This page turns a “simple” reed switch contact into an evidence-driven, production-ready node design: contact physics → debounce/threshold → event logic → EMC/ESD → ultra-low power and field debug.
You’ll learn exactly what to measure (waveforms, current, counters) to stop false alarms, avoid missed events, and hit multi-year battery life without drifting into system/platform topics.
H2-1. Definition & Boundary (What this page solves)
30-second scope lock
A magnetic contact (reed switch) is an event-input node that converts a door/window magnet state into a
debounced, threshold-qualified open/close/tamper event. This page stays at node level and uses an
evidence chain that always maps back to: contact → threshold/debounce → event → counters/logs.
This page owns: reed contact behavior, debounce/threshold front-end, MCU wake/event logic, node power/leakage,
node-level wireless retry evidence (if wireless), tamper switch inputs, and node-level EMC/ESD/EFT effects.
No platform/system deep-dive
Typical use cases (node viewpoint)
Open/Close sensing for doors, windows, cabinets
Tamper (cover open, device removal) as a separate event
Figure F1. Keep the discussion at the reed-contact event node. Any design choice must map back to
measurable evidence: bounce/threshold crossings, event counters, and power/leakage contributors.
H2-2. Reed Switch Fundamentals That Matter (engineering-only)
Physics that changes false/miss rates
Reed switches look binary, but field failures often come from geometry tolerance (gap/offset/angle),
operate/release distance spread, and contact chatter under vibration. Treat the reed input as an
analog phenomenon that must be qualified before becoming an event.
NO vs NC (fault-safety lens)
NC loop: tends to expose open-circuit faults sooner (fail-safe behavior), but may be more sensitive to wiring/connector intermittency.
NO loop: may simplify some nodes, but broken wires can look like “normal” unless supervision is present.
Practical rule: pick the contact mode that makes the worst fault unmistakable for the intended installation and supervision method.
Operate/Release distances are distributions
Operate distance: where the reed closes; release distance: where it opens.
They vary with magnet strength, part tolerances, mounting stress, and angle.
Design to the worst-case tail (e.g., P95/P99), not the “typical” distance.
Geometry variables that dominate behavior
Gap: primary sensitivity axis; small changes can flip the event outcome.
Offset: lateral misalignment reduces effective field at the reed.
Angle: orientation can shift both operate and release points noticeably.
Field direction: magnet polarity/orientation can improve or worsen repeatability.
Chatter / bounce (why “one action” becomes many edges)
Contact bounce: multiple rapid transitions at closure/opening.
Door rebound: mechanical oscillation near the threshold causes repeated close/open.
Vibration + near-threshold: turns marginal geometry into bursts of interrupts and false alarms.
Design implication: physics here dictates the minimum need for hysteresis and time qualification (implemented in H2-3/H2-5).
What to measure (fast, evidence-first)
Operate/release sweep: scan gap (and angle) and record the close/open points per unit.
Edge burst count: count raw transitions during a single door action (bounce/chatter indicator).
Environment corners: low temperature + low battery voltage can change margins indirectly.
How to frame pass/fail without system creep
Define a node-only window: “Must close by gap ≤ A, must open by gap ≥ B, across angle ±θ.”
Define a false edge budget: raw burst ≤ N within T ms for the expected door mechanics.
These become inputs to debounce/threshold design and validation (later chapters), while staying within node scope.
Figure F2. Reliability depends on gap/offset/angle margins and the spread of
operate/release distances. Near-threshold geometry amplifies vibration into chatter, which must be handled
by threshold hysteresis and time qualification (later chapters).
Cite this figure:Figure F2— “Magnet–Reed Geometry: gap/offset/angle and operate/release margin concepts.”
The reed input is not “clean digital.” Field issues usually come from bounce, near-threshold chatter,
or noise-driven threshold crossings. The front-end must convert that analog reality into
a single qualified event with measurable margins.
Rule: treat the reed pin as a signal-processing problem. Every topology choice must be justified by evidence:
raw waveform → crossing count → qualified output.
Topology map (what each block is good at)
Pull-up + RC: removes fast bounce energy; simplest “first line” filter. Boundary: RC too large adds latency and can hide fast valid actions.
Schmitt trigger: adds hysteresis around the threshold; strong against near-threshold chatter. Boundary: too much hysteresis can create missed events at large gaps.
Comparator + hysteresis: tunable thresholds/windowing; suitable when long cables or supervision windows demand repeatable thresholds. Boundary: input bias/leakage and reference drift must be budgeted.
Digital qualification (time-window / majority / edge-qualification): converts “edge bursts” into one event and enables counters. Boundary: a noisy pin can create wake/interrupt storms unless analog conditioning is present.
RCSchmittComparatorTime qualification
How to choose the time constant (evidence-first)
Measure bounce duration distribution on real mechanics (not only bench toggling). Record P50/P95/P99 bounce length.
Set a latency budget for event confirmation (e.g., acceptable delay for alarm/reporting) and keep debounce inside that budget.
Pick RC / time-window to cover the tail (P95/P99), then verify it does not suppress legitimate “quick open/close” actions.
Validate with counts: one physical action should yield raw_edges > 1 but qualified_events = 1.
Oscilloscope evidence (minimum set)
REED_RAW: direct pin (or before RC) to see bounce/chatter shape.
RC_NODE: after RC to confirm attenuation vs latency.
SCHMITT/COMP_OUT: thresholded signal to count crossings and check hysteresis behavior.
MCU_IRQ: interrupt line (if used) to detect wake storms.
Counter/log evidence (must exist later in firmware)
wake_count and interrupt_count (to reveal noise-induced wake storms)
debounce_reject_count (how often a burst is rejected)
Figure F2. Convert a noisy reed waveform into a single qualified event using RC + hysteresis and
time qualification. Validate using evidence: raw_edge_count vs qualified_event_count.
Cite this figure:Figure F2— “Debounce pipeline: bounce → RC/threshold → single qualified event.”
H2-4. Wired Loop & Line Supervision (EOL) — node-level only
Open / Short / Normal / Alarm without panel creep
Wired magnetic contacts often fail in the field due to broken wires, shorts/moisture, and
noise on long cables. Node-side EOL supervision turns those failures into
distinct resistance (or voltage) windows that can be tested by injection.
Scope note: this section covers node-side loop measurement and windowing only.
Zoning/arming logic and system-wide panel behavior are out of scope.
Single EOL vs Dual EOL (what changes at node level)
Single EOL: simpler ladder; can detect key faults but may have less separation depending on wiring mode.
Dual EOL: clearer four-state separation, but requires tighter window budgeting (tolerance + line resistance + drift).
Design goal: maintain non-overlapping windows across temperature, cable length, and component tolerance.
Comparator windowing: low-power, instant; requires stable references and hysteresis design.
Noise reality: long cables can inject transients; add input conditioning but avoid “smearing” windows into overlap.
Evidence that proves correct supervision
Resistance/voltage window map: Normal / Alarm / Open / Short boundaries defined as ranges.
Open/short injection test: force each fault and verify the node lands in the intended window.
False classification rate: under EFT/noise, log “window-flip” counts per hour (node-level statistic).
Long cable trade-offs (node-only)
Line resistance shifts measured windows; budget it explicitly.
Common-mode noise creates brief threshold crossings; use RC + hysteresis at the measurement node.
Protection leakage can distort high-impedance ladders; select parts and bias values with leakage in mind.
Figure F3. EOL supervision is a windowing problem: define non-overlapping ranges for
Normal / Alarm / Open / Short, then validate with open/short injection and log “window flips” under cable noise.
Cite this figure:Figure F3— “Four-state EOL windows: Normal/Alarm/Open/Short mapped to decision regions.”
A reed switch “event” is only trustworthy when the firmware can prove it survived bounce, chatter,
and noise-driven wakes. This chapter turns the chain into a reusable SOP:
wake source → sample → qualify → commit (counters/log) → report.
Interrupt & wake source selection (avoid wake storms)
GPIO IRQ: prefer a trigger mode that minimizes repeated firing under near-threshold chatter. Track irq_count separately from wake_count.
Wake gating: if available, use LP domain (e.g., low-power comparator / simple latch) so the main MCU only wakes after a stable condition.
Backoff: after a valid event, enforce a short “quiet” window to suppress mechanical rebound bursts (evidence via debounce_reject_count).
irq_countwake_countbackoff
Debounce algorithms (pick by failure mode)
Time-window: confirm state only if stable for T. Best against bounce + chatter. Risk: added latency if T is large.
Majority vote: N samples within a short window; accept if ≥K match. Best against sporadic noise. Risk: longer awake time.
Edge qualification: require minimum pulse width or minimum spacing. Best against narrow EMI spikes. Risk: miss slow/soft transitions if thresholds are marginal.
Minimum counters (field-debug essential set)
irq_count, wake_count (per hour/day)
raw_edge_count (per action window)
bounce_reject_count (or debounce_reject_count)
event_count and event_rate
tamper_count (if tamper input exists)
retry_count / tx_fail_count (wireless nodes only; still node-level)
Evidence chain requirement: one physical action should produce raw_edge_count > 1,
but event_count increments by 1. If wake_count grows faster than event_count,
suspect chatter/noise or wrong IRQ configuration.
Figure F4. Firmware event SOP and power evidence share the same backbone:
a state machine with counters/logs, plus an average-current breakdown that ties
sleep leakage, wake storms, and TX retries to measurable evidence.
Cite this figure:Figure F4— “Event state machine + average-current evidence model for reed nodes.”
H2-6. Ultra-Low Power Design (Leakage, Battery Life & Root Causes)
Years of battery life without false alarms
Reed nodes look simple, but battery life is usually destroyed by unexpected leakage and
unplanned wake/report activity. The only reliable method is evidence-driven budgeting:
sleep leakage + wake sampling + TX retry — each tied to a waveform and a counter.
Average current breakdown (turn symptoms into a budget)
Sleep leakage (uA): dominates long life. Measure after full settle and across temperature.
Wake sampling (mA × ms): driven by wake_count and qualification duration (N, T, K).
TX retry spikes: driven by retry_count; worsened by low RSSI and low-temp droop.
Iavg modelwaveform + counters
Leakage kill list (common “battery drops fast” root causes)
Divider / pull resistors: too low wastes current; too high becomes noise/leakage sensitive.
ESD/TVS leakage: high-impedance nodes shift levels; humidity/high temp makes it worse.
LDO IQ: an always-on regulator can dominate sleep current.
GPIO float / back-power: floating inputs or protection paths cause hidden current.
Wake storms: noisy thresholds cause frequent wake_count growth without real events.
RF retries: retry_count increases create “spike trains” and fast depletion.
Evidence checklist (what must be captured)
Current waveform: uA sleep baseline + wake plateau + TX spike train.
Event statistics: event_rate, wake_count, retry_count aligned to waveform spikes.
Voltage & resets: battery_mv under TX and reset_reason histogram (brownout correlation).
Battery selection (node-level only)
Internal resistance and pulse capability: determines droop during TX spikes.
Low-temperature behavior: IR rises, droop worsens, brownout becomes likely even with “remaining capacity”.
Acceptance test: verify battery_mv and reset_reason under worst-case low-temp + retry scenario.
Root-cause discriminator: if wake_count grows much faster than event_count, suspect
noise/chatter or qualification too permissive. If retry_count grows with temperature drop,
suspect low-temp droop and battery IR — confirm via battery_mv and reset_reason.
Misreports → wake storms → retries → fast battery drain
Wireless turns small input issues into large energy losses: extra wakes often become extra transmissions,
and weak signal turns each event into a retry spike train. This chapter stays strictly node-side:
only what the node can measure, limit, and prove with counters and logs.
Message policy (reduce TX count without hiding evidence)
Battery droop → battery_mv_min drops during TX spikes
Reset loop → reset_reason_hist correlates with retry spikes
Figure F5. Node-side evidence closes two hard problems without system creep:
wireless drain loops (weak RSSI → retries → droop → reset) and tamper credibility
(tamper switch, magnet spoofing, replay/inject) using counters, event_seq, and timestamps.
Cite this figure:Figure F5— “Wireless drain chain + node-level tamper evidence with event sequence.”
Prove events are trustworthy (without platform scope creep)
Security products must explain why an “open/close/tamper” event is credible.
This chapter stays node-level: physical tamper signals, magnetic spoofing evidence,
and simple trust primitives that create an audit trail without relying on any cloud or panel logic.
Tamper switch (cover open / removal)
Debounce tamper input like any mechanical contact. Track tamper_debounce_reject to distinguish bounce from real open.
Commit-first: store tamper event_seq + timestamp before reporting to survive resets.
Evidence: tamper_count, raw_edge_count (or reject_count), and a short log sample around the event.
Magnet spoofing (strong magnet hold / manipulation)
Detect anomalies using node-observable inconsistencies (unexpected persistence, abnormal patterns, or multi-input mismatch if available).
Record anomaly evidence as magnetic_anomaly_count and tag events with anomaly flags.
Boundary: do not promise “perfect prevention”; provide measurable confidence and a forensic trail.
Minimal trust primitives (node-side)
Monotonic event_seq: increments per event. Detect missing/injected events with seq_gap_count.
Timestamp: allows time-order evidence even if communication is intermittent.
Credibility discriminator: if the node sees many transitions but few qualified events,
treat it as mechanical bounce / noise. If event_seq gaps appear or signatures fail,
treat it as integrity risk. Always keep the evidence at node level.
H2-9. EMC/ESD/EFT & Cabling Reality (False Triggers in Outdoor/Long-Cable Builds)
Many reed “false opens/closes” are not sensor defects. Long cables collect ESD/EFT energy and convert it into
input threshold crossings (extra edges, extra IRQ wakes, sometimes resets). This section focuses on
node-level suppression choices and verification points that produce measurable evidence.
Failure mechanisms that actually create false events
Common-mode injection on long cabling shifts the input reference and produces short spikes.
Capacitive pickup turns fast transients into a visible pulse at the input node.
Threshold sensitivity: a “clean” spike becomes a logic transition when hysteresis/RC is insufficient.
After-event instability: ESD/EFT can cause latch-up-like behavior or brownout resets (node-level evidence only).
Front-end hardening (with the real tradeoffs)
RC filter: attenuates narrow spikes; too large can add delay and create slow-edge re-crossings.
Hysteresis / Schmitt: reduces multi-crossing; must match expected cable noise amplitude.
TVS/ESD device: clamps peaks, but leakage can bias high-impedance inputs and Cj can reshape pulses.
Return path awareness: provide a short, predictable path for transient current so it does not flow through the threshold reference.
What to verify (actionable, node-side)
EFT injection: measure false_event_rate and irq_count_delta under defined pulse bursts.
ESD touch points: record reset_reason_hist_delta; watch for correlated false events immediately after hits.
Waveform proof: confirm the transient actually crosses the input threshold (before and after suppression).
false_event_rateirq_count_deltareset_reason_hist
Discriminator (avoid guessing)
If irq_count rises but qualified event_count does not, qualification is working; the entry point is still noisy.
If event_count rises, threshold/hysteresis/RC or leakage bias is insufficient.
If resets correlate with EFT/ESD, investigate droop paths and clamp-induced leakage/capacitance effects.
Figure F6. A practical model for long-cable false triggers: interference couples into the cable,
becomes an input spike, crosses threshold, and manifests as IRQ wakes, false events, or resets.
Suppression choices (RC, hysteresis, TVS with leakage/Cj awareness) must be verified with counters.
Cite this figure:Figure F6— “Cable coupling path and suppression points that map to node evidence counters.”
H2-10. Validation Plan (Quantify Reliability with Tolerances, Statistics & Criteria)
A copyable matrix: variables × stress × pass/fail
“Reliable” must be measurable. Validation for reed/contact nodes is a matrix across
geometry, environment, power, cabling, and immunity.
This chapter provides a repeatable structure: what to sweep, how to log, and how to judge false/miss rates.
This SOP is node-side only: contact → threshold/debounce → event → counters/logs.
Each symptom starts with two minimum measurements (waveform/current/counters), then uses discriminators
to split the root cause into Mechanical, Electrical/EMC, Firmware, or RF/Wireless.
Always capture these two first
M1: Input waveform at the pin (does it cross threshold, how many times, spike width/shape?)
M2: One counter pair: irq_count_delta vs event_count_delta (same time window)
If wireless: add tx_retry_count + rssi_hist. If battery-powered: add battery_mv_min + reset_reason_hist.
Minimum tool set (field-friendly)
Scope probe (input pin + ground reference)
Current logger / DMM (sleep uA + TX spike mA)
Serial log / local dump (counters + reset_reason)
Symptom 1
Intermittent false alarms (random open/close events with no real motion)
Evidence (first 2 measurements)
E1: Compare irq_count_delta vs event_count_delta in the same window.
E2: Scope the input pin: look for spike-driven threshold crossings vs contact bounce.
Isolate (discriminators)
Mechanical: bounce-like multi-edges during real actuation; bounce_count rises.
Electrical/EMC: spikes occur without actuation; often cable-length / touch / switching correlated.
Firmware: irq rises but events do not (qualification rejects); or events rise due to weak thresholds.
Raise noise immunity at the threshold: add/upgrade a Schmitt buffer.
MPN examples: SN74LVC1G17 (TI, single Schmitt buffer), 74LVC1G17GW (Nexperia).
Clamp fast transients carefully (avoid leakage bias on high-impedance nodes):
MPN examples: PESD5V0S1UL (Nexperia, low-leakage ESD), ESD9B5.0ST5G (onsemi).
Harden the cable entry with series impedance / ferrite:
MPN examples: BLM18AG102SN1 (Murata ferrite bead), ACM2012-900-2P (TDK common-mode choke, if using twisted pair).
Symptom 2
Door is closed but still shows “open” (persistent wrong state)
Evidence (first 2 measurements)
E1: Measure gap/offset/angle relative to the trigger boundary (is it at a cliff?).
E2: Observe static input level vs threshold over time (humidity/temperature drift indicates leakage bias).
Isolate (discriminators)
Mechanical boundary: tiny door movement flips state; sensitivity is too close to edge.
Leakage bias: input sits near threshold and drifts with temp/humidity; ESD device Cj/leakage is suspect.
Wiring: only one door/loop affected; resistive window/connection quality differs.
Fix (first actions + concrete MPN examples)
Increase mechanical margin: reposition magnet/switch so “closed” is far from the switching cliff (leave geometry headroom).
Stabilize threshold with a comparator + hysteresis (when a raw MCU GPIO is too sensitive):
MPN examples: TLV3691 (TI, nano-power comparator), MCP6541 (Microchip comparator).
Replace high-leakage protection parts on the input (bias problems look like “stuck open/closed”):
MPN examples: PESD5V0S1UL (Nexperia), TPD1E10B06 (TI ESD diode).
Symptom 3
Battery life far below expectation (months instead of years)
Evidence (first 2 measurements)
E1: Measure sleep current (uA-level). Confirm it stays low over minutes, not only instantly.
Leakage dominant: sleep current is high; suspect ESD leakage, LDO IQ, divider resistors, floating GPIO.
Event storm dominant: sleep is low, but wake_count is high; input noise or mechanical bounce.
Retry storm dominant: tx_retry_count high + RSSI low; each event costs multiple transmissions.
Fix (first actions + concrete MPN examples)
Cut baseline power with a true low-IQ regulator (or bypass if topology allows):
MPN examples: TPS782 (TI low-IQ LDO family), MCP1700 (Microchip low-IQ LDO).
Stop droop-driven resets using a supervisor / brownout strategy and (if needed) a load switch:
MPN examples: TPS3839 (TI voltage supervisor), TPS22910A (TI load switch).
Protect the battery against fault/short paths with an eFuse / power switch (node-side only):
MPN examples: TPS25940 (TI eFuse), TPS2553 (TI current-limited switch).
Symptom 4
Only one long cable / one door fails (others are stable)
Evidence (first 2 measurements)
E1: Compare the “bad” line vs a “good” line: cable_len, routing class, shielding, near-noise sources.
E2: Under the same environment, compare false_event_rate and input spike amplitude/width.
Isolate (discriminators)
Follows the cable when swapped → coupling/entry path issue.
Follows the node when swapped → front-end margin/leakage/threshold issue.
Only after certain actions (relay switch nearby, motor start) → conducted/EMI coupling.
Use a resistor network for consistent thresholds (reduces tolerance drift across builds):
MPN examples: YC164-JR-07 series (Yageo resistor network family), EXB-38V series (Panasonic array family).
Re-evaluate TVS choice for leakage/Cj at this node (swap and compare with same test_id):
MPN examples: PESD5V0S1UL (Nexperia), TPD1E10B06 (TI).
Symptom 5
After ESD / thunderstorm, it starts toggling (unstable behavior appears after stress)
Evidence (first 2 measurements)
E1: Check reset_reason_hist_delta (did resets start clustering after the event?).
E2: Check static input bias and leakage symptoms (level drift, new sensitivity, slow re-crossings).
No resets but toggles → threshold/hysteresis margin degraded or leakage increased after stress.
Only long lines affected → coupling path stronger than design margin.
Fix (first actions + concrete MPN examples)
Replace / upgrade the ESD device on the input and re-run the same EFT/ESD test:
MPN examples: PESD5V0S1UL (Nexperia), ESD9B5.0ST5G (onsemi), TPD1E10B06 (TI).
Improve rail robustness using a supervisor and a controlled power path:
MPN examples: TPS3839 (TI supervisor), TPS22910A (TI load switch), TPS25940 (TI eFuse).
Strengthen thresholding (if GPIO-only is too fragile):
MPN examples: TLV3691 (TI nano-power comparator), SN74LVC1G17 (TI Schmitt buffer).
Decision tree
Use the tree to avoid guesswork: start from the symptom, capture two minimum evidence items, then isolate into
Mechanical / Electrical-EMC / Firmware / RF-Wireless, and apply a first fix with a measurable re-test.
Figure F8. A field-friendly decision tree that forces evidence first:
capture waveform + counter deltas, then isolate into mechanical/electrical/firmware/RF buckets,
apply one first fix, and re-test using the same counters.
Cite this figure:Figure F8— “Field debug root-cause tree for reed/contact nodes with evidence-driven fixes.”
MPN note
MPNs above are examples to make BOM discussions concrete (availability/ratings/package must match your design).
The key is to validate changes with the same test_id and compare false_rate, miss_rate, irq_count_delta,
and reset_reason_hist_delta.
Reed switch is closed, but it occasionally toggles—bounce or EMI first?
Usually EMI if you see narrow spikes; bounce if you see multiple slow crossings during a real closure. Measure: (1) input-pin waveform vs the Schmitt/comparator threshold, (2) irq_count_delta vs event_count_delta in the same 10–60 s window. First fix: add hysteresis (e.g., SN74LVC1G17) or a series bead (BLM18AG102SN1). See H2-3/H2-9.
Maps: H2-3Maps: H2-9
Will a long debounce time miss fast open/close events? How to pick the time window?
Yes—an overly long debounce window can miss fast open/close pulses. Measure: (1) the shortest legitimate contact-open time in your use case (P99), (2) qualify_reject_count (or equivalent) and miss_rate from validation runs. First fix: use edge qualification or a shorter window plus hysteresis instead of only RC. See H2-3/H2-5.
Maps: H2-3Maps: H2-5
NO or NC for security sensors—what exposes open-wire/short faults faster?
For fail-safe intrusion detection, NC is often preferred because a cut wire tends to look like an alarm/open circuit, but only if your loop supervision is designed for it. Measure: (1) loop resistance windows (normal/alarm/open/short), (2) open/short injection test confusion rate. First fix: implement single/dual EOL windows at the node. See H2-2/H2-4.
Maps: H2-2Maps: H2-4
With long cables and false alarms, add RC first or switch to Schmitt/comparator?
Start with evidence: if spikes are narrow and high, RC may not catch them; if edges are slow and hovering near threshold, RC helps. Measure: (1) spike width vs your RC time constant, (2) false_event_rate under EFT/long-cable conditions. First fix: add Schmitt/comparator hysteresis (TLV3691 class) plus modest RC, then retest. See H2-3/H2-9.
Maps: H2-3Maps: H2-9
How to set EOL resistance windows to separate open/short/alarm/normal reliably?
Pick windows by stacking tolerances: EOL resistors, cable resistance, contact resistance, ADC error, and temperature drift. Measure: (1) ADC code distributions for each state across temp, (2) worst-case min/max loop resistance per state. First fix: widen guard bands and add a ‘fault’ dead-zone between windows to reduce overlap. See H2-4.
Maps: H2-4
Battery drains fast—check LDO IQ first or a wake-up storm? What two evidences?
Separate baseline leakage from wake storms. Measure: (1) true sleep current over minutes (uA-level), (2) wake_count + tx_retry_count (or irq_count) per hour. If sleep is high, suspect LDO IQ/leaky ESD (PESD5V0S1UL class); if wakes are high, throttle events and cap retries. See H2-6/H2-7.
Maps: H2-6Maps: H2-7
More misses at low temperature—magnet gap issue or battery internal resistance causing threshold drift?
Low-temperature misses can be geometry margin or supply droop. Measure: (1) gap/offset at cold soak vs the trigger boundary, (2) battery_mv_min during TX/wake and the input threshold stability. First fix: move the magnet to increase margin first; then harden power with a low-IQ LDO (MCP1700 class) and brownout logging. See H2-2/H2-6.
Maps: H2-2Maps: H2-6
How to detect “strong-magnet spoofing” on the node side with lightweight evidence?
At node level, you can’t ‘prove intent’, but you can flag anomalies. Measure: (1) magnetic anomaly counter (sustained closed state with abnormal edge statistics), (2) tamper switch events and monotonic event_seq continuity. First fix: add a cover-tamper switch and sign events with a monotonically increasing counter stored in NVM/SE. See H2-8.
Maps: H2-8
Only one door false-triggers—installation offset or cable coupling? How to tell in one pass?
Use a swap test to isolate fast. Measure: (1) swap the cable or node and see if the failure follows, (2) compare (gap/angle) stats vs input spike stats on the bad door. First fix: if it follows the cable, add entry impedance and low-leakage ESD; if it follows geometry, re-mount for margin. See H2-2/H2-9/H2-11.
Maps: H2-2Maps: H2-9Maps: H2-11
After ESD it starts toggling—is the input clamp damaged or MCU reset/config?
After ESD, decide ‘reset’ vs ‘threshold drift’. Measure: (1) reset_reason_hist_delta and battery_mv_min, (2) post-event static input bias (near-threshold drift implies leakage damage). First fix: replace/upgrade the input ESD diode and add hysteresis; if resets cluster, add a supervisor (TPS3839 class) and improve hold-up. See H2-9/H2-11.
Maps: H2-9Maps: H2-11
Same hardware but different wireless protocol/report strategy gives very different battery life—why?
Protocol/strategy changes airtime and retries, which can dominate average current. Measure: (1) tx_retry_count distribution + rssi_hist, (2) duty_cycle and energy per event (current waveform integration). First fix: cap retries, add exponential backoff, and batch non-urgent reports; validate with the same counters before/after. See H2-7/H2-6.
Maps: H2-7Maps: H2-6
How to run a fast production test to baseline magnet gap margin and false-alarm rate?
For production, test margin and false/miss rates with a minimal matrix. Measure: (1) gap sweep (e.g., 3–5 points) at nominal temp, (2) false_rate/miss_rate with a standard jig motion profile plus an EFT spot check for long-wire SKUs. First fix: define pass/fail windows and record test_id, gap_mm, false_rate, miss_rate. See H2-10.
