Scope lock: “posture trainer patch” vs a generic EMG patch

A generic EMG patch can “look fine” on a bench while failing on-body. A posture trainer patch has a stricter boundary: it must produce repeatable posture cues (state + trigger) under movement, skin impedance drift, and electrical aggressors (charger ripple and BLE TX bursts).

Minimum viable architecture (what must exist on the patch)

• Flex electrodes + interface: contact stability, bias path, lead-off detection, ESD entry control.
• EMG AFE: high input impedance + high CMRR, gain staging, anti-alias, saturation recovery.
• ADC + on-patch DSP: baseline control, 50/60 Hz residual handling, envelope/feature extraction.
• IMU (6-axis) + fusion: bias/drift management, orientation proxy, posture state + hysteresis.
• BLE SoC + antenna: duty-cycled link, scheduling, RF/PDN coupling containment.
• Battery + charger/PMIC: analog/digital domain separation, charge-while-measuring constraints.

What “posture output” means in hardware terms

For a patch, “posture” is not a UI feature—it’s a set of measurable outputs derived from IMU (and optionally EMG features) that drive a state machine:

• Tilt/roll proxies (relative orientation) with calibrated axes and bounded drift.
• State classification (e.g., upright / slouch / twist) with hysteresis + debounce to prevent chatter.
• Event triggers (vibration cue) with a defined latency budget and false-trigger constraints.

Key engineering point: posture stability is limited by IMU bias/drift and mounting shift, not by “more math”.

The 3 coupling paths that create most “real-body failures”

Almost every field failure can be mapped to one of these coupling paths. This is why the architecture diagram must explicitly show them.

• Motion → electrode interface: contact impedance changes unbalance inputs, converting common-mode into differential error and inflating “EMG” during movement.
• Charger/PMIC ripple → analog rail / ADC reference: charge switching and ground bounce leak into microvolt-level front ends, creating periodic artifacts (especially charge-while-measuring).
• BLE TX burst → PDN dip → AFE/ADC disturbance: RF events cause peak current steps; if sampling overlaps, you see steps, spikes, or drifting baseline correlated with TX timing.

Design philosophy: treat motion, charger, and TX as first-class “noise sources” with explicit evidence hooks.

Why metrics first (the engineering-grade contract)

“Works on the bench” is meaningless unless you define pass/fail targets tied to real coupling paths. This page uses four domains—EMG analog, signal quality, posture (IMU+fusion), and system (wireless+power). Each metric below includes (1) what it means, (2) how to measure it, and (3) the failure signature that points to the root cause.

Domain A — EMG analog front-end metrics (microvolt survival)

• Input-referred noise (µVrms): measure with electrode simulator / input short, bandwidth-limited; failure signature is noise rising with TX/charge events (points to PDN/coupling).
• CMRR @ 50/60 Hz: common-mode injection + output residual; if it varies strongly across users/placements, suspect electrode imbalance and bias path sensitivity.
• Input impedance & bias path stability: sweep source impedance and observe offset/drift and recovery; weak bias paths cause slow baseline wandering or sudden saturation on motion.
• Headroom + saturation recovery time: inject step-like artifact and time-to-return-to-noise-floor; long recovery makes “posture cues” unreliable even if average SNR looks fine.

Domain B — Signal quality metrics (artifact separation)

• Baseline wander: quantify low-frequency drift after filtering; if drift aligns with charge cycles, suspect charger ripple or ground reference modulation.
• Power-line residual: PSD around 50/60 Hz and harmonics; persistent residual after notch often indicates CMRR loss from electrode imbalance or layout asymmetry.
• Motion artifact amplitude vs EMG band: compare artifact energy to EMG envelope (relative metric); if artifact dominates during movement, root cause is usually interface physics, not DSP.
• Repeatability (session-to-session): same posture action should produce bounded feature variance; large variance implies changing contact impedance, not “random noise”.

Domain C — Posture metrics (IMU + fusion realism)

• IMU bias stability & drift: log bias estimates and angle drift over 10–20 minutes; drift that correlates with skin/adhesive shift points to mechanical mounting, not firmware.
• Axis alignment error: quantify misalignment sensitivity; posture thresholds must include hysteresis to tolerate small mounting differences.
• Latency budget: measure time from motion to trigger (fusion + decision + BLE/haptic); spikes in latency often correlate with BLE scheduling collisions.

Domain D — System metrics (wireless + power as root cause indicators)

• Battery life (duty-cycle breakdown): partition current into AFE, IMU, compute, BLE, haptic; unexpected drain often comes from retries and high TX power.
• Packet Error Rate (PER) + retry counters: treat PER as an “aggressor amplifier”—more retries → more bursts → more PDN dips → more analog corruption.
• Rail ripple & TX-burst droop: measure analog rail ripple and digital droop with event markers; time correlation is the fastest proof of coupling.
• Charge-while-measuring performance: define a pass condition (artifact amplitude cap + no baseline steps) while charging; if it fails, isolate by changing charge mode and sampling slots.

“Two-point evidence” rule (fast isolation)

For every failure, always capture two evidence points: (1) an analog observable (AFE output / ADC code / PSD), and (2) an aggressor marker (TX event, charge switching, rail droop, lead-off toggles). If they correlate in time, you have a root-cause direction before touching algorithms.

Reality check: the interface dominates on-body behavior

In a posture trainer patch, the “signal source” is not a stable voltage. The electrode–skin interface behaves like a time-varying impedance network. During motion, micro-slips and pressure changes modulate the interface impedance, creating artifacts that can exceed the EMG amplitude and push the AFE into saturation.

Dry vs hydrogel: why contact variability dominates

• Dry electrodes: typically higher impedance and more sensitive to sweat/pressure; more likely to amplify imbalance-driven artifacts.
• Hydrogel electrodes: typically lower and more stable impedance; better repeatability, but lifetime and drying can become a constraint.
• Engineering takeaway: if performance varies strongly across users or placements, treat the interface as the primary suspect before blaming DSP.

Minimal impedance model (actionable, testable)

A useful “smallest model” per electrode is Rs + Cdl, plus a motion-induced modulation term. The imbalance between the two electrode impedances (ΔZ) is what converts common-mode interference into differential error at the AFE inputs.

• Rs: contact/skin resistance component (strongly motion- and sweat-dependent).
• Cdl: double-layer capacitance component (affects low-frequency baseline behavior and settling).
• ΔZ imbalance: the key “artifact amplifier” for common-mode pickup.

Design mindset: reduce ΔZ sensitivity and keep the AFE out of saturation; filtering cannot recover clipped information.

Biasing + driven reference: common-mode management on a patch

Stable measurement requires a defined input bias path and a controlled common-mode operating point. Without these, slow baseline drift, sudden saturation, and long recovery are common—especially when the interface impedance changes.

• Bias path: provide a stable return for input bias currents without creating large imbalance.
• Driven reference (RLD/DRL-style): reduce common-mode excursions and improve robustness under imbalance.
• Common-mode headroom: ensure the front-end does not hit rails during motion artifacts or environmental interference.

Lead-off detection: DC vs AC injection tradeoffs (and why false alarms happen)

• DC methods: simple and low power, but sensitive to sweat and motion-induced impedance steps → false lead-off during activity.
• AC injection: can be more robust, but requires careful frequency placement and symmetry to avoid self-injecting artifacts.
• False lead-off pattern: lead-off toggles correlated with IMU motion energy and EMG saturation events usually indicate interface instability, not “bad firmware”.

Placement & flex constraints (what is controllable in hardware)

• Cable-less flex routing: keep electrode traces short, symmetric, and low loop-area to reduce pickup and imbalance.
• Strain relief: micro-movement at the electrode pad behaves like impedance modulation → artifact bursts.
• Sweat ingress: changes impedance and leakage paths; design for repeatable contact and stable biasing across humidity states.

Practical rule: repeatability beats “best-case lab SNR” for posture triggers.

Design goal: survive artifacts first, then optimize noise

On-body EMG fails when the front end clips or recovers slowly under motion/common-mode stress. A robust architecture prioritizes headroom, fast saturation recovery, and imbalance-tolerant CMRR, then uses filtering and gain staging to reach the target noise floor.

Front-end choice: INA vs chopper vs dedicated EMG AFE

• INA-based: focus on CMRR retention under ΔZ imbalance, input bias behavior, and EMI tolerance at the input pins.
• Chopper-stabilized: helps offset/drift, but verify switching artifact immunity and EMI behavior in the presence of long electrode traces.
• Dedicated EMG AFE: often integrates lead-off and protection hooks; check input impedance, noise, dynamic range, and recovery behavior under large artifacts.

Selection rule: the “best” part is the one that stays linear during motion artifacts and returns quickly when disturbed.

Gain staging: avoid early clipping with multi-stage strategy

• Stage 1 (front-end): moderate gain to preserve headroom against motion/common-mode swings.
• Stage 2 (post-HPF/LPF): apply additional gain after baseline control and anti-alias conditioning.
• Evidence hook: if clipping appears only during movement or TX/charge events, reduce early gain and verify recovery time before tuning filters.

Filtering decisions: HPF, LPF/anti-alias, and notch placement

• HPF (baseline control): suppress slow drift and baseline wander; avoid setting corners so high that useful EMG content is removed.
• LPF + anti-alias: prevent out-of-band noise from folding into the EMG band; match to sampling rate and ADC front-end.
• Notch (50/60 Hz): analog notch can reduce early overload; digital notch is flexible but cannot recover information lost to saturation.

Practical order: keep the chain linear → prevent alias → then refine notch and feature extraction.

ADC choice: resolution, bandwidth, power—and reference stability

• Resolution: only helps if analog noise and reference noise are below the quantization floor in-band.
• Bandwidth / sampling rate: must support the target EMG band with margin for filtering and artifact handling.
• Reference integrity: TX bursts and charging ripple often enter via ADC reference / ground, producing steps or drift-like errors.

Input protection: ESD without killing noise (symmetry matters)

• Leakage + capacitance at the input becomes part of Rs/Cdl and can worsen drift and imbalance.
• Asymmetric protection creates ΔZ and degrades effective CMRR, turning common-mode pickup into differential artifacts.
• Rule: prioritize symmetric networks and controlled leakage over “stronger clamps” that destabilize the interface.

Scope: a bounded on-patch DSP stack (no ML platform)

The goal is not clinical-grade waveform reconstruction. The goal is stable, repeatable posture cues on a small patch: (1) a controlled baseline, (2) predictable latency, and (3) low false-trigger rate during dynamic motion. This chapter stays at patch level: simple filters, envelope/RMS, and deterministic trigger logic.

Core pipeline: envelope + RMS (fast, bounded, testable)

• Rectify + LPF: produces an EMG envelope with low compute cost; tuning trades responsiveness vs smoothness.
• RMS windowing: stabilizes amplitude estimates; window length trades noise rejection vs trigger delay.
• Median / spike handling: limits the impact of impulse artifacts (contact steps, ESD-like events) on envelope/RMS.

Engineering rule: choose parameters by trigger repeatability and delay, not by “filter order”.

Motion artifact mitigation: adaptive baseline control + IMU-gated updates

Motion artifacts often arise from electrode impedance modulation and can dominate the EMG band after saturation/recovery. A practical approach is to protect long-term stability: gate or down-weight envelope/threshold updates during high-motion segments, using IMU energy (acceleration magnitude variance, gyro magnitude, or jerk proxies).

• Adaptive HPF limits: constrain baseline update speed so slow drift is removed without “learning” motion artifacts.
• Artifact gate: when IMU energy is high, freeze baseline/threshold adaptation and suppress triggers.
• Evidence hook: if false triggers correlate with IMU energy bursts, gating is the first lever before more complex DSP.

50/60 Hz handling: notch + harmonics strategy with PSD proof

• Notch placement: use notch after artifact gating when possible, so gating decisions are not contaminated by line pickup.
• Harmonics: check 2nd/3rd harmonics if the environment or nonlinearity creates residual peaks.
• Verification: validate residual line peaks in PSD and confirm they do not drive envelope/RMS triggers.

Acceptance mindset: demonstrate “residual is below trigger influence”, not “a perfect-looking waveform”.

Posture-ready features (patch level): deterministic, drift-resistant

• Activation symmetry: compare channels (if 2-ch) or compare time segments (if 1-ch) to detect consistent activation patterns.
• Fatigue proxy: use low-compute trends (envelope statistics over time) to modulate cue intensity without medical claims.
• Event triggers: threshold + hysteresis + debounce; suppress during dynamic motion to reduce false cues.

Boundary: features are for cue triggering, not for diagnosis or rehabilitation protocols.

Placement reality: misalignment is normal and must be absorbed

A posture patch is not a rigid body fixed to a skeleton. Small placement angle errors, soft tissue movement, and micro-slips map directly to posture angle offsets and state flicker. Robust posture estimation starts by treating misalignment as an expected condition, not an exception.

Calibration: accel/gyro bias and “no-mag” clarity

• Gyro bias: estimate in a stationary window; track bias value as a logged variable over time.
• Accel sanity: use gravity vector consistency checks to validate scale/offset and detect gross placement changes.
• No magnetometer (common for patches): avoid magnetic disturbances; accept that yaw is weakly observable and focus on tilt/roll proxies.

Log to prove the cause: bias estimate, stationary flag, and posture angle drift slope.

Fusion choices: complementary vs EKF (decide by evidence + power)

• Complementary filter: low compute, few parameters, stable for patch-level tilt/roll; preferred when power budget is tight.
• EKF: stronger modeling but higher complexity; require explicit logging and validation to avoid “black-box tuning”.
• Proof logging: fused angle, gyro bias estimate, stationary detect, and drift monitor counters are sufficient to isolate drift vs placement issues.

Posture states: thresholds + hysteresis + debounce (avoid dynamic false cues)

• Angle thresholds: define posture boundaries using tilt/roll proxies relevant to patch placement.
• Hysteresis: prevent state chattering near boundaries when noise or micro-motions occur.
• Debounce: require persistence before state change; coordinate with IMU motion energy to suppress cues during dynamic motion.

Failure signatures: no hysteresis → flicker; too-short debounce → walking triggers cues; too-long debounce → sluggish correction feedback.

Drift loop + recalibration triggers (keep long-term stability)

A practical posture estimator includes a drift monitor: when the device is stationary, fused angles should remain stable. If a slow drift persists during stationary periods, trigger a recalibration action (bias update) and optionally a user-facing cue if placement change is suspected.

• Stationary-driven bias update: update gyro bias only when stationary detect is confident.
• Recal trigger: drift slope above a bound during stationary windows → recal step.
• Placement change hint: drift + interface instability (lead-off/EMG artifact counters) together suggest re-attachment rather than pure IMU drift.

Why power integrity dominates microvolt EMG

In a posture patch, EMG is often “broken” by power and return-path behavior rather than DSP. Charger switching ripple, rail dips from wireless bursts, and ground bounce can appear as baseline wander, false activation, or envelope inflation. This section builds an evidence-first power design approach: isolate, filter, and schedule.

Power domains: separate what injects noise from what measures microvolts

• Analog domain: EMG AFE rail + ADC AVDD/REF; treat as a “quiet island” with controlled return paths.
• Digital/RF domain: BLE SoC, storage, haptic drivers; expect burst currents and fast edges.
• Charger domain: switching node ripple, inductor current ripple, and charge current steps.

Design rule: isolation is not only about LDOs; return paths and reference points decide the outcome.

Return paths & star points: “ground” is not a single node in EMG systems

EMG inputs are sensitive to common-mode movement turning into differential error through mismatch. Ground bounce and reference movement couple into the AFE/ADC through supply pins, reference pins, and asymmetric bias/protection networks. A controlled star point and short, predictable return loops reduce CM-to-DM conversion.

• Watchpoints: AFE rail ripple, ADC reference ripple, and ground bounce during charger/RF events.
• Symptom signature: repeatable baseline steps aligned with switching or TX timing.

Charger coupling: why out-of-band ripple still corrupts the EMG band

• Ripple folding: switching ripple can alias into the EMG band through sampling and reference modulation.
• Load-step bounce: charge-current changes and mode transitions can create ground/reference steps.
• AFE vulnerability: PSRR and CMRR are not infinite across frequency and mismatch conditions.

Evidence-first: correlate EMG baseline/envelope artifacts with charger switch timing and mode changes.

Measure while charging: patch-level “fix levers” that actually work

• Slow-charge mode: disable fast-charge or reduce di/dt to cut ground bounce and ripple amplitude.
• Frequency shift: move switching frequency away from sensitive windows and sampling/alias zones.
• Scheduling: create clean sampling slots; freeze baseline adaptation during noisy periods.
• Filtering: strengthen local decoupling at AFE/REF nodes; keep the quiet island isolated from the burst island.

Fuel gauge / protection (patch scope): brownout looks like “EMG weirdness”

• UVLO / cutoff events: short resets or rail droops can create data gaps and envelope spikes.
• Brownout drift: ADC/reference movement can mimic activation changes.
• Proof: check reset flags / brownout counters aligned with the EMG anomaly timestamps.

Boundary: only power events that directly corrupt on-patch measurements are covered here.

First 2 measurements (minimal tools, maximum clarity)

• CH1: AFE rail or ADC reference node — look for ripple, steps, and mode-transition signatures.
• CH2: digital/BLE rail or ground bounce proxy — align to TX bursts and charger switch timing.

Decision: if CH1/CH2 anomalies time-align with EMG baseline/envelope corruption, prioritize isolation + scheduling before DSP tweaks.

Board-level coexistence: power bursts and RF injection are the two failure modes

Wireless can corrupt EMG through (1) burst current events that dip rails or move references, and (2) RF energy coupling into high-impedance electrode traces on a flex substrate. This section stays at board level: observe, correlate, then fix with scheduling and layout boundaries.

TX burst current → rail dip → ADC/reference error (how to observe it)

• Correlate timing: log BLE event timestamps (or toggle a GPIO) and align to EMG baseline/envelope artifacts.
• Measure rails: probe digital rail and AFE/REF nodes during TX bursts; look for short dips or steps.
• Typical signature: repeatable baseline step or envelope bump after each TX event, stronger when retransmissions rise.

Fix priority: stabilize references and schedule sampling away from bursts before increasing DSP complexity.

Antenna coupling into electrode traces: flex routing and keep-outs

• Keep-out zones: maintain distance between antenna/feed and electrode traces; avoid parallel runs on flex.
• Loop area control: route electrode pairs to minimize loop area; keep symmetry to reduce CM-to-DM conversion.
• Edge sensitivity: high-impedance nodes at the electrode interface are most susceptible; protect with careful routing and shielding strategy.

Boundary: this is layout coexistence, not antenna matching or protocol deep dive.

Duty-cycling: create “safe sampling slots” and buffer around RF activity

A practical patch strategy is to define explicit EMG sampling windows and place BLE activity outside them as much as possible. Buffer samples locally, then transmit in RF windows. When charging or haptic events occur, treat them as additional timing constraints.

• Safe slot: a time window with stable rails and low RF activity for EMG conversion.
• Buffering: store EMG blocks locally so sampling can be decoupled from transmit timing.
• Degradation mode: when safe slots are unavailable, freeze baseline adaptation and raise trigger robustness.

Robust link metrics: evidence-first debugging (PER/RSSI trends, retrans counters)

• Trend-based view: monitor PER/RSSI over time and align with “EMG corruption” timestamps.
• Retransmissions: rising retries increase TX density and raise the chance of rail disturbance and coupling.
• Decision: if link degradation and EMG anomalies co-occur, suspect a shared root cause (power/placement/layout) rather than DSP.

Minimal debug loop: isolate the culprit fast

• Step 1: reduce TX density (longer intervals or temporarily disable transmit) and check if EMG immediately cleans up.
• Step 2: re-enable TX but enforce safe sampling slots; verify artifact reduction.
• Step 3: if coupling persists, enforce keep-outs / symmetry and re-check with the same timing correlation method.

Always keep the same evidence alignment method; only change one lever at a time.

Flex reality: routing + shielding + stress decide EMG reliability

In a posture patch, the electrode interface is a high-impedance, motion-sensitive system. On flex, long traces, asymmetry, and uncontrolled return paths can convert common-mode disturbance into differential error. Mechanical micro-motion and sweat-driven leakage often appear as “EMG drift” or false activation. This section provides reviewable, zone-based rules.

Top-level zoning: place analog close, push RF/power away

• Electrode zone: pads and the shortest possible run to the front-end.
• Analog zone: input protection + AFE + ADC/ref; treat as the “quiet island”.
• Digital/RF zone: MCU/BLE and antenna; keep-out around the antenna feed and radiator.
• Power/charge zone: charge pads, switching components, buttons; treat as “dirty interfaces”.

Review rule: if an electrode trace crosses a digital/RF zone, expect artifacts first and debug later.

Electrode trace routing: symmetry, guard, and minimal loop area

• Symmetry: match length, bends, and environment so CM disturbances do not become DM error.
• Loop control: route the pair to minimize loop area; avoid wide separation on flex.
• High-impedance discipline: keep the electrode-to-AFE segment short, with minimal discontinuities.
• Guard concept: where appropriate, define a guard/shield zone around the most sensitive segment.

Failure signature: “one side triggers more” or “hand proximity changes EMG baseline” often indicates asymmetry + loop area issues.

Shielding strategy: grounded shield vs driven shield (bounded tradeoffs)

• Ground shield: simplest; must avoid creating a noisy return path near the electrode interface.
• Driven shield: can reduce leakage and capacitive pickup at high-impedance nodes, but requires stable, controlled drive behavior.
• Flex constraint: shield zones should respect antenna keep-outs and avoid bridging analog and digital return paths.

Evidence rule: compare PSD and 50/60 Hz residual with shield variants; keep only what improves measurable metrics.

Connectorless integration: stiffeners, stitching, and strain relief

• Stiffeners: protect solder joints and stabilize the AFE region near the electrode interface.
• Via stitching: reinforce shield/return continuity in defined zones (without shorting domains).
• Strain relief: ensure bend lines do not pass through high-impedance nodes or critical reference paths.
• Adhesive stack-up: avoid periodic micro-motion that modulates electrode impedance under posture movement.

Failure signature: “works on bench, fails when walking” often indicates mechanical micro-motion and flex stress concentration.

Sweat & corrosion (bounded): leakage paths and drift prevention

• Leakage control: keep high-impedance nodes protected from sweat-driven conductive films.
• Moisture boundary: define protective regions around electrode routing and AFE input structures.
• Trend check: test for baseline drift and lead-off instability under controlled humidity / sweat simulation.

Boundary: focus is on preventing electrical drift and false lead-off, not a materials dissertation.

ESD entry points: map where “human touch” injects energy

• Electrodes: direct user contact and large exposure area.
• Charge pads / USB: external interface with frequent plug/touch events.
• Buttons: repeated touches and edge exposure.

Design intent: define protection so AFE input integrity and power rails recover without lasting drift.

Validation mindset: stimulus → must-capture → pass criteria → failure signature

The goal is to catch “body failures” before field trials: motion artifacts, impedance drift, lead-off false positives, and coupling from charging/wireless bursts. Each test below defines what to inject, what to record, and what proves the root cause.

Test matrix overview (what to test, what to log)

• Bench emulation: electrode impedance box + motion/impedance modulation + common-mode injection.
• Noise/CMRR: PSD-based noise and 50/60 Hz residual; saturation & recovery time.
• Lead-off robustness: false-positive/negative under motion and sweat simulation.
• Charge + measure: ripple spectrum, rail steps, and artifact correlation to modes/events.
• Wireless robustness: PER/RSSI and retrans counters vs posture motion and sample scheduling.
• ESD/EMC sanity: where to zap, which rails/flags to monitor, and recovery behavior.

Pass criteria should map to the “metrics targets” defined earlier (noise, residual hum, drift, PER, recovery).

Bench emulation: make “human variability” controllable

• Electrode impedance box: adjustable Rs/C model to reproduce contact variability repeatably.
• Motion artifact injection: impedance modulation to mimic micro-motion and contact shift.
• Common-mode injection: controlled 50/60 Hz CM to validate CM-to-DM behavior under mismatch.

Must-capture: raw EMG, envelope/RMS, lead-off flag, IMU energy, and AFE/REF rail waveforms.

Noise/CMRR verification: PSD + residual hum + saturation recovery

• PSD checks: confirm noise floor and post-filter 50/60 Hz residual (and harmonics).
• Mismatch sensitivity: test with impedance imbalance to reveal CM→DM conversion.
• Recovery time: after saturation or injected disturbance, measure time to return to stable baseline.

Failure signature: clean bench noise but strong residual hum under mismatch indicates routing/shield/return-path issues.

Lead-off tests: avoid false positives during motion/sweat

• Motion + sweat simulation: combine impedance drift and modulation to reproduce field conditions.
• Measure: false-positive rate, false-negative rate, and recovery time after re-contact.
• Correlate: lead-off flags vs IMU energy and EMG baseline/envelope to distinguish contact vs power events.

Decision: if lead-off toggles align with IMU motion but not with rail events, prioritize mechanical/contact fixes.

Charge + measure + wireless: correlation tests that isolate coupling

• Charge modes: fast/slow/disabled fast-charge; switching frequency options if available.
• Wireless load: vary TX density and retransmissions; observe rail dips and EMG corruption signatures.
• Scheduling: enforce safe sampling slots; compare artifact rate against “no scheduling” baseline.

Must-capture: AFE rail/ADC REF, TX event timestamps, IMU energy, reset/brownout flags, and EMG anomaly timestamps.

ESD/EMC sanity checks: where to zap, what rails to watch

• Zap points: electrodes, charge pads/USB, buttons (human-touch entry map).
• Monitor: AFE rail, digital rail, ADC reference, reset/brownout flags, and baseline recovery behavior.
• Pass intent: system recovers without lasting drift or persistent lead-off instability.

Boundary: “sanity checks” to reveal gross vulnerabilities early; not a full certification walkthrough.

Minimal logging requirements (to avoid “mystery failures”)

• Timestamps: BLE events (TX/RX/retry), charge mode transitions, haptic cues.
• Counters: PER, RSSI trend, retransmissions, resets/brownouts.
• Signals: raw EMG (or pre-envelope), envelope/RMS, lead-off state, IMU energy metric.

If a failure cannot be time-aligned to a stimulus or rail event, it will be misdiagnosed as “algorithm noise”.