Practical boundary (what this page covers, and what it does not)

A fitness band is an ultra-compact, wrist-worn tracker optimized for very low average current and high robustness under real wear. The hardware scope here is the end-to-end chain: PPG (HR/SpO₂) + accelerometer/IMU → low-power BLE SoC (sampling schedule, epoch alignment, packetization) → power/charging/protection that prevents brownouts during burst loads.

Out of scope: smartphone app/cloud architecture, GNSS, ECG, Wi-Fi/LTE, medical compliance procedures, and BLE spec-level deep dives (ATT/GATT). Every claim should map to a hardware mechanism and measurable evidence.

Typical feature set (described as “function → hardware meaning”)

• HR / SpO₂: LED pulses + photodiode + AFE (TIA / ambient cancellation / ADC) → usable raw waveform and confidence window.
• Steps / sleep trends: IMU event-driven operation (ODR / FIFO / interrupts) → activity epochs without keeping the main core awake.
• Sync / upload: BLE advertising/connection events create burst current → average power is controlled by duty-cycle and connection parameters.
• Battery life & stability: low-load efficiency (µA–mA region) + burst headroom → determines brownout resets and “random dropouts”.

Hard constraints (why fitness bands “drift” more easily in the field)

Evidence-first (two measurable signals that explain most “it feels wrong” complaints)

• Current waveform evidence: baseline + bursts (BLE events and LED pulses). Explains battery life, dropouts, resets, and charging heat.
• PPG raw waveform evidence: clipping, baseline drift, strong-light ripple, motion artifact shape. Locates the physical cause (optics/AFE/wear).

Organize by IC roles, not product features

A fitness band succeeds when two loops are stable: the signal loop (optics + motion → aligned samples) and the power loop (battery → rails that survive burst current). The block map below is the anchor: every later section should refer back to a block or a link.

• What sets accuracy: optical coupling + AFE timing/range + motion-aligned epochs.
• What sets battery life & stability: radio burst duty-cycle + low-load efficiency + brownout resilience.

Role-by-role interpretation (what each block is responsible for)

Six drivers that most field issues map to (kept intentionally hardware-level)

Engineering takeaway (the “ceiling” argument)

A fitness band PPG signal is a small modulation riding on top of much larger DC/ambient components. Optical coupling and shielding determine whether that modulation exists with enough SNR to survive motion and sunlight. Algorithms can denoise and reject artifacts, but they cannot recover information that was never captured due to light leakage, crosstalk, or saturation.

Scope here is strictly hardware: wavelength choices, LED/PD geometry, window/adhesive stack-up, shielding walls, and the observable signatures of loose wear and leakage. No clinical accuracy claims or regulatory procedures.

Wavelength selection (green vs red/IR) as an optics + power trade

• Green (HR): strong reflectance modulation in typical wear, but highly sensitive to pressure/coupling changes (loose strap → amplitude swing).
• Red/IR (SpO₂): dual-wavelength ratio logic increases sensitivity to crosstalk/leakage and window material behavior; shielding and stack-up consistency become more critical.
• Practical boundary: wavelength is not “more accurate” by itself; it is “more controllable” under a given window material, shielding budget, and LED pulse energy.

Geometry that matters most: LED/PD placement, shielding walls, and dominant stray paths

The optical system must maximize the intended path while minimizing two failure paths: crosstalk (LED light reaches PD without going through tissue) and ambient leakage (external light reaches PD through gaps or window edges).

• Intended path: LED → tissue scattering → PD (the only path carrying physiological modulation).
• Crosstalk path: LED → internal reflections / window edge / adhesive meniscus → PD (creates “fake signal” and destroys SpO₂ ratio stability).
• Ambient leakage: sunlight / indoor flicker / display PWM → PD via gaps (consumes dynamic range, adds ripple, triggers saturation downstream).

Window & stack-up: small mechanical details that become electrical symptoms

• Window material: wavelength-dependent transmission and surface reflections affect red/IR more strongly; haze/texture increases stray scattering.
• Adhesive layer: thickness and refractive index mismatch can create internal reflection paths; edge fillets can “pipe” light into the PD area.
• Light barrier wall: height, continuity, and fit determine whether crosstalk is blocked or merely redirected.
• Gaps (gasket / vent / tolerance): the fastest route for ambient light leakage; often correlates with “works indoors, fails outdoors.”

Evidence checklist: what “loose wear” and “leakage” look like in raw data

• Loose wear: large amplitude changes with strap pressure, stronger baseline wander, and more frequent motion-synchronous spikes.
• Ambient leakage: residual ripple aligned to 50/60 Hz lighting or PWM bands, and elevated DC level even when LED drive is reduced.
• Crosstalk: PD output remains high when tissue coupling is poor; LED current scaling looks “too linear” with reduced pulsatile content.
• Outdoor strong light: DC climbs quickly, leaving little headroom; downstream shows clipping or recovery lag (ties directly into H2-4).
• Batch-to-batch drift: similar electronics but shifted baseline/amplitude distributions—often points to window/adhesive/shielding consistency.

What the AFE must guarantee (in one sentence)

The AFE must convert photodiode current into non-clipped, low-noise, time-aligned samples while ambient light and motion create large DC shifts and ripple. When optics cannot fully block stray light, the AFE is the last line of defense against saturation and residual flicker.

TIA selection: noise, bandwidth, and “real stability” with PD capacitance

• Input-referred noise: sets the minimum visible pulsatile component after subtraction; too high → HR waveform becomes “grainy.”
• Bandwidth / settling: must settle within the LED pulse sampling window; insufficient settling → amplitude error and phase distortion.
• Stability with capacitance: PD capacitance + routing parasitics alter phase margin; layout and feedback compensation decide whether ringing appears.

A common field failure pattern is not “TIA spec too small,” but “effective bandwidth and stability degraded by parasitics,” especially in thin wearables.

LED pulse + sampling schedule: reclaim SNR in the time domain

• Ambient slot (LED OFF): measure background light baseline.
• LED ON settle: allow TIA/ADC to settle after switching LED current.
• Sample window: capture the intended signal + remaining ambient.
• Reset/discharge: prevent memory effects and speed recovery after near-saturation events.

Clean subtraction requires consistent timing; if sleep/wake jitter moves the sampling window, residual ambient ripple rises and shows up as periodic artifacts.

Ambient cancellation & anti-saturation: keep headroom before it is too late

• Headroom first: reduce optical leakage (H2-3) to stop DC from consuming dynamic range.
• Timed subtraction: remove ambient baseline using OFF/ON slots; watch for residual flicker at lighting/PWM frequencies.
• Controlled gain: manage TIA gain and LED pulse energy so the ADC does not clip under bright conditions.
• Fast recovery: after a clip event, design reset/bleed paths so the baseline returns quickly (avoids long “dead time”).

ADC, ENOB, and dynamic range: why “more bits” can still fail

• ENOB vs nominal bits: quantization is rarely the limit; noise floor and headroom dominate.
• Dynamic range consumption: ambient DC and leakage reduce effective resolution for the pulsatile component.
• Multi-wavelength switching: channel switching and recovery time can create apparent baseline drift if settle/reset is insufficient.

Evidence signatures (what to look for in traces)

• Saturation: flat-topped waveform (clipping), non-linear scaling vs LED current, and slow baseline recovery.
• Ambient ripple: strong components at 50/60 Hz or PWM bands; subtraction leaves residual periodic “wobble.”
• Thermal drift: baseline shifts with temperature even when motion is minimal; often linked to bias/reference/TIA offset behavior.
• Timing jitter: increased residual noise correlated with sleep/wake transitions; sampling window effectively moves within the pulse.

Why the accelerometer is not “optional” in a fitness band

Motion artifacts are primarily caused by optical coupling changes (micro-slips, pressure variation, light leakage shifts) that distort the raw PPG waveform. An accelerometer (IMU) is the hardware channel that makes motion observable and therefore controllable: it enables event-driven wakeups, epoch-aligned quality gating, and predictable degradation under high motion rather than unstable readings.

Scope is engineering implementation and measurable evidence: ODR/range selection, FIFO + interrupts, epoch alignment with PPG sampling, and motion-tiered confidence. No algorithm-paper deep dives.

ODR and full-scale range: selection logic that preserves correlation

• ODR (output data rate): choose the lowest rate that still captures the dominant motion energy for the target activities; excessive ODR increases data movement and wake costs.
• Range (full-scale): ensure headroom for worst-case peaks (running / hand shaking). If the IMU clips, RMS no longer tracks artifact intensity, breaking correlation-based gating.
• Practical target: stable, repeatable correlation between accel RMS and PPG residual distortion is the pass/fail signal for this tuning.

FIFO + interrupt wake: turning “continuous motion” into “batch evidence”

• FIFO watermark: accumulate samples in the IMU with minimal MCU activity.
• Interrupt wake: wake the SoC only when enough motion evidence exists to classify the epoch.
• Burst read: read a block of IMU data once per epoch instead of frequent short transactions.

This approach reduces wakeups while still producing a high-quality motion descriptor for gating PPG quality.

Epoch alignment: the minimum “same-time” loop that makes gating reliable

The IMU becomes valuable when its motion evidence is aligned to the same time window as PPG sampling. Define fixed epochs (e.g., 0.5–2 s windows) using an RTC/low-power timer, then bind: PPG samples and IMU RMS (and/or peak metrics) to that same epoch.

• Aligned evidence: one epoch produces one quality decision, avoiding “good IMU data but wrong time.”
• Stable behavior: motion-driven confidence changes become consistent (no random oscillation of HR/SpO₂).

Motion-tiered confidence: make degradation predictable instead of noisy

• Low motion: normal HR update; SpO₂ confidence normal.
• Medium motion: SpO₂ confidence reduced or update rate slowed; HR updates with stricter quality gating.
• High motion: SpO₂ update held or shown as trend-only; HR may continue but flagged low quality.

Evidence: what to measure to prove the mechanism

• Correlation: motion artifact level vs accel RMS (correlation should strengthen once ODR/range/epoch alignment are correct).
• Activity signatures: running tends to create more periodic distortion, cycling often shows steadier low-frequency energy, and hand shaking produces impulsive spikes that break waveform continuity.
• Regression signal: if IMU clips or epochs drift, correlation drops and gating fails—this is a direct, testable indicator.

The core model: average current = baseline + spike duty

Battery life in a fitness band is dominated by average current, which is the sum of a deep-sleep baseline and short, periodic spikes: BLE advertising/connection events, PPG LED pulses, sensor reads, and CPU wake windows. Optimizing lifetime means reducing spike count, spike width, and preventing rail droop resets during spikes.

Sleep states and domains: deep sleep, RTC domain, and retention cost

• Deep sleep: high-frequency clocks and most peripherals off; wake sources limited and controlled.
• RTC domain: keeps epoch timing and wake schedules alive at minimal power.
• RAM retention: preserves critical state (sequence counters, buffers) at a leakage cost; trade retention vs wake re-initialization time.

Advertising and connection intervals: spike density control (without spec-level detail)

• Advertising interval sets how often radio wake spikes occur when not connected.
• Connection interval and slave latency set event frequency and allow skipping events to cut average current.
• Sync feels “slow” when intervals are too conservative for the data volume or when batching is not used.

Sensor hub + timers: batch work to reduce wake frequency

• Low-power timers define epochs and schedule sampling windows.
• Batch reads from IMU FIFO and PPG buffers reduce bus transactions and CPU wake time.
• Wake for meaning: wake on epoch completion or FIFO watermark, not on every sample.

The minimum “no-loss” loop: sequence, log, buffer, confirm

“No data loss” is achieved with a small, hardware-friendly loop: generate epoch records with a sequence number and integrity check, store locally first, then transmit; mark confirmed records after successful delivery. This prevents duplicates, gaps, and reordering after disconnects.

• Epoch record: epoch_id + sequence + quality flags + CRC.
• Ring buffer: retention RAM / FRAM / flash depending on budget and write endurance.
• Resume after drop: reconnect continues from last confirmed sequence instead of restarting blindly.

Dropout evidence: separate RF issues from power droop

• Current spikes: advertising/connection events appear as repeatable peaks; power droop failures cluster around these peaks.
• RSSI distribution: if RSSI is stable but drops still happen at spike timing, prioritize rail headroom and recovery logging.
• Power-driven signature: brownout reset counters, gap in sequence, and “restart at spike” patterns point to transient power integrity.

Battery life is a ledger, not a guess

Real battery life is determined by average current, which comes from multiple domains (sensor, AFE/LED, radio, MCU) and their duty cycles, plus conversion losses and event spikes. A practical design treats lifetime as a ledger: each domain has an “active cost,” a “time share,” and a conversion efficiency penalty.

Scope: cell capacity realities, buck/LDO trade-offs under light load, domain partitioning, DVFS/gating, shipping mode, and evidence signatures of UVLO/brownout.

Cell capacity reality: nominal mAh vs usable energy

• Nominal capacity is measured under specific conditions; usable runtime depends on temperature, load profile, and cutoff behavior.
• Event-driven spikes (radio/LED/CPU) can pull the rails closer to UVLO even if the average current looks small.
• Practical implication: design for rail headroom at spike peaks and log resets to separate RF issues from power integrity.

Domain partitioning: who spends the battery

A fitness band should be analyzed as four power domains with independent duty cycles:

• Sensor: IMU and auxiliary sensors (mostly low baseline, occasional bursts).
• AFE + LED: PPG analog chain and LED pulses (short, repeatable peaks).
• Radio: BLE advertise/connection events (spiky, highly schedule-dependent).
• MCU/SoC: wake windows for processing and storage (often the “hidden” cost).

Effective average is driven by “active current × duty,” not by peak current alone.

Buck vs LDO: light-load efficiency is the runtime trap

• LDO: simple and quiet, but efficiency is limited by voltage drop; leakage and quiescent current still matter in long sleep.
• Buck: high efficiency at moderate load, but light-load efficiency and quiescent current can dominate in wearables.
• Key point: fitness bands spend most of their life in light-load or intermittent load. A regulator’s light-load curve can change real runtime more than its headline peak efficiency.

DVFS and gating: reduce the cost of being awake

• DVFS: lower voltage/frequency when performance is not needed; shorten wake time with predictable bursts.
• Clock gating: turn off clocks to idle blocks to cut dynamic power.
• Power gating: shut down entire domains when inactive (with retention only where needed).

Battery life improves most when work is batched per epoch so domains do not repeatedly power up for tiny tasks.

Shipping mode: controlling leakage during storage and logistics

Shipping mode is a deliberate low-leakage state that disables most domains and leaves only a minimal wake path. It prevents unpredictable drain during storage and makes “out-of-box” battery life consistent.

Evidence: UVLO/brownout signatures and “weird” step/log behavior

• Brownout clustering: resets occur near event peaks (radio events, LED pulses, flash writes), not uniformly in time.
• Data anomalies: step counts and epoch logs show discontinuities, sequence jumps, or sudden gaps after a droop event.
• Debug priority: check rail headroom and recovery before blaming RF instability.

The system goal: stable power-path + safe thermal window + trustworthy SOC

A fitness band charging design must do more than accept charge: it must keep the system rail stable under load, control heat in a tiny enclosure, and produce a state-of-charge (SOC) reading that does not jump. This requires a closed loop across charger topology, power-path, temperature control, fuel gauge, and protection.

Scope is on-device charging and power-path. High-power USB-C PD/QC/PPS behavior is excluded (charger/adapter pages).

Linear vs switching charging: heat is the first constraint in wearables

• Linear charging: simple, but heat rises with (charge current × voltage drop). Small enclosures hit thermal limits quickly.
• Switching charging: higher efficiency and less heat at comparable power, but requires tighter layout discipline and careful system integration.
• Engineering outcome: charge current and taper behavior must be set by thermal headroom, not only by “fastest possible.”

Power-path (charge-while-use): why instability happens and what “stable” means

• Two consumers: the system load and the battery. Without proper power-path control, load steps can steal current and destabilize charging.
• Stable definition: VSYS does not brown out during load peaks; charge current does not oscillate; SOC does not jump due to path switching.
• Peak events: radio events, LED pulses, and writes can cause VSYS droop if headroom and regulation are insufficient.

NTC thermal control and simplified JEITA behavior (engineering view)

• Temperature windows: cold / normal / hot bands define charging limits.
• Actions: current limit, voltage limit, or suspend charge depending on band.
• Logging: record thermal-limit events to explain “slow charge” and avoid false “battery defect” conclusions.

Fuel gauge: coulomb counting vs OCV (and why SOC jumps)

• Coulomb counter: smooth short-term tracking, but needs calibration to prevent drift.
• OCV-based estimation: requires rest and is sensitive to load and temperature; can mislead during dynamic use.
• SOC jumps often come from voltage sag under peaks, temperature transitions, and model corrections during path switching.

Protection chain: stability depends on OVP/OCP/short and battery FET behavior

• OVP/OCP/short protection prevents damage but can present as sudden shutdowns or charge interruptions.
• Battery protection FET actions can look like “random power loss” if not correlated with logs and rail measurements.
• Debug focus: correlate protection flags and rail droop timing with load peaks and charging state.

Evidence triad: SOC vs VBAT/VSYS vs temperature

• SOC: fuel gauge output and sequence/log continuity.
• VBAT/VSYS: sag under peaks, path switching transitions, and UVLO thresholds.
• Temp: NTC band crossings that explain charge current reduction and apparent “stall.”

Why “same BOM, different enclosure” creates big field variance

In a fitness band, measurement stability is dominated by coupling paths: optical leakage in the PPG window, analog ground return contamination, LED pulse ground bounce, antenna coupling to metal parts and the battery, ESD return routing, and sweat-driven leakage on high-impedance nodes. These paths can change dramatically across mechanical revisions even when the PCB schematic stays the same.

Scope is engineering evidence and physical implementation: shielding, return paths, keepout, ESD current routing, and moisture leakage mechanisms. No app/cloud discussion and no BLE spec-level deep dive.

Optical window & shielding: stop leakage before “fixing” algorithms

• Window material: scattering and reflections can raise the noise floor under ambient light.
• Shield wall & gasket: prevents LED-to-PD direct leakage and reduces sensitivity to strap tightness.
• Stack-up discipline: lens/window, adhesive, shield, PD/LED spacing should produce repeatable coupling across production.

Typical leakage signature: baseline drifting with fit changes and an abnormally large “ambient slot” reading that tracks room light.

Analog ground return: protect the AFE reference from “dirty” currents

• Analog ground island: keep TIA/ADC reference return short and predictable.
• Single-point tie: define where analog return meets system ground to avoid random current paths.
• High-impedance nodes: PD/TIA input vicinity should avoid contamination from digital or pulse returns.

LED pulse ground bounce: the hidden cause of “waveform tearing”

• Large di/dt pulses create ground bounce and rail ripple that can couple into the TIA input and ADC reference.
• Return loop control: LED pulse current must close on a high-current loop that does not pass through the AFE reference region.
• Decoupling placement: local energy must be placed close to the pulse path to reduce loop inductance.

Typical bounce signature: spikes aligned to LED pulses, slow recovery after each pulse, and elevated noise floor under high drive.

Antenna keepout & coupling: RF instability feeds back into power and analog quality

• Keepout matters: metal parts and the battery can detune the antenna and reduce efficiency.
• Field symptom: more retries and longer on-air time increase average current and shrink rail headroom.
• System impact: reduced headroom makes LED/radio peaks more likely to trigger droops that degrade analog performance.

ESD return path: “passes ESD” can still mean “measurement degraded”

• Goal: route ESD current to chassis/system reference without forcing it through AFE input vicinity.
• Common field case: functionality remains (connects, steps), but PPG quality drops—noise floor rises and baseline drift accelerates.
• Debug approach: compare pre/post ESD waveform metrics, quality flags, and baseline drift under the same light and fit conditions.

Sweat / moisture ingress: leakage and drift on high-impedance nodes

• Ion-rich moisture forms resistive bridges that bias high-impedance nodes (PD/TIA/reference networks).
• Typical symptom: drift correlates with humidity/sweat; improvement after cleaning/drying; long-term can cause calibration instability.
• Mitigation direction: sealing strategy, controlled venting/drain, and surface protection near sensitive nodes.

The purpose of a decision tree: convert goals into hardware roles

A practical reference design does not start with part numbers. It starts with constraints—battery life, motion ratio, UX priorities, enclosure/thermal limits, and yield consistency—then maps each constraint to IC roles and integration choices. Each branch should include the field symptom of a wrong choice, so validation is measurable.

Scope: selection logic by IC roles only. No brand/model recommendations and no charger-adapter protocol discussion.

Step 1 — Requirements gate: five questions that define the branches

• Battery life target: days vs weeks (sets how aggressive sleep/intervals must be).
• Motion ratio: mostly static vs high-motion dominant (sets IMU needs and PPG gating aggressiveness).
• UX priority: HR responsiveness vs SpO₂ trend stability (sets update policy and confidence gating).
• Size/thermal headroom: what heat can be tolerated during charging and peaks.
• Yield consistency: how sensitive the design can be to mechanical variance and calibration.

Step 2 — PPG branch: integrated module vs discrete AFE + LED driver

• Integrated module: faster integration and better production repeatability; less mechanical sensitivity if the window stack is standardized.
• Discrete AFE + LED driver: higher tuning freedom for dynamic range, timing, and optical constraints; higher responsibility for layout and mechanics.

Wrong-choice symptom examples: using an over-simplified solution leads to motion instability and saturation recovery issues; over-customizing increases yield variance and mechanical sensitivity.

Step 3 — IMU branch: ultra-low-power wake type vs higher-performance type

• Wake-centric IMU: best for long static periods—minimizes wakeups via FIFO/interrupt and keeps average current low.
• Higher-performance IMU: best for high-motion activity—preserves correlation and epoch alignment under strong dynamics.

Wrong-choice symptom: clipping or insufficient ODR breaks RMS-to-artifact correlation and makes quality gating unreliable; over-spec IMU increases baseline drain.

Step 4 — BLE SoC branch: memory and low-power behavior for “no-loss data loop”

• Memory (RAM/Flash): enough to buffer epochs, attach sequence numbers, and resume after disconnect without duplication or reordering.
• Sleep/wake cost: low RTC-domain overhead and predictable wake windows for batching work.
• Peripherals: sufficient I²C/SPI and timers to align PPG + IMU epochs without frequent CPU wake.

Wrong-choice symptom: data gaps/duplicates after dropouts, slow sync, and night-time disconnects that correlate with peak events.

Step 5 — Power branch: regulator + charging + gauge/protection as a stability system

• Regulators: prioritize light-load efficiency and clean domain partitioning (sensor/AFE/radio/MCU).
• Charging: decide whether power-path is required; enforce thermal windows via NTC logic.
• Gauge/protection: SOC credibility and protection behavior must match the expected load peaks and environmental conditions.

Wrong-choice symptom: unexpectedly short runtime, charge-while-use brownouts, SOC jumps, and “random” shutdown events that cluster at peaks.

Reference profiles (role sets): output templates without part numbers

• Yield-first profile: integrated PPG module + wake-centric IMU + SoC with strong retention/buffering + light-load-optimized rails + stable power-path.
• High-motion stability profile: discrete AFE + strong LED timing control + higher-performance IMU + tighter grounding/mechanics discipline + conservative rail headroom.
• Ultra-long runtime profile: aggressive batching, strict sleep domains, minimal wake sources, and regulators optimized for microamp-level idle behavior.