123 Main Street, New York, NY 10001

Submit Your BOM

Phase Noise Analyzer (Cross-Correlation & Discriminators)

Phase Noise Analyzer (Cross-Correlation & Discriminators)

← Back to: Test & Measurement / Instrumentation

A phase noise analyzer measures tiny phase fluctuations and reports them as a normalized L(f) curve plus integrated jitter, not just a spectrum trace. Credible results come from a calibrated discriminator and a verified residual floor, with cross-correlation used to push the measurement below the instrument’s own noise.

H2-1 · What this page solves: making phase-noise measurements trustworthy

A phase noise analyzer is not “just looking at spectrum.” It measures phase fluctuations by translating small phase error Δφ(t) into a clean baseband signal, then normalizes the result into L(f). When the DUT is quieter than a single measurement chain, a two-channel cross-correlation architecture pushes below the instrument’s own residual noise floor.

What you should get from this page
  • L(f) curve reading rules: how close-in vs far-out shapes map to mechanisms (not just a pretty plot).
  • Residual noise floor discipline: how to prove you are seeing the DUT, not the analyzer chain.
  • Spur tagging rules: how to label discrete tones and avoid “averaging them away.”
  • Integrated jitter results: how integration limits change the number, and how to report it responsibly.
Scope boundary (to prevent topic overlap)

This page stays inside the phase-noise analyzer chain: cross-correlation, phase discriminator, LO/reference cleanliness, ADC/FFT settings, calibration/self-check, and error sources that create false noise floors or spurs.

F1 — Phase noise analyzer measurement chain overview Block diagram showing a DUT oscillator split into two independent channels, each with a discriminator, baseband chain, ADC and FFT. A cross-correlation engine combines both channels to reduce uncorrelated instrument noise and output L(f). COMMON (DUT) UNCORRELATED (INSTRUMENT) CROSS-CORRELATION DUT Oscillator Carrier + Δφ(t) S Split CH-A Discriminator Phase → Baseband Baseband Amp + Filter ADC Low Noise FFT RBW / Window CH-B Discriminator Independent Chain Baseband Amp + Filter ADC Low Noise FFT RBW / Window Cross-Correlation Keeps common DUT noise Averages down uncorrelated noise Residual floor decreases with N Output L(f) plot · Spurs tagged · Integrated jitter spur Trustworthy measurement = clean phase-to-baseband conversion + calibrated FFT + correlation that removes uncorrelated instrument noise.
F1 — Two independent channels measure the same DUT; cross-correlation keeps the common DUT noise and averages down uncorrelated channel noise.

H2-2 · Core outputs & terms: L(f), residual floor, and integrated jitter

Fast definitions (one sentence each)
  • L(f): single-sideband phase-noise density at offset f, normalized to the carrier (commonly shown in dBc/Hz).
  • Residual noise floor: the analyzer’s own equivalent phase-noise limit when the DUT is not the dominant noise source.
  • Integrated jitter: a time-domain number derived by integrating phase noise over a specified offset-frequency range.
  • Spurs: discrete tones that must be tagged and handled by a stated rule (include/exclude, or report separately).
Plot symptoms → engineering meaning (4-line cheat sheet)
Close-in region rises steeply (1/f³ or 1/f²)
Often indicates flicker-related mechanisms or discriminator/reference sensitivity at low offsets. Verify stability of the discriminator operating point and reference cleanliness before blaming the DUT.
Mid offsets show ~1/f or a gentle slope
The chain transitions from flicker-dominated behavior to broadband noise. Check baseband gain planning and ensure ADC noise is below the expected DUT region.
Far-out region becomes flat (white floor)
A flat floor is usually set by broadband noise (baseband electronics + ADC + DSP). If the floor is above expectation, verify filtering, ADC full-scale utilization, and alias control.
Discrete spikes appear (spurs)
Spurs must be tagged and traced (reference-related, power-related, coupling). A trustworthy report states a spur handling rule: include/exclude in jitter integration, or report spur jitter separately.
Integrated jitter: the number changes with your limits (so limits must be stated)

Integrated jitter is computed over a specific offset-frequency band. Changing the lower/upper integration limits can change the jitter number dramatically, especially when close-in flicker or spurs dominate. Always report the limits and the spur rule with the result. 

Approximate relationship (SSB L(f) form):
σt ≈ (1 / (2π f0)) · √( 2 · ∫[f1..f2] L(f) df )

Report with:
- Carrier frequency f0
- Integration limits [f1..f2]
- Spur handling rule (include/exclude or separate spur jitter)
F2 — Phase-noise plot zones and typical slopes Stylized phase-noise plot showing close-in to far-out regions and typical slope labels 1/f^3, 1/f^2, 1/f, and flat floor. A spur is highlighted as a discrete tone that should be tagged. L(f) Offset frequency (f) close-in mid far-out white floor 1/f³ 1/f² 1/f flat spur (tag) Plot shape hints mechanisms; jitter depends on stated integration limits and spur handling rules.
F2 — A trustworthy report labels regions, tags spurs, and states integration limits for any jitter number.

H2-3 · Measurement modes map: why “discriminator + FFT” (not just spectrum)

Phase noise is a time-varying phase error Δφ(t). A phase noise analyzer measures it by converting small phase fluctuations into a linear baseband signal, then using FFT to estimate noise density and normalize it into L(f). This avoids “carrier-dominance” problems where the huge carrier masks tiny phase modulation if only a direct spectrum view is used.

Mode 1 — Single-channel (fast, but limited by residual floor)
  • Best when the DUT is clearly above the analyzer’s floor across the offsets of interest.
  • Key limitation: the result becomes “DUT + instrument residual” once the curve approaches the floor.
  • Quick sign: increasing averages does not push the floor lower after a point.
Mode 2 — Dual-channel cross-correlation (for DUT below single-channel floor)
  • Two independent channels see the same DUT; most channel noises are uncorrelated.
  • Correlation processing retains the common DUT contribution while averaging down uncorrelated noise.
  • Hard limit: any correlated leakage (crosstalk/shared reference coupling) will not average down.
Selection rule (practical, repeatable)
  • If the measured curve “sticks” to a flat floor and extra averaging does not lower it, the test is floor-limited (single-channel) or correlation-limited (leakage).
  • If the floor decreases as averaging increases, cross-correlation is working and can reveal quieter DUT noise.
  • If the DUT is far above the floor, single-channel is usually sufficient for fast sweeps and comparisons.
F3 — Single-channel vs cross-correlation measurement modes Side-by-side block diagrams comparing single-channel measurement (instrument noise adds to result) and dual-channel cross-correlation (uncorrelated channel noise averages down with more averages N). Single-channel Cross-correlation DUT Carrier + Δφ(t) Discriminator Phase → Baseband ADC FFT Instrument noise Measured result DUT + residual floor DUT Carrier + Δφ(t) S CH-A Disc + ADC/FFT CH-B Disc + ADC/FFT Cross-Correlation uncorrelated noise ↓ Averages N ↑ → Floor ↓ Revealed DUT noise below single-channel floor Correlated leakage will not average down
F3 — Single-channel is limited by residual floor; cross-correlation reduces uncorrelated channel noise as averages increase, but correlated leakage remains.

H2-4 · Choosing a phase discriminator: mixer, delay-line, and digital detectors (and their traps)

The discriminator is the “transducer” that turns phase error into something measurable. Its sensitivity (V/rad), linear range, ability to stay near quadrature, and susceptibility to AM-to-PM leakage decide whether close-in noise is truly from the DUT or created by the measurement chain.

Selection criteria (use these to prevent false floors)
  • Sensitivity (V/rad): must lift DUT noise above baseband + ADC noise without clipping.
  • Linear range: overload or compression can create fake spurs or flatten slopes.
  • Quadrature maintenance: drift away from 90° changes conversion gain and corrupts comparisons.
  • AM-to-PM leakage risk: amplitude noise can masquerade as phase noise if rejection is weak.
  • Low-frequency limit: 1/f, drift, and microphonics often dominate the close-in region.
Discriminator type × key criteria (no part numbers)
Type Strength Common trap Best offsets Quick validation
Mixer discriminator Simple, high sensitivity near quadrature. Quadrature drift + AM leakage creates false close-in rise. Wide, but close-in needs stable operating point. Sweep phase trim: gain and floor should behave predictably.
Delay-line discriminator Good far-out conversion; strong phase-to-time mapping. Cable/temperature microphonics corrupt close-in region. Mid/far offsets; close-in limited by drift. Tap the fixture / warm it slightly: watch close-in change (should not).
Digital phase detector Convenient scaling, easy digital calibration & logging. Reference-related spurs & sampling artifacts appear as tones. Depends on sampling; far-out can be strong if aliasing is controlled. Change sample rate / anti-alias settings: true DUT slopes stay consistent.

Tip: A discriminator is “good enough” only if its sensitivity stays stable and AM leakage does not reshape the close-in slope.

Minimal checks that prevent AM-to-PM mistakes
  • Power sweep: change input level slightly; if close-in floor moves unexpectedly, suspect AM leakage or compression.
  • Quadrature sweep: small phase trim around 90°; sensitivity should peak near quadrature and behave smoothly.
  • Channel consistency: in correlation mode, CH-A and CH-B should show similar shapes before correlation is applied.
F4 — Discriminator operating point and AM-to-PM leakage path Left panel shows a simplified conversion gain vs phase offset with a marked quadrature point at 90 degrees. Right panel shows an amplitude-noise path leaking into a mixer discriminator and appearing as false phase noise at baseband. Quadrature point AM → PM leakage Sensitivity Phase offset 90° (quadrature) Drift away from 90° changes gain and increases nonlinearity risk Amplitude noise AM(t) Mixer discriminator phase → baseband Baseband output looks like Δφ(t) False phase noise risk AM leakage reshapes close-in slope leakage Check by: power sweep · quadrature sweep · channel consistency
F4 — Keep the discriminator near quadrature for stable V/rad; treat AM-to-PM leakage as a first-class error source in close-in measurements.

H2-5 · Cross-correlation explained: why the floor drops, and when it fails

Why it works (intuition that matches bench reality)
  • Common term: DUT phase fluctuations are seen by both channels, so correlation keeps them.
  • Uncorrelated term: most channel electronics noise differs between CH-A and CH-B, so it averages down.
  • Averages N ↑ → floor ↓: increasing correlation averages reduces only the uncorrelated floor; the DUT contribution remains.
  • Hard limit: any noise that becomes correlated (leakage, shared reference coupling) will not drop with more averages.
Why it fails (checklist)
  • Channel leakage / shared coupling: spur or floor becomes correlated, so it stops averaging down.
  • Synchronization errors: mismatched timing/processing causes unstable correlation results or “algorithm-shaped” artifacts.
  • Environmental correlation: temperature drift, cable microphonics, or vibration changes both channels together.
Fast bench checks
  • Increase averages: if the floor never trends downward, suspect correlated leakage or a shared injection path.
  • Swap CH-A/CH-B inputs: if the same spur stays “locked” to both, it is likely shared/correlated.
  • Stabilize cables/fixtures: if close-in changes with touch or airflow, environmental correlation is dominating.
F5 — Classifying common vs uncorrelated noise in cross-correlation Diagram labeling DUT phase noise as common to both channels and channel electronics noise as uncorrelated. A red leakage path shows how shared coupling makes noise correlated and prevents floor reduction with averaging. COMMON (kept) UNCORRELATED (averages down) LEAKAGE → CORRELATED (won’t drop) DUT Common phase noise S CH-A chain Disc · BB · ADC/FFT CH-B chain Disc · BB · ADC/FFT Noise_A uncorrelated Noise_B uncorrelated Cross-Correlation Averages N ↑ → Floor ↓ keeps common DUT term Output L(f) common DUT noise + reduced uncorrelated floor leakage → correlated If the floor stops improving with N, look for correlated injection: coupling, shared references, or environment.
F5 — Cross-correlation reduces uncorrelated channel noise; leakage paths make noise correlated and block further floor reduction.

H2-6 · Ultra-low-noise LO/reference: when the LO becomes a “fake DUT”

A phase noise analyzer can only be as clean as the reference energy and distribution feeding the discriminator paths. If LO/reference noise or its distribution leakage dominates the close-in region, the measurement can report the instrument’s behavior as if it were the DUT.

Symptom → likely cause → quick verification (engineering checklist)
Symptom Likely cause Quick verification
Close-in floor rises or drifts over time Reference distribution drift, ground coupling, cable microphonics Stabilize cables/fixtures; change distribution path; see if close-in stabilizes
Identical spurs appear in both channels Shared reference/divider spur injection, shared power ripple Change ref/divider settings; isolate power/ground; spur should move or weaken
Cross-corr floor stops improving with more averages Reference leakage makes noise correlated across channels Add isolation / change splitter / break shared ground; watch floor trend return
Far-out floor too high vs expectation Baseband/ADC noise dominates, alias control or scaling is weak Increase baseband gain safely; adjust anti-alias/RBW; ensure ADC uses healthy full-scale
Spur hygiene rule of thumb

Spurs are often created by reference/divider activity or power coupling. Treat them as “named mechanisms”: tag frequency and suspected origin, and do not assume averaging will remove them.

F6 — LO/reference distribution and correlated leakage paths Diagram of an ultra-low-noise LO/reference feeding a splitter into CH-A and CH-B. Red paths show shared ground, power ripple injection, and splitter leakage that can create correlated noise and spurs across channels. Ultra-low-noise LO / Reference close-in critical Splitter / Dist. Isolation matters CH-A Ref In feeds discriminator CH-B Ref In feeds discriminator Risk outcome correlated noise + spurs cross-corr floor stops dropping shared ground power ripple leakage spur The cleanest LO is still dangerous if distribution creates shared coupling that makes noise correlated across channels.
F6 — Treat LO/reference distribution as part of the measurement chain: isolation, grounding, and ripple control decide whether cross-corr can keep improving.

H2-7 · Baseband chain + ADC: dynamic range, 1/f, clipping, and aliasing define the floor

The measurable noise floor is set by a practical budget: baseband amplifier noise (especially 1/f close-in), ADC noise/quantization, and any aliased noise folded back from beyond Nyquist. Gain staging must use ADC full-scale efficiently without clipping, or the measurement will manufacture false spurs and reshape the floor.

Design criteria (each has a clear cause → effect)
  1. Gain to fill ADC wisely: too small wastes ENOB; too large clips and creates spurs that look “real”.
  2. 1/f dominates close-in: lower offsets amplify the impact of drift/1/f, so baseband parts and bandwidth must be chosen intentionally.
  3. Bandwidth is noise bandwidth: opening baseband BW integrates more noise and raises the floor with no measurement benefit.
  4. Limiter behavior must be predictable: hidden saturation/recovery can elevate close-in and introduce intermittent spur artifacts.
  5. Anti-alias sets far-out truth: insufficient roll-off folds wideband noise into baseband and lifts the far-out floor.
  6. Sample rate and filtering must match: higher Fs helps only if Nyquist is pushed out and the anti-alias corner follows the plan.
  7. ADC noise must not dominate: if ADC sets the floor, changing gain will not improve the curve the way the budget predicts.
  8. Prove the budget on the bench: use terminated/shorted input and a known tone to separate amp/ADC/alias contributions.
Fast bench checks (what to change and what it means)
  • Gain sweep: if floor and slope do not respond predictably, suspect ADC limit or alias folding.
  • Bandwidth toggle: wider BW should raise integrated floor; if not, a hidden limiter or processing artifact may be active.
  • Fs / anti-alias toggle: if far-out floor moves strongly, alias folding is present and must be controlled.
  • Clip fingerprint: reducing headroom should grow spurs/skirts; if this happens, the prior “clean” result was margin-sensitive.
F7 — Baseband/ADC dynamic range budget and must-be-below thresholds Horizontal budget bars showing DUT noise, baseband amplifier noise (including 1/f), ADC noise/quantization, and aliased noise fold-back, compared against a target floor line and a headroom/clipping line. Dynamic range & floor budget lower noise → better DUT noise AMP noise ADC noise ALIAS fold-back DUT (to be measured) AMP + 1/f close-in ADC noise + quantization Aliased noise (fold-back) TARGET floor must be below HEADROOM avoid clipping Use ~60–90% ADC FS Improve the floor by reducing AMP 1/f, avoiding clipping, and preventing alias fold-back with proper anti-alias filtering.
F7 — A practical noise budget: baseband 1/f, ADC noise, and alias folding must sit below the target floor while preserving headroom.

H2-8 · FFT/DSP: how RBW, windows, and averaging change close-in truth and spur credibility

FFT settings do not just “pretty up” plots. Bin width, window sidelobes, and averaging math can change the measured floor and make spurs look cleaner or worse. Reliable phase-noise work chooses settings that control leakage and statistical bias, then validates that changes affect the display in expected ways.

Setting → impact → recommended approach
  1. NFFT / bin width → narrower bins reduce RBW but need more averages for stable floors → start medium, then tighten only where needed.
  2. RBW vs ENBW → window choice changes ENBW and noise-density scaling → keep the window consistent for comparisons.
  3. Window sidelobes → high sidelobes leak spur energy into the “floor” → use low-sidelobe windows for spur credibility.
  4. Main-lobe width → low-sidelobe windows widen the main lobe → accept slightly worse resolution to avoid false skirts.
  5. Leakage sanity check → if spur skirts reshape strongly when switching windows, the prior “floor” may be leakage.
  6. Averaging type → log averaging can bias the displayed floor → prefer power/linear-domain averaging for quantitative floors.
  7. Average count → more averages reduce variance but increase time → first verify settings, then run longer for final curves.
  8. Display smoothing → can hide tones and distort slopes → use for visualization only, not for pass/fail numbers.
Quick credibility rule for spurs

A real spur keeps its center frequency; its apparent skirts should improve when sidelobes are reduced. If the “spur neighborhood” changes wildly with window choice, spectral leakage is being observed, not true broadband noise.

F8 — Window sidelobes and spectral leakage around a spur Three simplified plots show the same narrow spur under different windows. Higher sidelobes cause more leakage into adjacent bins; lower sidelobes improve spur credibility but widen the main lobe. Spectral leakage vs spur credibility RECT high sidelobes HANN medium sidelobes LOW-SIDELOBE cleaner spur Resolution main lobe width Leakage sidelobes drive spur skirts Use low-sidelobe windows to trust spur amplitude and avoid leakage posing as “noise floor”.
F8 — Window choice trades resolution for sidelobe leakage; spur credibility improves when sidelobes are reduced.

H2-9 · Calibration & “residual noise” method: proving you measured the DUT, not the instrument

Calibration is the credibility layer of a phase-noise analyzer. A complete run must show (1) the residual floor is known, (2) discriminator sensitivity is correctly scaled (V/rad or equivalent), and (3) self-tests confirm symmetry and robustness. The steps below form a traceable, repeatable workflow.

Steps 1–6 (each step outputs a recordable metric)
Step 1 — Normal baseline
Purpose: establish a baseline L(f) + spur list under the intended RBW/window/AvgN so later checks have a reference.
Record: L(f) curve, SpurList, AvgN, RBW, Window, Temp.
Step 2 — Residual floor check
Purpose: measure the instrument residual noise floor so the DUT result is not mistaken for analyzer self-noise or coupling artifacts.
Action: replace the DUT with a terminated/known substitute or a reference path, keep the same processing settings, and run the same averaging plan.
Record: ResidualFloor (close-in & far-out summary), Δ vs baseline, SpurList.
Step 3 — Sensitivity calibration (V/rad)
Purpose: convert measured baseband voltage spectrum into phase noise correctly by calibrating discriminator gain (V/rad) or an equivalent scale factor.
Action: apply a known phase perturbation / injected tone at a defined offset and extract the scale factor from the observed baseband response.
Record: V_per_rad, CalVersion, CalDate, tolerance band.
Step 4 — Injection self-test
Purpose: validate DSP settings (RBW/window/averaging) so spectral leakage does not impersonate “noise floor” and tones are labeled consistently.
Action: run two reasonable window/RBW configurations and confirm the injected tone keeps the same center offset while skirts behave as expected.
Record: InjectionTone (offset & amplitude), window/RBW pair, consistency check.
Step 5 — Channel swap (symmetry check)
Purpose: remove channel bias and expose crosstalk/coupling; a believable result does not depend on which path is called “A” or “B”.
Action: swap CH-A/CH-B input routing (and key cables if practical), then repeat baseline with unchanged processing.
Record: ΔSwap (overlay difference), Spur migration (yes/no), SymmetryOK.
Step 6 — Decision & archive
Purpose: produce an auditable deliverable package with all settings, calibration state, environment and raw evidence preserved.
Record: Pass/Fail, ThresholdRef, SetupID, LORef, AvgN, RBW, Window, Temp, RawDataFile, RawDataHash.
F9 — Calibration & self-test state machine with recordable outputs State machine: Normal baseline → Residual floor check → Sensitivity calibration → Channel swap → Pass/Fail. Each state outputs specific metrics to record for traceability. Calibration & self-test workflow Normal L(f), SpurList Residual ResidualFloor SensCal V/rad, CalVer Swap ΔSwap, OK? Pass / Fail ThresholdRef Troubleshoot see H2-10 Each state produces a recordable metric so results remain traceable across days, cables, and operators.
F9 — Calibration/self-test state machine with explicit outputs to record for auditable phase-noise measurements.

H2-10 · Common artifacts & troubleshooting: AM-to-PM, ground loops, microphonics, and false spurs

Troubleshooting should start from the curve’s “fingerprint.” The boxes below map each symptom to likely causes and the fastest checks that confirm or eliminate each hypothesis.

1) Symptom: close-in “hump” or unexpected low-offset lift
Likely causes: baseband 1/f and drift, discriminator operating-point drift, airflow/temperature gradients, cable microphonics.
Fast checks: log temperature and warm-up time, narrow baseband bandwidth, fix cables mechanically, repeat after airflow changes (fan on/off).
2) Symptom: a sharp spur at a fixed offset that never moves
Likely causes: reference/divider leakage, digital clock coupling, supply ripple, LO feedthrough into discriminator/baseband.
Fast checks: change RBW/window to see if only skirts change, swap reference source or divider settings, isolate/clean supplies, reroute grounds.
3) Symptom: comb spurs (regular spacing “teeth”)
Likely causes: switcher frequency/harmonics, periodic digital activity, fractional-N patterns, periodic calibration loops.
Fast checks: change switcher frequency (or load) and check spacing, disable periodic tasks, compare day-to-day and cable-to-cable repeat.
4) Symptom: far-out floor is too high and refuses to improve
Likely causes: alias fold-back from insufficient anti-alias filtering, ADC/noise budget dominated by digitizer, baseband bandwidth too wide.
Fast checks: change Fs/anti-alias settings, reduce baseband bandwidth, run a gain sweep to see whether ADC limit is dominating.
5) Symptom: cross-correlation does not lower the floor as AvgN increases
Likely causes: channel crosstalk/shared leakage (making noise correlated), timing/sync error, shared LO/reference contamination.
Fast checks: increase physical isolation and check routing, repeat channel swap, isolate LO distribution, verify synchrony and trigger alignment.
6) Symptom: the curve drifts with time (same settings, different answer)
Likely causes: temperature drift, connector contact noise, cable motion, discriminator quadrature drift.
Fast checks: record Temp and warm-up, re-seat connectors and secure cables, rerun sensitivity calibration, repeat after mechanical stabilization.
7) Symptom: changing window function “changes the floor”
Likely causes: spectral leakage is dominating, RBW/bin width is too aggressive, averaging count is insufficient.
Fast checks: use a low-sidelobe window and compare, increase AvgN, validate with an injected tone that center offset stays fixed.
8) Symptom: AM-to-PM artifact (amplitude noise masquerades as phase noise)
Likely causes: insufficient AM rejection in the discriminator, nonlinearity in mixers/amps, operating point offset from quadrature.
Fast checks: change input level/attenuation, restore quadrature/operating point, compare results with AM suppression enabled/disabled if available.
9) Symptom: the plot looks “too smooth” but lacks physical consistency
Likely causes: over-smoothing, biased averaging (e.g., log average), hidden post-processing, incorrect sensitivity scaling.
Fast checks: review raw spectra, switch averaging type (power vs log), rerun sensitivity calibration and verify the global offset does not jump unexpectedly.
F10 — Symptom fingerprint atlas for phase-noise troubleshooting Four simplified thumbnails show typical curve fingerprints: close-in hump, comb spurs, flat high floor, and random wander. Each thumbnail includes short keywords to guide fast checks. Troubleshooting fingerprints Close-in hump drift · 1/f · microphonics Comb spurs switcher · clock · divider Flat high floor alias · ADC · bandwidth Random wander temp · contact · cable motion Match fingerprint → run fast checks Use the fingerprint to choose the quickest verification step before changing many settings at once.
F10 — Symptom atlas: recognize the curve shape first, then validate with targeted checks to isolate the root cause.

H2-11 · Validation checklist: how to prove the phase-noise measurement is “done”

A deliverable phase-noise report is more than a pretty L(f) plot. It must show that the residual floor is understood, the result is repeatable, and the full setup can be reconstructed later. The checklist below structures proof into R&D validation, reproducibility, and delivery-grade logging.

Layer 1 — R&D validation (prove the curve is physically correct)
  1. Residual floor baseline: run a residual/noise-only condition and save the curve and spur list so the instrument floor is visible.
  2. Average trend sanity: increase AvgN and confirm the floor improves only until it reaches a stable platform (a sign of true limits).
  3. Channel swap consistency: swap CH-A/CH-B inputs (and key cables if practical) and verify the curve shape and spur pattern remain consistent.
  4. Injection verification: inject a known, traceable tone/modulation and confirm its offset and amplitude are detected and labeled correctly.
  5. Anti-alias sanity: change Fs or anti-alias bandwidth and confirm far-out behavior responds in an explainable way (flagging fold-back risk).
  6. Clipping fingerprint: adjust gain/headroom to ensure spurs/skirts do not appear due to saturation; keep operation in a safe full-scale window.
  7. Window/RBW robustness: repeat with a second reasonable RBW/window set; the main shape should stay consistent while variance/leakage changes predictably.
Evidence to archive (minimum)
  • Overlay plots (baseline vs DUT), plus a spur list with thresholds.
  • Channel-swap overlay and a short note on what changed (if anything).
  • Injection run files (before/after) with the injected offset annotated.
Layer 2 — Reproducibility (prove someone else can reproduce it)
  • Day-to-day repeat: repeat on different days after power cycling; record the typical drift band for close-in and far-out regions.
  • Cable/fixture repeat: repeat after re-connecting cables and fixtures; document any microphonic/thermal sensitivity observed.
  • Setting repeat: repeat with an alternate “reasonable” RBW/window combination; the main trend should remain stable.
  • Operator-proof steps: keep a short SOP (connect → warm-up → calibrate → measure → save → judge → archive) so results do not depend on experience.
Practical pass/fail wording

A “pass” statement should reference the exact threshold and conditions (offset range, RBW/window, AvgN, temperature and connection), not just a single dBc/Hz number.

Layer 3 — Delivery-grade log (record fields for traceability)
Category Must-record fields Why it matters
Setup identity SetupID, DUT_ID, ConnectionMap, DiscType, LORef, ChannelMap Allows reconstruction of the exact routing and the main leakage risks.
DSP / sampling Fs, AntiAliasCfg, RBW, Window, AvgN, AvgType, SpurPolicy Prevents “settings-dependent” results from being mistaken for DUT behavior.
Environment & calibration Temp, DateTime, CalVersion, CalDate, WarmupTime Captures drift drivers and ensures the calibration state is traceable.
Result & decision ResidualFloor, SpurList, PassFail, ThresholdRef, RawDataFile, RawDataHash Makes the conclusion auditable and repeatable from raw evidence.
Verification BOM anchors (example parts used to build repeatable checks)
Clock / reference distribution
  • Jitter cleaner / clock tree: ADI HMC7044, TI LMK04828, ADI AD9545
  • Low phase-noise oscillator anchors: Crystek CCHD-957 (family), SiTime SiT5356 (family)
  • 2-way splitter (distribution baseline): Mini-Circuits ZFSC-2-1+ (family)
Discriminator / injection helpers
  • Phase/gain detector anchor: ADI AD8302
  • I/Q modulator anchors (controlled injection): ADI ADL5370, ADI ADL5375
Baseband & ADC chain anchors
  • Low-noise baseband op-amps: ADI ADA4898-2, TI OPA1612
  • Zero-drift (close-in drift control): ADI ADA4522-2
  • High-resolution ΣΔ ADC anchor: ADI AD7768 / AD7768-4
  • High-speed ADC anchors (if downconversion is used): ADI AD9208, ADI AD9689
Low-noise supplies & references (spur hygiene)
  • Low-noise LDO anchors: ADI/LT LT3042, ADI/LT LT3045
  • Precision reference anchors: ADI ADR4550, ADI/LT LTC6655

Part numbers above are engineering anchors (commonly used classes). Equivalent-grade substitutes are acceptable if isolation, noise, and drift targets are met and recorded.

F11 — Delivery log template for phase-noise measurements Field template showing which parameters must be recorded to reproduce and audit a phase-noise measurement: setup identity, processing settings, environment/calibration, and result/decision. Measurement Record (Deliverable Fields) SETUP SetupID DUT_ID ConnectionMap DiscType LORef ChannelMap PROCESSING AvgN RBW Window Fs AntiAliasCfg SpurPolicy RESULTS ResidualFloor SpurList Pass / Fail Temp CalVersion RawDataHash Archive RawDataFile DateTime ThresholdRef (spec / requirement) + Decision note A “done” measurement is one that can be re-run and audited: settings, environment, calibration state, and evidence are all preserved.
F11 — Record template: Setup identity, processing settings, environment/calibration, and result fields required for reproducible phase-noise deliverables.

Request a Quote

Accepted Formats

pdf, csv, xls, xlsx, zip

Attachment

Drag & drop files here or use the button below.

H2-12 · FAQs × 12 (with answers + Google FAQ structured data)

These FAQs target practical measurement credibility: separating DUT noise from instrument limits, choosing DSP settings that do not bias L(f), and producing a deliverable report that can be reproduced and audited.

1) Why does cross-correlation sometimes fail to lower the noise floor?
Cross-correlation only averages down noise that is truly uncorrelated between channels. If the two paths share leakage (shared LO/reference contamination, crosstalk, ground coupling), that noise becomes correlated and will not drop with AvgN. Verify by increasing isolation, swapping channels, separating reference distribution, and checking time alignment/synchrony.
2) How can DUT phase noise be distinguished from the instrument residual floor?
Run a residual-floor measurement with the same RBW/window/AvgN, using a terminated or known substitute input (or a reference path), then overlay it with the DUT result. Where the DUT curve sits on the residual floor, the measurement is instrument-limited. Record ResidualFloor and the delta-to-DUT curve for both close-in and far-out regions.
3) What are the three most common causes of close-in “lift” (a low-offset hump)?
The most common causes are (1) drift/1/f noise in the discriminator or baseband chain, (2) AM-to-PM leakage from imperfect operating point (e.g., quadrature error), and (3) environmental effects such as cable microphonics or temperature gradients. Fast checks include mechanical stabilization, warm-up/temperature logging, and input-level sweeps to expose AM sensitivity.
4) Should spurs be included in jitter integration?
It depends on the reporting objective. If a spur is a real system tone that affects the application inside the integration band, include it (and document the spur threshold and method). If a spur is an instrument artifact or a removable setup coupling, exclude it and report a “clean” broadband jitter number. A strong practice is to publish both “spur-in” and “spur-out” results with a clear policy.
5) How should the window function be chosen to avoid underestimating spurs?
Spur accuracy depends on sidelobes and effective noise bandwidth (ENBW). Use a lower-sidelobe window when spur readout matters, and keep RBW/ENBW consistent when comparing runs. Confirm the choice with an injected tone: the tone’s center offset should stay fixed, and the measured amplitude should remain stable within tolerance across reasonable window/RBW combinations.
6) How should ADC full-scale be set to avoid false spurs or clipping?
Set gain so the baseband signal uses meaningful ADC range without approaching saturation. Too little amplitude wastes ENOB and raises the effective floor; too much causes clipping that creates wide skirts and fake spurs. A quick method is a gain sweep: true DUT noise trends smoothly, while clipping artifacts change abruptly with small headroom adjustments. Always log gain, headroom, and any limiter/overload flags.
7) How can AM-to-PM leakage be verified quickly?
AM-to-PM is likely when the measured “phase noise” changes strongly with input level or attenuation while the DUT is unchanged. Sweep input power across a safe range and watch whether close-in behavior or specific spurs scale with amplitude. Re-center the discriminator operating point (e.g., quadrature) and repeat. If available, inject controlled AM and confirm whether it appears as phase noise at the expected offsets.
8) What is the acceptance criterion for a channel-swap self-test?
After swapping CH-A and CH-B (and key cables if practical), the main L(f) shape should remain consistent and the spur pattern should not “follow” a channel. Channel-following spurs indicate instrument path issues or coupling. Acceptance should be stated as an overlay delta band (ΔSwap) within a defined tolerance over a specified offset range, with the exact ChannelMap and ConnectionMap recorded.
9) What are typical fingerprints of LO/reference contamination?
Common fingerprints include fixed-offset spurs that remain locked across runs, comb spurs spaced by reference/divider or switcher activity, and cross-correlation that refuses to improve with averaging (because the contamination is correlated). Fast checks include isolating LO distribution, separating references between channels, cleaning supplies, and rerouting grounds. Always log LORef topology and any shared paths.
10) What errors occur when RBW/FFT bin width is too narrow or too wide?
Too-narrow bins produce high variance and slow convergence, so the floor can look unstable and drifting spurs may be misread. Too-wide bins increase leakage, smearing tones into the floor and biasing spur and close-in estimates. Choose RBW based on the offset region and expected spur density, then validate with an injected tone and a second “reasonable” RBW/window pair to confirm stability of conclusions.
11) When is a delay-line discriminator a good choice (by offset region)?
A delay-line discriminator is strongest where the delay converts small phase changes into a clean baseband signal with good sensitivity and bandwidth, typically benefiting mid-to-far offsets depending on carrier frequency and delay stability. Close-in offsets are often limited by drift, microphonics, and operating-point sensitivity. Always calibrate sensitivity (V/rad), control temperature/mechanics of the delay path, and confirm residual floor before trusting very low-offset results.
12) What is the minimum field set for a deliverable phase-noise test report?
Minimum fields should make the run reproducible: SetupID, DUT_ID, discriminator type, LO/ref topology, ChannelMap/ConnectionMap, AvgN and averaging type, RBW, window, Fs and anti-alias configuration, temperature and warm-up time, calibration version/date and sensitivity (V/rad), residual floor curve, spur policy, integration band, pass/fail thresholds, and raw data file + hash. This transforms a plot into an auditable deliverable.
F12 — FAQ coverage map for phase-noise measurement credibility Three-lane map showing how the 12 FAQs cover: (A) credibility and calibration, (B) settings and DSP, and (C) troubleshooting and deliverables. FAQ coverage map A) Credibility B) Settings & DSP C) Delivery & Fix Residual floor: Q2 Channel swap: Q8 Cross-corr limits: Q1 Window/RBW: Q5, Q10 ADC headroom: Q6 Delay-line fit: Q11 Close-in hump: Q3 AM-to-PM: Q7 LO contamination: Q9 Jitter policy: Q4 Report fields: Q12 Use the map to link each FAQ to the exact evidence: residual floor, calibration scale, settings, and traceable records.
F12 — FAQ coverage map: credibility, DSP settings, and deliverable/reporting topics covered by the 12 questions.
icnavigator

ICNavigator helps engineers find verified ICs, alternatives, and fast quotes.

Explore
Categories
  • Power Management
  • MCU
  • Automotive ICs
  • Sensors
  • Interface & Transceivers
  • Analog Front-End
Get in Touch
  • Email: hello@icnavigator.com
  • WeChat/WhatsApp: +86-xxxx

© 2025 ICNavigator. All rights reserved.