• X (in-phase): the baseband component aligned with the reference phase; strongest when phase is correct.
• Y (quadrature): the 90°-shifted component; useful to detect phase error or orthogonal physical response.
• R: magnitude from X/Y (robust amplitude readout when phase may drift).
• θ: phase angle derived from X/Y (phase information relative to the chosen reference).

• Time constant (τ) sets how much the demodulated baseband is averaged: larger τ lowers noise but slows response.
• Roll-off (filter order) sets how aggressively out-of-band baseband noise is rejected: steeper roll-off reduces noise faster, but can increase settling time/overshoot.
• Practical takeaway: choose τ and roll-off based on the slowest signal change that must be tracked and the noise floor required.

• A reference exists (internal, external, or derived from a controlled modulation), so the wanted signal is coherent with it.
• The wanted signal is narrowband around the reference compared with the broadband noise/ interference.
• The front-end can be kept out of overload; otherwise distortion products can create false baseband outputs.

• Phase is not stable (reference drift or loss of coherence): X/Y rotate, θ becomes noisy, and amplitude can “wander.”
• Large interferers saturate the input chain: the lock-in can output a clean-looking number that is actually a demodulated artifact.

1. Set a moderate τ (not too fast) and view X and Y simultaneously.
2. Adjust phase (or auto-phase) until |Y| is minimized for a purely in-phase response.
3. Switch to R for amplitude reporting if residual phase drift is expected.

• Internal reference (NCO): best when the system can apply controlled modulation to the DUT (cleanest coherence).
• External reference input: best when the DUT already has a stable drive/clock; lock to it to keep phase alignment.
• Derived reference (limited): possible only if the signal contains a stable pilot/marker; otherwise phase drift corrupts results.

• PLL lock: removes slow frequency/phase drift to maintain coherence with an external timebase; exposes a measurable “LOCK” state.
• Digital NCO: generates precise sin/cos with programmable phase φ; enables multi-tone and harmonic references without analog drift knobs.
• Phase adjust (φ): aligns the detector to put the dominant response in X and minimize Y (critical for stable readings).

• Fast phase jitter broadens the effective demodulation process and raises the baseband noise floor (worse sensitivity).
• Slow phase drift rotates X/Y over time; amplitude may look steady in R, but θ becomes unreliable and X/Y can appear to “wander.”
• Loss of lock often looks like sudden jumps in X/Y, increased Y leakage, or an R that changes with τ settings rather than physics.

• Why use harmonics: isolate non-linear responses, shift detection away from strong low-frequency noise, or separate mechanisms by harmonic order.
• Main risks: harmonic contamination from the drive path, intermodulation when multiple tones are present, and false baseband artifacts if the input chain overloads.
• Practical guardrails: keep front-end out of saturation, validate with a reference injection check, and confirm that results are stable across τ choices.

1. Confirm a stable reference path: lock indicator OK (if using PLL) and reference level within range.
2. Run phase alignment: minimize Y for a known in-phase stimulus, then hold φ fixed.
3. Change τ by 2–4×: a real coherent signal keeps consistent R; a coherence problem often changes apparent amplitude/phase with τ.

• Voltage mode (diff amp / in-amp): best for low-to-moderate source impedance and voltage-output sensors; preserves phase when input impedance is high and CMRR is strong.
• Current mode (TIA): best for current-output sources or very high impedance sources where voltage pickup and bias errors dominate; converts input current directly into voltage with a defined transimpedance gain.
• If the source impedance is high, prioritize bias current and leakage control first; noise density alone will not predict drift and offset errors.

If the input chain saturates, clips, or recovers slowly, it can create distortion products that the PSD translates into a stable baseband value. This is dangerous because the output can look “clean” while being wrong.

• Saturation/clipping → generates harmonics and intermodulation → PSD demodulates part of that distortion into X/Y.
• Clamp recovery (ESD diodes / TVS / limiter) → asymmetric settling → appears as an offset or slow drift in baseband.
• Range switching transients → repeatable steps or ringing → can masquerade as a coherent response at certain τ values.

• Ensure a linear operating region for expected interferers; do not rely on the PSD to “clean up” overload.
• Place a gentle pre-limit and band-limit before any hard clamp to reduce recovery artifacts.
• Expose an overload flag/counter so field data can separate physics from saturation-induced artifacts.

1. Verify no overload: input limiter not active and overload indicator stays clear under worst-case interference.
2. Check range changes: R and θ should not jump in a way that depends on the selected input range.
3. Short or terminate the input: X and Y should fall close to the expected noise floor for the chosen τ.
4. Apply a known coherent stimulus: minimize Y by phase alignment and confirm repeatability after large transients.

• Analog multiplier: simple conceptually and continuous-time, but limited by multiplier noise, linearity, and drift.
• Chopper / synchronous switching: often better low-frequency stability (less 1/f sensitivity), but requires careful band-limiting to control switching artifacts.
• Digital DDC (ADC + digital sin/cos): flexible for multi-tone and calibration; performance depends on ADC headroom, anti-alias filtering, and reference timing quality.

• With only one reference phase, amplitude and phase are entangled; a phase shift can look like an amplitude change.
• I/Q makes the signal vector observable: R reports amplitude, while θ reports phase relative to the reference.
• Y becomes a diagnostic channel: if a response should be in-phase, a large Y indicates phase misalignment or path mismatch.

Any phase error (reference delay, imperfect quadrature, gain mismatch) rotates the I/Q axes. A purely in-phase response that should sit in X will spill into Y, and the reported θ becomes biased.

1. Apply a known coherent stimulus expected to be mostly in-phase.
2. Adjust φ (or run auto-phase) to minimize |Y| and maximize stable X.
3. Validate by changing τ: R should remain consistent (only noise reduces with larger τ).
4. If residual leakage remains, correct I/Q gain and orthogonality (digital rotation/scaling) and re-check Y floor.

• If R changes dramatically when τ changes, suspect non-coherence or artifact energy entering baseband.
• If Y rises after transients, suspect phase drift or front-end recovery asymmetry rather than real physics.
• If a strong interferer is present, verify overload flags; PSD can demodulate distortion into a stable baseband number.

• Larger τ averages longer → smaller ENBW → lower output noise, but slower response.
• For common LPF implementations, ENBW scales roughly inversely with τ (τ ↑ → ENBW ↓).
• Noise RMS at the output tends to scale with √ENBW: cutting ENBW by ~4× typically lowers noise by ~2× (rule of thumb).
• Use this to estimate “how much noise improvement is worth how much time penalty” before tweaking roll-off.

• 1st order: simplest settling, least risk of surprising behavior; weaker suppression of out-of-band baseband noise.
• 2nd order: practical balance for many measurements; cleaner readout at similar τ, with moderate settling cost.
• 4th order: strongest rejection of baseband noise away from DC; can require more time to fully settle and may appear “slower” after steps.

Practical rule: meet the required response time first with a conservative order (1st/2nd), then increase τ or roll-off only if the noise floor is still above the target.

1. Define the fastest change that must be tracked: scan dwell per point, sweep speed, or smallest transient that matters.
2. Pick a settling requirement: for example, “stable enough for display” vs “stable enough for logging and control.”
3. Start with 1st or 2nd order and choose τ so the output stabilizes within the allowed time window.
4. If noise is still too high, increase τ (primary knob) or increase roll-off (secondary knob) while re-checking settling.
5. Lock the configuration only after step-response and noise-statistics checks agree with expectations.

• Step check: apply a repeatable amplitude step and measure time to reach a stable plateau (same test for each roll-off).
• Noise check: with no signal (or shorted input), confirm X/Y RMS drops predictably when τ is increased.
• Consistency check: R should not change materially with τ for a truly coherent steady signal—only noise should shrink.

• Non-coherence: the interferer is not phase-locked to the reference, so correlation rejects it after demodulation.
• Linear headroom: the interferer does not drive the front-end into compression/clipping; otherwise distortion creates demodulatable artifacts.

1. Front-end headroom + limiter strategy: prevent saturation and ensure fast, symmetric recovery if limiting occurs.
2. Pre-filtering: notch or band-limit large known interferers before PSD (especially 50/60 Hz and harmonics in many lab environments).
3. Reference frequency planning: choose modulation/reference frequencies that avoid interference-rich regions and spur clusters.
4. Consistency validation: confirm results remain stable across τ changes and under controlled addition/removal of interference.

• τ-dependent amplitude: R shifts significantly when τ is changed (beyond noise reduction) → artifact energy is entering baseband.
• Y/θ instability under interference: Y jumps or θ becomes erratic when a large interferer appears → phase rotation or nonlinear mixing.
• Slow recovery after transients: readings remain biased after the interferer is removed → limiter/clamp recovery or range switching imprint.

1. Establish a baseline with a coherent stimulus and record X/Y/R/θ.
2. Add a large non-coherent interferer gradually and monitor overload flags and Y/θ behavior.
3. Change τ by 2–4×; true coherent amplitude stays consistent in R, while artifacts often shift with τ.
4. Remove the interferer and confirm fast return to baseline; slow return indicates recovery-induced baseband bias.

• Oversampling + noise shaping push quantization noise away from the low-frequency region of interest.
• Decimation filters set a predictable baseband bandwidth, matching the lock-in’s time-constant-driven averaging.
• Low-frequency behavior (drift and 1/f-like effects) often dominates real lock-in accuracy more than headline sample rates.

• Analog anti-aliasing still matters: large out-of-band content can saturate the front-end or modulator before any digital filter can help.
• Decimation sets the effective baseband bandwidth: choose it to comfortably cover the intended measurement bandwidth; overly aggressive decimation can “slow” the measurement.
• Keep headroom under interferers: if a strong interferer drives nonlinear behavior, distortion products can become partially correlated and appear as false baseband.
• Timing quality affects baseband purity through the reference path; keep this as an impact path and validate with τ-sweep consistency rather than relying on headline specs.

Range changes (gain/attenuation, relay or switch matrix, digital scaling) can inject transients and bias shifts that settle slowly. With long τ, these effects can masquerade as a stable coherent output.

• Predictable settling time after a range change should be specified and testable.
• No directional bias: X/Y should return to the same baseline regardless of switching direction.
• Phase sanity: Y and θ should not jump in ways inconsistent with the physical signal path.

1. Under worst-case interferers, verify no overload and no slow recovery offsets in X/Y.
2. Change τ by 2–4×: R should remain consistent for a steady coherent signal; only noise should shrink.
3. Perform a controlled range switch: confirm settling time and absence of directional bias in X/Y.
4. With input shorted/quiet, ensure X/Y noise follows the expected τ trend and does not show spur-driven “DC” offsets.

• Amplitude (gain): correct the end-to-end gain per range so the same coherent stimulus maps to the same input-referred value.
• Phase: define a zero-phase point and correct I/Q quadrature and gain mismatch so X/Y axes are stable.
• Offset / drift: suppress long-term bias from front-end thermals, reference amplitude drift, and ADC reference drift.

• Reference injection: inject a known calibration tone/amplitude through an internal switch into the measurement path.
• Known gain path: verify gain consistency across range steps (analog gain/attenuation and digital scaling).
• Executable acceptance: input-referred amplitude should match across ranges after correction, and should not change materially with τ (only noise shrinks).

Phase alignment requires a defined reference condition for “in-phase,” plus correction of quadrature errors (non-ideal 90° separation) and I/Q gain mismatch. These errors rotate the measurement axes and create X→Y leakage.

• Auto-phase: adjust φ to minimize |Y| for a stimulus expected to be in-phase.
• Quadrature correction: apply digital rotation/scaling to restore orthogonality and equalize I/Q gain.
• Executable acceptance: for an in-phase stimulus, Y should approach the noise floor and θ should stay stable across temperature and reboot.

• Front-end thermal drift → X/Y baseline wanders with temperature → periodic zero/self-cal + temperature-tagged constants.
• Reference amplitude drift → R slowly scales over time → reference monitoring and gain recalibration interval.
• ADC reference drift → global amplitude bias across ranges → reference monitor + calibration constants with versioning.
• Range-switch bias → step-like offsets after switching → enforce settling time and store per-range offset trims.

• Triggers: power-up, periodic runtime timer, temperature delta beyond a threshold, and after large range changes.
• What to run: zero/offset first, then amplitude gain, then phase alignment (in that priority order).
• What to store: calibration version ID, timestamp, temperature, ranges involved, pass/fail status, and updated constants summary.

1. Choose reference/modulation frequency: avoid 50/60 Hz harmonics and local spur clusters; pick a quiet window.
2. Select input mode and range: voltage vs current mode to match the source; ensure headroom under worst-case interferers and transients.
3. Set τ and roll-off: hit the required response time first; then narrow ENBW for noise reduction without violating settling.
4. Enable I/Q and phase optimization: auto-phase for a known in-phase stimulus, then lock φ to keep Y near the noise floor.
5. Check X/Y/R/θ behavior: look for τ-dependent amplitude, θ jumps, or large Y under interference—these point to artifacts or overload.
6. Run a minimal validation: perform a reference injection (or self-cal) and confirm amplitude/phase consistency before logging real data.

• Noisy readout → τ too small / ENBW too wide → increase τ or roll-off after confirming settling.
• Slow or “frozen” sweep response → τ too large / over-filtered → reduce τ or roll-off, then re-check stability.
• Stable-looking but inconsistent R → overload or distortion folding into baseband → increase headroom, pre-filter, or move frequency window.
• Bias after range changes → switching transients and recovery memory → enforce a settling delay and re-run a quick self-cal.

1. Choose reference/modulation frequency: avoid 50/60 Hz harmonics and local spur clusters; pick a quiet window.
2. Select input mode and range: voltage vs current mode to match the source; ensure headroom under worst-case interferers and transients.
3. Set τ and roll-off: hit the required response time first; then narrow ENBW for noise reduction without violating settling.
4. Enable I/Q and phase optimization: auto-phase for a known in-phase stimulus, then lock φ to keep Y near the noise floor.
5. Check X/Y/R/θ behavior: look for τ-dependent amplitude, θ jumps, or large Y under interference—these point to artifacts or overload.
6. Run a minimal validation: perform a reference injection (or self-cal) and confirm amplitude/phase consistency before logging real data.

• Noisy readout → τ too small / ENBW too wide → increase τ or roll-off after confirming settling.
• Slow or “frozen” sweep response → τ too large / over-filtered → reduce τ or roll-off, then re-check stability.
• Stable-looking but inconsistent R → overload or distortion folding into baseband → increase headroom, pre-filter, or move frequency window.
• Bias after range changes → switching transients and recovery memory → enforce a settling delay and re-run a quick self-cal.