Capnography measures the time-varying CO₂ waveform in the breathing cycle, while capnometry reports numeric CO₂ values derived from that waveform. EtCO₂ is the end-expiratory CO₂ value near the plateau end, so it is only meaningful when the system response, transport delay, and waveform phase structure allow the plateau to be captured without distortion or drift.

• Phase I (baseline): inspired gas / near-zero CO₂ region. Baseline lift usually indicates “zero shift” or true inspired CO₂.
• Phase II (upstroke): transition from dead-space to alveolar gas. A very rounded/slow upstroke often signals bandwidth limits or mixing in the sampling path.
• Phase III (plateau): near-alveolar CO₂. EtCO₂ is typically taken near the end of this plateau; poor plateau definition is a warning that EtCO₂ is not trustworthy.
• Phase IV (inspiratory downstroke): return toward baseline. Downstroke shape can be distorted by delay/response and baseline drift.

Mainstream places the optical cell at the airway for minimal transport delay and crisp phase structure, but it must manage condensation, contamination, and heating at the patient interface. Sidestream moves gas through a sampling line for flexible placement, but it introduces transport delay and physical bandwidth limits from line volume, mixing, and moisture management—these cannot be “recovered” by algorithms.

NDIR capnography does not measure “CO₂ directly.” It measures optical attenuation that depends on CO₂ concentration, optical path length, and pressure/temperature. The transfer curve is nonlinear and drifts with source aging, window contamination, and temperature gradients. The AFE must therefore prioritize stable baseline, reference/ratio handling, and calibration hooks over raw ADC bit count.

• Absorption depends on concentration and path length: more CO₂ or longer path → more attenuation (not a perfect line).
• Pressure and temperature reshape the curve: changing gas density and absorption characteristics shifts the same optical reading to a different CO₂ estimate.
• Electronics and optics add their own “gain drift”: source output, detector sensitivity, and front-end gain drift can mimic real CO₂ change unless normalized.
• Calibration is part of the measurement: the output is only comparable across devices/time when zero/span (and compensation coefficients) are controlled.

• Wider bandwidth increases signal energy but can admit more neighboring absorption and scattering effects (notably water vapor and contamination scatter), raising cross-sensitivity risk.
• Narrower bandwidth improves selectivity but becomes more sensitive to optical alignment, incidence angle, and temperature-related shifts; manufacturing tolerance can dominate drift.
• Practical rule: treat the filter as a system drift component, not only a “spectral selector,” and validate with humidity/condensation stress tests.

Moisture and contamination are not edge cases in capnography—they are the main drivers of baseline drift, attenuation events, and “good waveform but drifting numbers.” Heating (or another anti-condensation strategy) must be specified as a system requirement, with defined warm-up time, allowed temperature gradients, and serviceability for optical windows and sampling-path components.

• Transmission: intuitive geometry; window contamination directly reduces signal; serviceability matters.
• Folded reflection: compact path; surface condition becomes more influential; requires robust contamination strategy.
• Multi-pass: longer effective path (more sensitivity) but amplifies scatter/film effects; must be paired with strong normalization and condensation control.

Modulation is the practical way to move CO₂ information away from low-frequency drift and ambient interference. The source drive method, modulation frequency, and detector interface must be chosen as a single chain—otherwise common failures (aging, contamination, saturation, bias drift) will appear as “CO₂ changes” in the waveform and EtCO₂.

• Constant-current modulation stabilizes optical amplitude against supply variation and device temperature drift, improving repeatability for ratio/normalization.
• PWM / switching modulation is easy to implement, but it can inject edge-related noise into the detector/AFE unless the driver return path and supply filtering are tightly controlled.
• Both approaches need a defined modulation reference for synchronous demodulation (H2-6), otherwise amplitude drift and ambient effects leak into the baseband.

1. Pick above the 1/f-drift-dominated region so baseline drift is reduced after demodulation.
2. Pick within detector + AFE usable bandwidth so the modulated amplitude is not rolled off or phase-distorted.
3. Avoid dominant interference clusters (mains-related ripple and common lighting flicker harmonics) and ensure the demod/ADC window can lock cleanly.

The hardest requirement is not raw resolution—it is baseline stability and usable dynamic range while humidity, motion, and interference continuously perturb the chain. Synchronous demodulation (lock-in) moves the measurement away from drift and ambient effects, and the ADC + digital filtering must be tied to t90 response and breath-rate dynamics so the plateau remains accurate and timely.

• TIA/PGA noise sets plateau jitter; bias stability sets FiCO₂ baseline credibility.
• Humidity and contamination can create leakage paths; input bias planning and guarding prevent “fake CO₂ offsets.”
• Gain strategy must prevent clipping during strong breaths while still resolving low-amplitude regions and small breaths.

• Modulation moves the useful signal near f_mod, away from low-frequency drift and slow baseline wander.
• Sync demod multiplies by the known reference and then low-pass filters, rejecting ambient light flicker and a large part of 1/f behavior.
• The low-pass corner and averaging window must preserve breath dynamics: overly aggressive filtering smears phase transitions and biases EtCO₂.

• Sampling rate (Fs) must support stable demodulation and a clean digital filter window without aliasing.
• Effective resolution (ENOB) must cover baseline + plateau + transient attenuation events without frequent saturation or quantization collapse.
• Filter window should reduce plateau noise while preserving the time placement of phases—tie it to measured t90 and expected breath rate range.

• Keep driver switching currents out of the detector/AFE return path; isolate the driver supply locally and control edge energy.
• Give ADC reference a clean, quiet path; treat reference noise as direct measurement noise after demodulation.
• Place a simple clipping/baseline monitor in the chain so the UI/logic can flag “not trustworthy” rather than silently drifting.

NDIR does not measure CO₂ directly—it measures optical attenuation. To make EtCO₂ comparable across bedside environments and sampling conditions, the signal must be normalized and then corrected using time-aligned pressure/temperature/humidity inputs, with quality gates that prevent condensation or sensor faults from becoming “CO₂ changes.”

• Pressure shifts can come from local restrictions, partial occlusions, or pump-induced pressure drops in the sampling path.
• Pressure changes gas density and absorption line behavior, so the same optical attenuation can map to different CO₂ unless corrected.
• A usable implementation needs time-aligned pressure (timestamped near the cell/sampling path), not a generic ambient reading.

• Source drift: slow amplitude changes (often common-mode if a reference band exists).
• Detector drift: baseline and gain wander; can look like slow FiCO₂ changes.
• Optical parts: window gradients and pre-condensation behavior change effective throughput.
• AFE drift: bias and gain temperature coefficients translate into baseline offset unless guarded.

• Water vapor affects the relationship between volume fraction and partial pressure, especially near saturated exhaled gas.
• Ignoring humidity can turn real humidity swings into apparent EtCO₂ drift, even when the optical chain is stable.
• Implementations can use a humidity sensor or a bounded model, but the output should carry a correction mode and quality flag.

1. Acquire CO₂-band and REF-band amplitudes (and quality indicators such as clipping/SNR), each timestamped.
2. Normalize using ratio/scale to reduce common-mode drift (source aging, throughput changes) before environmental correction.
3. Align P/T/H samples to the same time base (handle missing/out-of-range readings deterministically).
4. Correct normalized signal with P/T/H to map to standardized CO₂.
5. Output waveform + EtCO₂ + FiCO₂ with quality flags (condensation suspected, contamination suspected, sensor missing, saturation).

Stable capnograms come from two decisions: (1) protect the baseline and plateau with filters that respect breath dynamics, and (2) extract EtCO₂ from a validated plateau region instead of a single peak sample. Motion, leaks, and sampling-path faults must be detected as waveform signatures and converted into quality flags rather than silently reshaping the reported EtCO₂.

• Moving average reduces random jitter, but an oversized window smears phase edges and biases plateau timing.
• Median filtering rejects spikes (motion/EMI-like bursts), but excessive use can round real rising edges.
• Adaptive filters help when noise changes over time, but they require a freeze/rollback rule when quality drops.

• Use a combination of slope/energy checks to detect cycle start, not a single threshold that motion can trigger.
• Define the plateau as a low-slope stable region; compute EtCO₂ from a robust statistic over that region.
• Reject cycles that violate expected phase ordering (burst spikes, clipped segments, or implausible durations).

• Fixed delay is simple, but it breaks when sampling flow changes or tubing compliance/partial occlusion shifts the transport time.
• Adaptive delay adjusts alignment using measured sampling flow/pump status or waveform feature matching, keeping phase boundaries consistent.
• Delay errors show up as phase timing distortion and can make EtCO₂ appear to “jump” even when the plateau level is stable.

• Leak: unstable or missing plateau, reduced amplitude, inconsistent cycle boundaries.
• Rebreathing: elevated baseline (FiCO₂ rises), inspiratory segment does not return to low level.
• Occlusion / water trap full: amplitude drops, delay increases, waveform becomes intermittent or “laggy.”

Long-term EtCO₂ comparability comes from treating the optics + AFE + algorithms as one sensor: align the baseline with zero, align the slope with span, and continuously watch trend gates (ratio drift, baseline drift rate, plateau noise) so contamination, humidity shifts, or aging trigger maintenance or recalibration before they show up as “clinical CO₂ changes.”

• Zero aligns the baseline so slow bias terms do not become FiCO₂ drift (optical throughput changes, AFE bias/leakage, temperature-dependent offsets).
• Span aligns the gain/mapping so amplitude changes do not become EtCO₂ gain error (source aging, detector sensitivity shift, optical efficiency drift).
• Zero/span do not “fix everything” by themselves: cross-sensitivity and condensation still require correction + quality gating.

• Zero reference must be stable and repeatable (baseline alignment is only as good as the reference stability).
• Span reference should be a known CO₂ point (single-point span is common; multi-point is used to validate nonlinearity and residuals).
• Interval rule: shorten calibration intervals when trend gates accelerate (ratio drift rate rises, baseline drift rate rises, plateau noise increases) or when contamination risk is high.
• Maintenance rule: if contamination signatures dominate, cleaning/consumables replacement should be performed before repeating span calibrations.

• Humidity: changes the meaning of standardized output (partial pressure vs vol%) and can shift baseline behavior near saturation.
• Pressure: pump/tubing state changes local pressure, altering absorption mapping even when true CO₂ is steady.
• Temperature: source/detector/optics/AFE drift create slow errors that look like long-term CO₂ trends.
• Contamination: window films and deposits change throughput and can shift CO₂/REF ratio relationships.

• Reference ratio trend: track normalized CO₂/REF behavior; sudden slope change often indicates contamination or optical changes.
• Baseline drift rate: monitor FiCO₂ baseline stability and how quickly it moves over time.
• Plateau noise: rising plateau jitter is an early warning for SNR collapse (aging, water, alignment issues).
• Self-test gates: clipping, amplitude-too-low, sensor-missing, temperature-out-of-range should force a clear degrade mode.

A trustworthy EtCO₂ chain is validated by measurable dynamics (t10/t90, delay), accuracy envelopes across P/T, robustness under realistic disturbances (light, vibration, condensation, contamination), and failure-injection behavior that produces correct flags and safe degrade modes rather than silently shifting the reported value.

• t10/t90 step response verifies how quickly the chain reaches a new CO₂ level without overshoot or long settling tails.
• Delay should be measured as the sum of transport delay (sidestream tubing) and algorithm/display latency.
• Record both rise and fall behavior; asymmetry often indicates transport effects or condensation behavior.

• Multi-point curve: validate nonlinearity and residuals after calibration across the expected range.
• P/T matrix: validate that the P/T/H correction keeps errors bounded across realistic bedside conditions.
• Report results as an error envelope (worst-case bounds), not only an average accuracy number.

• Ambient light: verify baseline drift and plateau noise remain bounded after demodulation.
• Vibration / motion: verify “fake breath” triggers are rejected and quality flags engage.
• Condensation / contamination simulation: verify the chain detects ratio shifts or amplitude collapse and does not silently bias EtCO₂.

• Occlusion: amplitude down + delay up → flag low amplitude/lag; freeze or withhold EtCO₂ as configured.
• Leak: plateau unstable → flag invalid plateau; reject cycle-level EtCO₂ extraction.
• Water trap full: intermittent waveform + lag → flag sampling fault; prompt service/maintenance.
• Source aging: CO₂/REF common-mode drop → aging trend; reduce confidence and trigger maintenance/calibration workflow.

This checklist converts capnography performance drivers (drift, moisture, motion, response time, and comparability) into device categories + key parameters + example part numbers. Each block below includes “failure signatures” to help engineering and sourcing align quickly during design reviews.

Note: part numbers are starting points. Final selection must be validated against modulation frequency, sensor type, temperature range, packaging, and supply availability.

• Current stability: drift here becomes amplitude drift (REF channel can reduce impact, but cannot erase all failure modes).
• Modulation capability: frequency range + edge control must support synchronous demod/lock-in.
• Pulsed current headroom: IR emitters often use pulsed drive for higher optical power.
• Protection: open/short detect and temperature derating to avoid silent long-term degradation.

• Offset and drift: baseline stability (FiCO₂) is usually limited here.
• Low-frequency noise: plateau jitter and “fake breath” triggers often come from 1/f + moisture/motion coupling.
• Input bias current: critical for high-impedance detector interfaces.
• Programmable gain: preserves margin as optics get dirty or emitters age.

• Why it matters: moving the signal away from DC reduces sensitivity to ambient light and slow drift.
• Implementation options: analog switch demod + low-pass, or digital demod after ADC (requires timing alignment).
• Key parameters: switch leakage, charge injection, on-resistance, and timing determinism.

• ENOB + input noise: directly impacts plateau stability and breath metric repeatability.
• Sampling rate: must support the demod/filter window and the desired response time (t90) without “smearing” phase features.
• Trigger/sync: deterministic timing reduces demod phase error and delay ambiguity.

• Pressure: accuracy + response; place so it reflects the sampling path pressure (not a distant enclosure).
• Temperature: prioritize thermal “closeness” to optics/AFE; PCB ambient temperature is often misleading.
• Humidity: evaluate high-humidity drift and protection method (filter/membrane) for long-term stability.

• Timing: stable modulation/trigger timing reduces demod phase error and delay ambiguity.
• Parameter storage: calibration coefficients, CalVersionID, trend thresholds, last QC results.
• Write integrity: avoid half-written coefficients (use atomic update strategy and CRC).
• Supervision: watchdog/reset prevents “silent stale outputs” when software or peripherals hang.

• Reference stability: reference noise/drift can become apparent CO₂ noise/drift after demod and filtering.
• Local decoupling and partitioning: keep the analog reference/AFE domain away from pulsed drive current returns.