An optical-scattering dust monitor becomes reliable only when the system is designed as an evidence chain: stable light source → low-noise AFE → temperature-aware calibration → edge filtering that preserves truth → wireless reporting that is traceable.

Definition of “done” (mechanically verifiable evidence fields)

• Baseline is stable and explainable: both dark baseline (light off/blocked) and ambient baseline (light on with scatter removed or equivalent) remain bounded, and their drift over temperature is modeled instead of hand-waved.
• Response is repeatable: repeated exposures to a reference aerosol (or a controlled surrogate standard) produce consistent outputs across days and across temperature—without hidden saturation.
• Telemetry is sufficient to reproduce any reported number: raw photodetector current (or proxy), TIA output, light-source current (set & measured), temperature, optional fan/flow status, plus a fault bitmask and the current gain/baseline state.

In optical-scattering designs, specification defines architecture. If targets are vague (“measure PM2.5”), the design will drift into untestable assumptions. This chapter converts PM targets into hard constraints on dynamic range, time response, and ambient rejection—each tied to measurable evidence.

2.1 Decide the outward-facing output (what the product promises)

• Mass bins: PM1 / PM2.5 / PM10 as reported values (what users will trend and compare).
• Event vs baseline behavior: whether short “dust bursts” must be preserved as events or smoothed into a stable trend.
• Health / self-diagnosis: whether the device must expose sensor health (contamination suspicion, saturation frequency, light-source stability).

2.2 Dynamic range: treat saturation as a state, not a surprise

• Low end (clean air): noise floor must be below the smallest meaningful change in background; otherwise PM1/PM2.5 becomes dominated by baseline jitter.
• High end (heavy dust): define what happens when the signal grows beyond headroom: auto-ranging (gain steps), integration control (sampling strategy), or flagged clipping (explicit saturation reporting).
• Non-negotiable evidence: saturation must set a loggable flag and preserve the gain state so later analysis can separate “true dust” from “measurement ceiling.”

2.3 Time response: reporting latency is an engineering choice

• 1–3 s class: reacts quickly to bursts but is sensitive to EMI, wireless TX artifacts, and ADC/reference perturbations.
• 5–10 s class: produces a calm trend line, but can hide short dust events unless a parallel fast channel exists.
• Design implication (evidence-based): the chosen window must be visible in telemetry (window length, filter mode), so “firmware changes” cannot silently change product behavior.

2.4 Environmental constraints: map each stressor to a mandatory mitigation

• Temperature range: requires temperature-aware calibration (coefficients or table) and explicit baseline rules (freeze vs adapt).
• Ambient light: requires either modulation/synchronous sampling or a two-baseline strategy (dark + ambient) with documented recovery behavior.
• Humidity / contamination exposure: requires drift detection and maintenance triggers (baseline slope, saturation frequency, light-source efficiency proxy).
• Vibration / electrical noise: requires a coupling test plan (wireless TX on/off, supply ripple injection) with recorded deltas.

2.5 Hard mapping: targets → architecture → validation items

• Dynamic range → TIA gain strategy + ADC resolution + gain/saturation states logged.
• Time response → sampling rate + filtering window(s) + window mode visible in telemetry.
• Ambient light robustness → modulation/synchronous sampling OR baseline subtraction + sunlight/flicker stress test results recorded.

Optical-scattering monitors do not “measure particles directly.” They measure time-varying scattered light, translate it into a detector current, and then convert that electrical signal into features and counts that can be mapped to PM bins under a specific calibration version. This chapter defines where the signal is born, where it is corrupted, and which evidence fields keep the design testable.

3.1 From photons to electrical current: the three-component model

• Baseline component: detector dark current + light leakage + slow optical/aging drift. This component must be bounded and modeled, not “filtered away.”
• Event component: particle-driven fluctuations (pulses/variance changes). This component must survive filtering windows without being mistaken for drift.
• Corruption component: EMI pickup, rail ripple, wireless TX coupling, and sampling/alias artifacts. This component must be detected and labeled in logs.

3.2 Noise injection points mapped to evidence fields

• Ambient light injection: track ambient baseline and recovery time after step changes.
• Laser intensity drift: correlate output drift with light current (set/measured) and temperature.
• Detector dark current drift: maintain a dark-baseline model vs temperature.
• TIA noise / headroom limits: estimate noise floor and expose saturation + gain range states.
• EMI pickup: run TX on/off A/B tests and record deltas, not anecdotes.

In scattering systems, the light source is part of the measurement instrument. Stability is achieved by choosing an appropriate drive mode (CW or modulated), enforcing constant-current behavior across temperature and aging, and exposing telemetry and fault states that make field behavior traceable.

4.1 Drive modes: CW vs modulated (selection rule, not preference)

• CW drive: simplest implementation, but more susceptible to ambient-light drift and slow baseline shifts.
• Modulated/pulsed drive: enables synchronous sampling and improves ambient rejection; requires timing alignment between modulation and ADC windows.
• Engineering rule: if operation includes strong ambient changes (sunlight, flickering indoor lighting) or low-concentration stability is required, prefer modulated drive with explicit sampling windows and logged mode states.

4.2 Constant-current driver requirements (must be defined and measured)

• Compliance headroom: insufficient headroom forces non-linear current behavior and amplifies temperature/aging effects. Evidence comes from I_set vs I_meas tracking at temperature extremes.
• Edge behavior (rise/fall, overshoot): edges that are too sharp inject EMI into the AFE; edges that are too slow reduce usable synchronous windows. Evidence comes from “light toggle” tests and TIA-out impulse signatures.
• Current monitor path: a measurable I_meas is mandatory. Without it, drift cannot be separated into optical vs electrical causes.

4.3 Safety and shutdown (system-level, traceable)

• Interlocks: lid/open-path, over-temperature, timeout. These must drive a hardware-visible shutdown.
• Fault shutdown: open/short detection and thermal faults must be latched or timestamped for field diagnosis.
• Current limit: a hard upper bound prevents software misconfiguration from damaging the light source.

4.4 Evidence captures that prevent “mystery drift”

• Optical power proxy vs temperature: use monitor photodiode or closed-loop proxy to show stability across temperature, not just current stability.
• Modulation vs sampling plan: define modulation frequency, ADC sampling rate, and window length as a coherent plan to avoid aliasing artifacts. Keep the plan visible in telemetry (Drive_mode, Mod_freq, Window_len).

The analog front-end (AFE) turns detector current into digital features while preserving evidence: low noise at clean-air baseline, headroom during dust bursts, and traceable gain/offset calibration. AFE design is successful only when noise floor, saturation behavior, and state changes are observable and repeatable.

5.1 Detector choice tradeoffs: photodiode vs SiPM

• Photodiode (PD): predictable linear behavior and simpler biasing; performance depends heavily on TIA noise and leakage control.
• SiPM: effective gain improves weak-signal visibility; requires bias management and tighter temperature control due to stronger temperature-dependent behavior.
• Selection rule: PD favors stable baselines and straightforward calibration; SiPM favors very low-light detection but demands more bias/temperature evidence fields.

5.2 TIA architecture: gain strategy, bandwidth, and stability margin (conceptual)

• Gain strategy: single high gain improves clean-air resolution but risks clipping; dual-gain or auto-ranging preserves headroom while keeping low-end sensitivity.
• Bandwidth strategy: bandwidth must align with the modulation/sampling plan (see H2-6). Excess bandwidth amplifies EMI and switching edges without adding measurement value.
• Stability margin concept: insufficient margin appears as ringing or “false pulses,” especially when temperature shifts component values. Verification uses light-toggle step tests and impulse signatures at TIA output.

5.3 Input protection and leakage control

• Protection must be invisible in-band: input clamps and ESD paths must not introduce measurable leakage or temperature-dependent offsets that resemble dust.
• Placement matters: protection and guard paths should prevent transient damage while avoiding extra parasitics at the summing node.

5.4 Saturation management: make clipping a state, not a surprise

• Dual-gain path: low-gain channel preserves bursts; high-gain channel resolves clean-air baseline.
• Auto-ranging: change gain with hysteresis; log the gain range state and the reason for switching.
• Clamp strategy: if clipping occurs, it must raise a saturation flag and preserve context (gain state + headroom margin).

5.5 Analog filtering targets (scoped)

• Reject supply ripple and switching edges: prevent rail-coupled artifacts from entering the ADC.
• Suppress EMI pickup: reduce high-frequency bursts that correlate with wireless TX and digital transitions.
• Do not “erase” baseline reality: slow drift is handled by baseline strategy (H2-6). Analog filtering should prevent corruption, not hide evidence.

Most field drift is caused by ambient/light-leak combined with baseline tracking failure. Robust designs treat ambient light as a measurable input and use synchronous sampling plus a baseline estimator to separate net scattering from ambient conditions.

6.1 Product-class methods (choose and document the assumption)

• Optical/mechanical shielding: treated as a constraint and leakage limit, not a guarantee. Designs still require electrical observability of ambient conditions.
• Electrical rejection: modulation (pulsed light) paired with correlated/synchronous sampling to remove ambient components.
• Baseline estimator: maintain dark and ambient samples and apply subtraction rules with explicit freeze/recovery states.

6.2 Baseline estimator rules (prevent event corruption)

• Dark sample: capture when light is off (or path blocked) to estimate detector/AFE baseline.
• Ambient sample: capture ambient-only component (light off) and track its changes over time.
• Subtraction path: compute net scattering as (ON-window − OFF-window) under a known timing plan. Baseline updates must freeze during particle events to avoid “learning the dust.”
• Recovery behavior: after a light-step (sunlight / switching on room lights), recovery time must be bounded and logged.

6.3 Validation checklist (stress tests that reveal real drift sources)

• Sunlight stress: different angles/intensity; verify ambient baseline tracking and recovery time bounds.
• Flicker stress: indoor flicker and dimmers; verify no alias patterns appear in net scattering output.
• Light on/off step response: measure step response and recovery curve; confirm baseline-state transitions are logged.

Drift becomes manageable only when it is classified, measured, and constrained by guardrails. Practical designs combine temperature-aware calibration with slow baseline adaptation and strict rules that prevent real dust events from being absorbed as “drift.”

7.1 Drift sources (symptoms mapped to evidence fields)

• Light-source efficiency drift: long-term baseline slope. Compare output trends to Light_I_set/Light_I_meas and temperature.
• Detector dark current vs temperature: baseline changes persist even with light off. Track Dark_baseline vs Temp.
• TIA offset / leakage drift: low-end bias shifts and temperature sensitivity. Track baseline offset and stability over a temperature sweep.
• ADC reference / gain drift: apparent scaling across the range. Track ADC_ref_proxy and calibration gain/offset versions.
• Contamination (optical path): reduced responsiveness and longer recovery after light steps. Track Recovery_time and contamination heuristics.

7.2 Compensation layers (stacked, versioned)

• Temperature sensing placement & sampling: measure temperatures that correlate with drift sources (light source, detector/AFE region). Log both raw and filtered temperature to avoid time-constant confusion.
• Per-temperature calibration: use coefficients or LUT tables to correct gain/offset over temperature. Store as versioned assets (Cal_version, Temp_coeff/LUT_id).
• Slow baseline adaptation: allow bounded long-term adjustments with a maximum drift rate and a rollback checkpoint.

7.3 Guardrails: “Do not chase dust as drift”

• Freeze during events: when event detection triggers, baseline updates must stop (Baseline_state = FREEZE).
• Recovery mode for light steps: allow controlled recovery after ambient steps (Baseline_state = RECOVERY) while preventing dust misclassification.
• Rate limits: baseline slope per hour must be bounded to avoid runaway adaptation.
• Traceable reasons: baseline state transitions must carry a reason code (event, light step, stable, fault).

7.4 Chamber-test evidence package (minimum acceptance)

• Baseline vs temperature plots: up/down sweep with hysteresis observation; compare “before vs after compensation.”
• Temp coefficient logs: coefficients or LUT version attached to each firmware build and telemetry packet.
• Residual drift metric: bounded drift after compensation under stable dust conditions.

Edge processing succeeds when the digital pipeline is reproducible: a telemetry packet can replay the same derived metrics. This chapter defines ADC requirements, a scoped DSP pipeline, mapping dependencies, and fault detection that turns “odd behavior” into traceable flags.

8.1 ADC requirements (beyond resolution)

• Resolution (effective bits): sets the clean-air quantization floor and low-end sensitivity.
• Sampling rate: must align with modulation and windowing (ON/OFF) to avoid alias patterns.
• Reference stability: drifting Vref creates “scale drift” that looks like dust. Track a reference proxy where possible.

8.2 Scoped DSP pipeline (replayable steps)

• Pre-clean: remove clipped samples and impossible codes; tag them for diagnostics.
• Smoothing: short-window smoothing for stability without large latency.
• Outlier rejection: suppress spikes correlated with EMI or wireless TX.
• Event detection: detect bursts using variance/peaks and emit Event_flag and event counters.
• Rolling windows: compute 1–10 s windows and longer trends for reporting.

8.3 Mapping: counts → concentration (explicit model dependence)

• Concentration mapping depends on the calibration aerosol/model and optical geometry. Therefore mapping must be versioned and attached to telemetry.
• Store and report Mapping_model_id and Cal_version so metrics remain comparable across builds and batches.

8.4 Fault detection (bitmask, not anecdotes)

• Saturation: ADC/TIA clipping detected and flagged with context (gain range and headroom).
• Stuck-at baseline: baseline does not move and noise becomes abnormal—indicates disconnection or algorithmic lockup.
• Light-source failure: mismatch between I_set and I_meas, or monitor proxy not responding to control changes.
• Contamination heuristic: responsiveness decreases and recovery time increases over time.

A dust monitor becomes deployable when sampling and diagnostics remain deterministic under wireless activity, and when every field unit reports enough context to explain its readings. This chapter focuses on node partitioning, power states, OTA hooks, and interface-level integrity—without RF/antenna design details.

9.1 Node partitioning templates (choose by measurement sensitivity)

• Split architecture (sensing MCU + wireless SoC): sensing stays deterministic; wireless stack bursts are isolated from the sampling timeline.
• Combo architecture (single SoC): requires strict scheduling: sampling windows must preempt wireless tasks, and TX bursts must be rate-limited around measurement windows.

9.2 Scheduling and data path (deterministic sampling, buffered reporting)

• Sampling as a real-time task: ON/OFF windows and baseline state transitions must stay phase-stable.
• Wireless as a background task: batch reporting from a ring buffer reduces frequent TX bursts that correlate with measurement noise.
• TX-aware guard time: optional policy: avoid TX during critical sampling windows and log TX activity when unavoidable.

9.3 OTA hooks (update strategy with rollback behavior)

• Versioned assets: firmware version and calibration/mapping versions must be reported together.
• Rollback point: update failures should revert to a known-good image and preserve calibration constants.
• Update window policy: schedule updates outside critical sampling periods to avoid baseline disturbance.

9.4 Interface-level integrity (basic, practical)

• Monotonic counter: detects replay and duplicates.
• Packet integrity check: detect corruption and drop invalid payloads.
• Version attachment: each packet carries FW/Cal/Model versions for traceability.

9.5 Field diagnostics (minimum remote triage capability)

• Last fault: fault code and timestamp (or sample index).
• Sensor health score: derived from noise floor, saturation rate, recovery time, and contamination heuristics.
• Calibration version: ensures cross-node comparability and correct model interpretation.

Dust monitors often fail silently: the device stays “alive,” but power noise and EMI corrupt micro-amp measurement paths. Robust designs treat power/EMI as measurable inputs: define domains, identify coupling paths, and prove immunity using A/B tests such as wireless TX on/off impact on baseline and noise floor.

10.1 Rail planning (domains and reference cleanliness)

• Separate domains: keep analog sensing rails isolated from digital/wireless rails where possible.
• Reference stability: ADC reference cleanliness is often more important than bulk supply regulation.
• Ground boundaries: define where noisy return currents flow so TIA and Vref nodes are not modulated by system activity.

10.2 Coupling paths (what injects false dust)

• Laser switching edges → rail/ground bounce → TIA corruption: residual artifacts can survive ON/OFF subtraction if windows are misaligned.
• Wireless TX bursts → Vref/ADC disturbance: raises the noise floor and can create periodic patterns in net_scatter.
• Digital clocks → electric/magnetic coupling: introduces narrowband artifacts that appear as false events.

10.3 Robustness strategy (separation + filtering + time-slotting)

• Separation: keep the measurement chain’s reference and summing nodes protected from noisy domains.
• Filtering: limit high-frequency junk entering ADC inputs and references.
• Time-slotting: avoid measuring during high-interference activities (TX bursts, high di/dt edges) and log activity when overlap occurs.

10.4 ESD/surge philosophy for exposed ports (conceptual)

• Energy termination at the boundary: clamp and route energy at the port entrance, not through the sensing area.
• Leakage discipline: protection must not create temperature-dependent offsets at the TIA input.
• Post-event detectability: after ESD, self-tests and health scores must reveal silent degradation.

10.5 Evidence fields and field tests (minimum acceptance)

• Wireless TX on/off A/B test: compare noise floor, baseline slope, and false event rate under fixed conditions.
• Baseline disturbance observations: tag measurement artifacts when rails or system states change.
• Recovery-time checks: verify bounded recovery after interference steps.

This chapter turns “it seems to work” into a manufacturing and field method. The objective is a closed evidence chain: calibration artifacts are versioned, production tests are fixture-reproducible, and field validation can trigger maintenance or recalibration.