks.

4) Local retention (encryption + erase policy)

Encrypt at rest

Protect buffered ECG segments and event logs if stored beyond short RAM windows.
Prefer session-derived storage keys over direct use of long-term identity secrets.

Erase posture

Support an explicit “clear sensitive data” action (unlink/return/service).
Erase should be engineered (invalidate keys / wipe blocks), not just “delete pointers.”

Acceptance checklist (on-device security)

Unsigned firmware never runs; verified boot fails closed.
DFU verifies before activation; interruption does not create a half-installed state.
Long-term identity secrets are not stored as readable plaintext in general Flash.
Encrypted BLE sessions are enforced; downgrade to plaintext is blocked by policy.
Any retained local data is encrypted and has an explicit erase action for unlink/return/service flows.

H2-9 · Reliability, skin interface & real-world pitfalls (hidden costs in wearables)

Wearable reliability improves when every “random” field issue becomes a measurable signal: contact quality, time continuity, buffer health, and reset reasons. The goal is not perfect conditions, but predictable behavior under sweat, motion, and dropouts.

1) Skin contact drift (observe, classify, act)

Moisture/sweat: impedance may drop while common-mode disturbance rises; baseline can drift faster.
Adhesive aging: edge lift causes micro-motion artifacts that cluster during movement.
Skin irritation: frequent repositioning creates near-off states that look “connected but unusable.”

Minimum observables

contact_quality_flag · impedance_trend · saturation_count · baseline_drift_index

2) Contact quality monitoring (avoid “half-contact” traps)

Lead-off vs near-off: full open is easy; “almost open” is more damaging because it produces misleading waveforms.
Stable decision: use hysteresis and time windows to prevent chattering around thresholds.
Output interface: expose quality grade (0–3), not a raw number; pair it with counters for diagnostics.

Useful flags

lead_off · near_off · quality_grade · quality_stable_ms

3) Dropouts and “leave-phone” behavior (device-side)

When disconnected: prioritize sampling + buffering; avoid wasting energy on repeated scans and failed sends.
When reconnected: use a gap-fill policy that preserves time order (sample_counter/sequence) before “fresh” data.
When congested: rate-limit retransmits; cap backfill size to protect battery and avoid long “catch-up” loops.

4) Reset/hang observability (WDT + evidence, not magic)

Reset cause: WDT vs brownout vs software reset must be recorded every time.
Last state: capture whether the system was in Sample / Flash write / BLE TX when it failed.
Continuity: detect and mark sample_counter gaps so downstream reconstruction remains deterministic.
Crash breadcrumbs: compact error code, memory watermark, and last N-second counters (optional but valuable).

Minimum logs

reset_cause · last_state · last_error_code · sample_gap_detected

5) Symptom triage (fast checks that reduce guesswork)

Sudden noise ↑: check quality_grade and impedance_trend; then check saturation_count and charging/derate tags.
Clipped waveform: check rail-hit counters and saturation recovery; confirm gain state did not stick high.
Gaps without reboot: check dropout logs, buffer overflow flags, and sample_counter continuity markers.
Frequent reconnect: check reconnect histogram and backfill cap; confirm policy prevents battery drain spiral.

Desired outcome

Each field issue maps to a small set of root causes with logged proof and a predictable fallback mode.

Acceptance checklist (reliability)

Quality and continuity are observable: quality_grade, gaps, overflow, and reset causes are logged.
Dropout policy prefers buffering and deterministic backfill over repeated failed realtime sends.
Peak stacking is prevented by scheduling or gating burst TX and Flash writes.
Triage starts from logged evidence, not waveform guesswork.

H2-10 · Validation & production test (what to measure, how to test, and pass criteria)

Validation proves the design meets performance and robustness targets. Production test compresses those proofs into fast, repeatable, and traceable steps: stimulus → measurement → pass/fail → record.

1) AFE validation (repeatable conditions, not anecdotes)

Noise: define bandwidth and input condition (shorted / simulator); record RMS and spectral indicators.
CMRR: verify mains-band rejection under realistic imbalance; record margin and stability across gain.
Bandwidth: confirm HPF/LPF corners do not distort baseline behavior; record corner tolerance.
Saturation recovery: force rail-hit and measure recovery time back to usable range.
Lead-off: verify threshold + hysteresis prevents chatter in near-off conditions.

Pass/fail should include

measurement condition · limit · margin · repeatability across units

2) Power profiling (state-split: average + peak)

Measure by states: Sleep → Sample → Buffer → Burst TX → Flash write.
Record average current and peak current per state, plus dwell time distribution.
Validate “no peak stacking”: TX peak and Flash peak do not overlap under normal policy.
Validate consistency: runtime variance across units stays within expected policy-driven ranges.

3) Wireless robustness (throughput, loss, reconnect, backfill)

Throughput: test in distance/obstruction/typical interference; record usable payload rate.
Loss + reorder: collect packet loss and reorder stats; verify reconstruction from sequence/sample_counter.
Reconnect: measure reconnect time histogram; verify backfill cap prevents battery-drain spirals.
Continuity: gap-fill success rate and maximum tolerated gap before downshift policy triggers.

4) Production test (fast, repeatable, traceable)

Electrode simulator fixture: switchable impedance + open + near-off modes for quick screening.
Calibration: program constants, read-back verify, and store calibration version.
Device identity: confirm unique ID and record key fingerprints (no secrets exposed).
RF/BLE test: short throughput + reconnect test with pass/fail thresholds.

5) Evidence package (what logs/records should exist)

AFE performance records: noise, CMRR, bandwidth, saturation recovery, lead-off behavior.
Power state records: per-state averages, peaks, dwell times, and policy configuration.
Wireless records: throughput/loss/reconnect/backfill statistics.
Production trace: calibration version, unique ID, test pass/fail summary, and firmware build ID.
Device event logs: reset_cause, WDT, brownout markers, buffer overflow markers, DFU success/fail counters.

Acceptance checklist (test)

AFE tests specify conditions and limits, and can be repeated across units without ambiguity.
Power tests are state-split and include peak capture to prevent brownout surprises.
Wireless tests validate reconstruction and backfill behavior, not only “connected once.”
Production flow writes calibration, verifies it, and stores trace fields for later audits.

H2-11 · BOM / IC selection checklist (criteria-driven, not “model stacking”)

A wearable ECG patch/band BOM should be selected by pass/fail criteria that protect signal quality, battery life, and reliability under sweat, motion, and dropouts. The shortlist below is organized by layers (AFE / BLE SoC / PMIC+Charger / Fuel Gauge / Memory / Sensors / Protection), and highlights system coupling traps that often dominate real-world performance.

How to use this checklist

Start with architecture: single-lead vs multi-lead, continuous vs event-first, rechargeable vs disposable.
For each layer: lock 5–8 criteria + a realistic “threshold class” (not a single number).
Run a coupling review: peaks, thermals, and dropouts can negate the best individual IC.
End with production reality: calibration, traceability, and test fixtures must be supported.

Recommended “threshold class” writing

Use classes: “single-digit µA sleep class”, “fast recovery class”, “low-leakage protection class”.
Define conditions: bandwidth, gain state, and typical policy (buffer + burst TX) must be stated.
Require evidence: per-unit repeatability and margin, not only a datasheet headline.
Keep scope local: device-side only; no cloud/gateway assumptions are needed.

Layered checklist (what to lock first)

ECG AFE: noise + recovery + contact diagnostics define “usable signal”.
BLE SoC: sleep class + deterministic DFU + time continuity define “usable data”.
PMIC/Charger: peak handling + thermal derate interface define “usable runtime”.
Fuel gauge: reliable low-battery gating prevents silent data breaks.
Memory: buffering + write peaks must match dropout/backfill policy.
Sensors: low-power motion context supports artifact handling without draining battery.
Protection: ESD/transients must not inject leakage or bias into the ECG front end.

1) ECG AFE — key criteria (5–8)

Noise class + test condition: define ECG bandwidth and gain state; require repeatable RMS noise evidence across units.
Input bias & impedance robustness: maintain baseline stability when electrode impedance changes with sweat and lift.
CMRR / mains resilience: verify stability under realistic imbalance; avoid “good only in ideal fixtures”.
Saturation recovery class: fast return to usable range after rail-hit; count and log recovery events.
Lead-off / near-off diagnostics: require stable hysteresis + time window; export quality_grade and counters.
Integration flexibility: channels, sample alignment, and interface options that keep time continuity simple.
Power modes: predictable wake/sleep transitions for duty-cycled architectures.