H2-1 — Scope, System Boundary, and What This Page Solves

Central idea (1–2 sentences): A smart thermostat is a coupled hardware system—environmental sensing, low-power compute, Wi-Fi/Thread radio, HVAC 24VAC I/O, and power robustness—where failures propagate across domains. This page focuses on measurable evidence (rails, waveforms, logs, and pass/fail criteria) to validate designs, debug field issues, and choose ICs confidently.

What is inside the boundary (hardware domains)

• Sensing domain: T/H/env sensors + AFE/ADC + layout/thermal reality that controls accuracy, drift, and condensation sensitivity.
• Compute domain: low-power MCU/SoC, RTC, wake sources, reset reasons, and diagnostic hooks that make failures observable.
• Radio domain: Wi-Fi/Thread hardware behavior tied to RF power integrity, antenna/layout constraints, and coexistence noise.
• HVAC I/O domain: 24VAC sensing and relay/triac/SSR drive, with dv/dt immunity and false-trigger control.
• Power domain: 24VAC front-end, rectified bus ripple/valleys, inrush, UVLO/BOR margins, sequencing, and brownout behavior.

What is outside the boundary (intentionally excluded)

This page does not expand into cloud/app platform architecture, automation rules engines, or protocol-stack deep dives. When platform-side symptoms exist (e.g., “disconnects”), the only goal here is to prove whether the root cause is hardware (supply droop, resets, RF rail noise, or I/O transients) using measurable evidence.

What will be gained (3 deliverables)

• A system map that is testable: a block diagram with labeled rails, critical I/Os, and test points (TPs) so measurements stay consistent.
• A diagnosis path that is repeatable: symptom → minimum evidence set → likely root-cause domain → hardware fix knobs.
• A validation + selection checklist: red-line parameters and tests that reveal field failures early (brownouts, false triggers, drift, and RF sensitivity).

H2-2 — Hardware Architecture Overview (Figure F1)

Purpose: The architecture must be readable as a test plan. Four chains are tied together in one diagram—sensing, power, radio, and HVAC I/O—with labeled rails and test points so measurements remain consistent across validation and field debug.

Critical rails and I/O (what must be visible)

• Rails: rectified bus (VRECT), main digital rail (e.g., 3V3), RF rail, sensor/reference rail (if separate).
• I/O: 24VAC sense inputs, relay/triac/SSR drive outputs, sensor buses (I²C/SPI/ADC), optional diagnostics header.

Test points (TP) — minimum evidence set

• TP1 (VRECT): rectified bus ripple/valley and wiring transient capture.
• TP2 (3V3): regulator droop prior to UVLO/BOR events.
• TP3 (MCU VDD): confirmation that resets correlate to supply behavior.
• TP4 (RF rail): TX peak droop/noise and coexistence sensitivity.
• TP5 (24VAC sense): false detection, threshold chatter, and common-mode noise evidence.
• TP6 (Drive node): relay coil current or triac gate/reference timing to prove false triggers vs control logic.

Failure propagation map (three repeatable chains)

• Chain A — energy deficit → reset → reconnect → output risk: 24VAC switching or wiring transient deepens VRECT valleys (TP1) → DC rail droop (TP2/TP3) → reset reason indicates BOR/UVLO → Wi-Fi/Thread rejoin bursts → HVAC output timing becomes uncertain during recovery.
• Chain B — HVAC dv/dt → ground/reference shift → sensing/I/O misread: relay/triac transitions create common-mode noise at sense pins (TP5) → ADC/I²C artifacts or threshold chatter → incorrect “call” interpretation or unstable control decisions.
• Chain C — RF burst without reset → link instability: RF TX peaks pull down RF rail (TP4) while MCU rail stays stable (TP3) → RSSI/packet retries rise → perceived “disconnects” occur without BOR evidence, pointing to RF power integrity/coexistence rather than firmware logic.

H2-3 — Environmental Sensing Chain: T/H/Env AFEs and Layout Reality

Goal: “Accuracy” is not a single spec. It is the sum of sensor tolerance, AFE/ADC noise, layout & thermal coupling, and sampling/filters. The sensing chain must be written as a measurable pipeline so field complaints can be mapped to evidence rather than assumptions.

Break down accuracy into four measurable buckets

• Sensor (device-level): initial tolerance, long-term drift, humidity saturation behavior, and response time.
• Interface (AFE/ADC + reference): ADC quantization, noise density, reference stability, input bias/leakage, and conversion cadence.
• Physics (layout/thermal/airflow): self-heating paths, thermal gradient across the PCB, enclosure airflow, and wall-mount “cold bridge” effects.
• Sampling strategy: raw sample cadence, digital filtering, outlier handling, and event-aligned logging.

Minimum evidence set (what to capture first)

Raw counts vs filtered value (why both matter)

Raw counts answer “is the sensing chain stable?” while filtered values answer “does the user view look stable?”. When only filtered values are logged, a filter can mask a real instability (spikes, bus retries, reference noise) until the unit is installed and exposed to HVAC switching and real airflow.

ADC/AFE signatures that distinguish root causes

• White-noise signature: mean stable, variance increases → typical of analog noise injection or reference instability.
• Spike signature: rare large outliers aligned to events (HVAC switching, RF bursts) → typical of coupling or transient ground/reference shifts.
• Step/plateau signature: discrete steps or plateaus → typical of quantization, insufficient resolution, or conversion cadence interacting with filtering.

Layout/thermal reality (why wall-mount differs from bench)

Wall mounting changes heat paths and airflow. A nearby regulator, RF PA, or copper plane can introduce a repeatable thermal gradient that is invisible on the bench. The key proof is correlation: reading error moves predictably with heat-source state or enclosure airflow changes, not randomly.

Error budget table (useful for priority decisions)

H2-4 — Calibration, Drift, and Condensation: Making Readings Trustworthy

Goal: Factory-accurate readings often degrade after wall-mount because the installation changes airflow, heat paths, and condensation risk. Calibration must be treated as an evidence-gated process—only applied when conditions are stable and the sensing chain is trustworthy.

Three-layer calibration model (hardware-friendly)

• Factory baseline: establish a predictable offset/slope behavior using one-point or two-point references.
• In-field correction: small adjustments only in stable windows; never “chase noise”.
• Guardrails: block updates when condensation, bus instability, power droop, or heavy HVAC switching is present.

Two-point method and temperature compensation (method, not math)

A two-point approach creates a stable baseline across unit-to-unit variation. The important part is not the formula; it is the evidence that the correction is valid: residual error must shrink and remain stable across a short observation window. If residual error does not converge, the root cause is likely thermal coupling, noise injection, or condensation—not a missing coefficient.

Condensation: field evidence that explains “RH looks wrong”

Condensation creates a distinctive pattern: humidity readings stick near saturation, recover slowly, and correlate to PCB temperature gradients. The proof is time behavior (saturation and recovery time) plus thermal evidence (local board temperature and enclosure conditions), not a single snapshot value.

Calibration triggers (time / temp / event) with stability gates

• Time-based: run on a schedule only when recent variance and outlier rate are low.
• Temp-based: trigger after crossing a temperature band, only if bus errors stay below a threshold window.
• Event-based: after install or after a long power interruption, only once rails and readings are stable.

When calibration must be blocked (do not update)

• Condensation suspicion: RH near saturation with abnormally long recovery.
• Bus instability: I²C/SPI error counters rising in the last observation window.
• Power droop: brownout/reset indicators or rail valleys during measurement windows.
• Switching turbulence: frequent HVAC switching producing event-aligned spikes in raw counts.
• Thermal gradient shifts: sensor area temperature changes with nearby heat-source state (buck/RF TX).

H2-5 — HVAC Drive I/O: Relays, Triacs/SSR, and 24VAC Sense Without False Triggers

Objective: HVAC I/O must survive long wiring, real 24VAC waveforms, and switching transients while avoiding false “call” detection and unintended output conduction. This chapter treats the I/O chain as an evidence-driven system: every claim maps to a waveform alignment, a current signature, or a timing relationship.

Three false-trigger mechanisms (map symptoms to evidence)

• Input mis-detection (24VAC sense): threshold chatter or transient injection makes “call” appear/disappear. Evidence: 24VAC vs sense-node aligned in time.
• Switch conduction without command (triac/SSR): dv/dt-induced triggering or leakage paths create short unintended on-time. Evidence: MT1/MT2 vs gate alignment shows conduction begins without valid drive.
• Mechanical chain instability (relay): coil undervoltage or bounce causes intermittent contact behavior. Evidence: coil current vs load alignment shows bounce windows and undervoltage correlation.

Evidence captures that settle debates fast

Zero-cross detect and turn-on timing (waveform alignment)

Zero-cross detect (ZCD) is only “correct” when timing stays stable across wiring and load conditions. The validation method is simple: align 24VAC, ZCD output, and drive enable on the same time base. If the relative timing shifts between events, the root cause is usually sense-node noise, reference/ground movement, or transient injection—not missing firmware logic.

Snubber/TVS/RC boundaries (symptom → countermeasure)

• Triac dv/dt false trigger: RC snubber across load reduces dv/dt. Boundary: adds loss/heat; too large C increases leakage effects.
• Sense-node spike injection: series-R + RC filter + TVS limits transient amplitude. Boundary: slows response and shifts timing margins.
• Relay bounce & coil noise: ensure coil voltage margin and proper flyback control. Boundary: flyback choice trades release speed vs EMI.
• Zero-cross chatter: add hysteresis/window filtering at the detector front-end. Boundary: too much filtering can miss valid edges.

Call detection and protection hooks (hardware-first)

• Call detection robustness: design a sense front-end with controlled thresholds, hysteresis, and bandwidth (RC) so noise does not look like a call.
• Open/short detection (minimum approach): use abnormal sense-node statistics (stuck-high/low, missing zero-cross, excessive spikes) to flag wiring faults without deep control algorithms.
• “Protection must be provable”: every protection action should have a measurable trace: clamping evidence, counters, or event markers aligned to waveforms.

H2-6 — Power from HVAC: C-Wire, Power-Stealing, Brownouts, and Rail Sequencing

Objective: “Average power looks fine” can still fail in real installations because the rectified bus is pulsed and the system lives or dies on energy valleys. Brownouts, resets, RF dropouts, and unintended I/O behavior usually follow a repeatable chain: bus valley → rail droop → UVLO/BOR event → recovery burst → secondary symptoms.

Three measurements that explain most field failures

• Rectified bus ripple & valley depth (TP1): peak-to-peak ripple is less important than the lowest valley and its duration.
• Buck peak current + UVLO/BOR threshold (TP2/TP3): peak load events can push the buck into limit or droop that triggers BOR/UVLO.
• Wi-Fi TX burst correlation (TP4): align RF burst windows to rail behavior to separate “reset-driven dropout” from “RF-rail droop dropout”.

Why “power enough” still resets (typical chain)

• Valley formation: 24VAC switching/load attach or power-stealing windows deepen VRECT valleys.
• Conversion stress: buck sees a lower input headroom and higher peak current demand.
• Threshold crossing: UVLO/BOR triggers, or rail droop corrupts RF/baseband operation.
• Recovery side effects: reconnection bursts and re-initialization increase instantaneous demand, repeating the cycle if energy margin is small.

Sequencing and “stability windows” (hardware-verifiable)

Sequencing is not about an ideal order on paper; it is about rails reaching a stable window (not just a threshold) before sensitive actions occur. A verifiable approach is to measure startup and recovery: VRECT (TP1) → 3V3 (TP2) → MCU VDD/BOR status (TP3) → RF rail (TP4). If any rail reaches its stable window late or collapses during peak events, symptoms will appear as intermittent resets, link drops, or I/O anomalies.

Supply scenarios comparison (power vs risk vs validation)

H2-7 — Wi-Fi + Thread Hardware: RF Power Integrity and Coexistence

Objective: Wireless reliability in a thermostat is primarily a hardware problem when the symptoms correlate with switching events, TX bursts, or installation constraints. This section focuses on RF power integrity, noise coupling, and antenna/ground reference—without protocol-stack explanations.

Three hardware-rooted failure patterns (symptom → evidence → first knob)

• PA rail droop during TX: unstable throughput or elevated retries without full MCU resets. Evidence: TX peak current aligned with PA rail droop.
• Noise injection from switching domains: retries spike when HVAC drive or buck operating point changes. Evidence: event-aligned bursts in retry rate with matching rail ripple/noise.
• Antenna/reference issues: performance varies strongly by wall box/metal proximity/orientation. Evidence: installation-dependent RSSI/retry statistics with weak correlation to rail droop.

Evidence captures (what to measure and why it closes the loop)

The “coexistence triad” (hardware-only)

• Power isolation: separate or filter RF supply (LDO/LC) and place decoupling to minimize loop area and shared impedance.
• Frequency & clock avoidance: avoid placing switching frequencies/harmonics on sensitive bands; keep noisy clocks away from antenna reference paths.
• Antenna keep-out & ground reference: enforce keep-out zones, maintain a stable reference ground, and isolate high dv/dt regions.

H2-8 — Low-Power MCU Platform: Always-On Scheduling, RTC, and Hardware Diagnostics Hooks

Objective: Low power and debuggability depend on hardware-visible state transitions. A practical platform can explain every wake-up, every peak current event, and every reset with minimal persistent evidence—without relying on OS or application tuning guidance.

Make low power measurable: current profile + wake causality

• Current profile: sleep (µA) → wake (mA) → peak (TX/drive) → return to sleep. Peak amplitude and “time-to-µA” define energy cost.
• Wake sources: RTC, GPIO, comparator/threshold events, and radio/module interrupts must be explicitly identifiable.
• Stability windows: heavy actions should occur only when rails are stable; brownout-prone windows should not trigger high-demand bursts.

Hardware diagnostics hooks (small set, big leverage)

Always-on domain design intent (hardware-visible)

• Always-on domain: RTC + backup domain retains minimal context while the main domain stays off.
• Main domain gating: power-gate RF and high-noise blocks; only enable them after rails reach stable windows.
• Fail-safe logging: keep persistent writes minimal and robust against brownouts (short records, bounded frequency).

H2-9 — EMC/ESD/Surge Reality in HVAC Wiring and Touch Interfaces

Objective: A thermostat lives on long HVAC wiring, switches inductive loads, and is frequently touched. Field failures are best explained by energy entry paths and hardware-visible signatures, not by abstract compliance checklists.

Field harshness model (what makes thermostats electrically “unfair”)

• Long cable as an energy channel: 24VAC terminals connect to meters of wiring that behave like antennas and common-mode conduits.
• Inductive switching: relays/contactors/valves introduce dv/dt and kickback that can couple into rails and sense inputs.
• User touch: touch edges and bezels become direct ESD entry points that can disturb references or trigger resets.

“Port failure leaderboard” (ports → signatures → first suspicion)

• 24VAC terminals (C/R/W/Y/G…): common signatures include VRECT valley deepening BOR/UVLO resets false trigger risk. First suspicion: clamp/RC boundary, shared ground return, and power-entry margin.
• Touch edge / display interface: common signatures include touch glitch I²C errors WDG/BOR resets. First suspicion: ESD return path crossing sensitive references and insufficient edge protection.
• Debug/USB/service port (if present): common signatures include I/O upset port clamp heating intermittent comms. First suspicion: exposed connector clamping and inadequate ground return segmentation.

Protection stack (from port to system)

• Port layer: clamp the surge/ESD early (TVS/RC/snubber) and control where the return current flows.
• Power-entry layer: ensure the rectified bus and downstream rails keep margin during transients (valley depth vs UVLO/BOR).
• Ground return layer: keep high dv/dt loops closed; route ESD return away from sensitive analog and reference points.
• Spacing / isolation layer: enforce clearance/creepage and isolate noisy partitions from touch and sensor references.
• Sensitive analog zone: last-line filtering and reference hygiene for AFEs/ADCs and touch references.

H2-10 — Validation Test Plan: What to Measure, Fixtures, and Pass/Fail Criteria

Objective: A thermostat is “validated” only when key behaviors are measurable, repeatable, and tied to clear pass/fail criteria. This plan emphasizes fixtures, measurement points, and failure signatures—without becoming a certification tutorial.

Four test clusters that cover real-world acceptance

• Sensing acceptance: accuracy, response time, and drift under realistic airflow and self-heating conditions.
• 24VAC switching & brownout: no unintended triggers and no resets across load changes and transient injections.
• Wireless stress: RSSI/retry windows vs TX peak current and rail droop, including occlusion and distance variations.
• ESD/EFT/Surge: defined stress levels with observable outcomes (reset/false trigger/damage/error spikes).

Fixtures (minimum set + add-ons)

• Minimum set: controllable 24VAC source, repeatable load switching, oscilloscope, current measurement (shunt/probe), and a repeatable airflow/temperature stimulus.
• Add-ons: FFT/spectrum capability, ESD/EFT/surge equipment, RF shielding/occlusion fixtures for repeatability.

Validation matrix (Test / Setup / Measure / Pass-Fail / Signature)

H2-11 — Field Debug Playbook: Symptom → Evidence → Likely Root Cause → Fix

Center idea: Field failures become solvable when each symptom is forced into the same workflow: capture a small, fixed evidence pack (2 waveforms + 2 registers/counters), map signatures to likely root causes, apply a layered hardware fix, then re-run the same captures to verify.

Standard evidence dictionary (reuse across all symptoms)

• Waveforms: TP1 VRECT (rectified bus valley), TP2 3V3, TP4 RF rail, 24VAC terminal waveform, coil current / triac gate timing, sensor raw/ADC noise proxy.
• Registers/counters: reset reason (BOR/UVLO/WDG/POR), brownout counter, Wi-Fi/Thread retry counters + RSSI window stats, I²C error counters (NACK/timeout), GPIO snapshot (HVAC outputs + call inputs), last wake source.

Symptom lookup table (10 common field signatures)

H2-12 — IC Selection Checklist + BOM Examples (by Function Block)

Center idea: Selection should be driven by field failure signatures. Each function block below includes a selection red line, common pitfalls, what to verify, and concrete MPN examples that match typical thermostat constraints (24VAC environment, long wiring, low power, and RF coexistence).