123 Main Street, New York, NY 10001

Submit Your BOM

Spurs & Masks: Fractional-N Spurs and DDS Images

Spurs & Masks: Fractional-N Spurs and DDS Images

← Back to:Reference Oscillators & Timing

Spurs are discrete spectral lines that often fail system masks long before broadband noise looks “bad.” This page shows how to identify spur families, measure them correctly, and apply a mask-driven debug and mitigation workflow that closes with clear pass/fail criteria.

Scope & definitions: spur vs noise, mask vs spec

This section locks the vocabulary and measurement “units of truth” used throughout the page. The goal is repeatable interpretation across instruments, labs, and product states—without drifting into phase-noise shape, RMS-jitter integration, or loop transfer theory.

Boundaries (prevent cross-page overlap)

In scope
  • Spur (discrete spectral line), image (mirror/folded families), and practical identification rules.
  • Mask and limit reading: “allowed maximum line/sideband” vs offset/band.
  • Reporting units: dBc, dBm, RBW/VBW, detector/averaging—what must be written into pass/fail criteria.
Out of scope (link only)
  • Phase-noise curve shapes (dBc/Hz), RMS jitter integration windows, and SNR-from-jitter math.
  • Jitter transfer / loop filter theory and stability derivations.

See: Phase Noise & Jitter and Loop Bandwidth.

Definitions that prevent misreads

Spur: a discrete line (deterministic energy) at a specific frequency or offset from the carrier. A spur typically has a narrow spectral footprint that does not behave like noise when RBW/averaging changes.

Noise floor: continuous spectral density (commonly in dBc/Hz). It rises/falls with instrument settings in predictable ways (RBW, averaging), unlike a true discrete line.

Mask (limit): the maximum allowed spur/sideband at a given offset or within a band. Masks may come from datasheets, system specs, RF line-up requirements, or test standards.

Measurement conventions (what “the number” actually means)

  • dBc is relative to the carrier level at the measurement point. Always record carrier frequency and power.
  • dBm is absolute power into 50 Ω (or stated impedance). Use it when comparing across carriers or across different nodes.
  • RBW/VBW and detector/averaging change displayed spur height and noise floor. A mask is incomplete unless it states these conditions.
  • RBW reconciliation: when comparing different RBWs, treat the goal as consistency (repeatable setup). Only use “equivalent density” conversions to avoid miscomparison—not to do jitter integration here.

Deliverables (reuse these everywhere)

Spur report fields (minimum)
  • Carrier: frequency, power level, node name.
  • Spur: absolute frequency and/or offset, amplitude (dBc), margin vs mask.
  • Instrument: RBW, VBW, detector, averaging, input attenuation/coupling.
  • State: fractional ratio / fPFD / DDS settings / SSC on-off / output standard.
  • Conditions: temperature, supply rails, load/termination.
Pass/fail sentence template

Pass criteria: “Spur ≤ X dBc at offset/band Y, measured with RBW=…, carrier=…, state=…, temperature=….”

Spur vs Noise vs Mask A spectrum view showing a carrier, discrete spurs, a continuous noise floor, and an overlaid mask limit line with offset labeling. Spectrum view (concept) Frequency offset → Level (dBc) ↑ Noise floor Carrier Spurs Mask limit Report with setup: RBW / VBW / Avg offset offset
A discrete spur is a line; the noise floor is continuous. A spur mask is a limit line that must be evaluated with a stated setup (RBW/VBW/averaging).

Why spurs matter: SFDR, EVM, and “jitter looks good but RF fails”

Deterministic spurs often dominate real-world failures because a single line can violate a mask even when the noise floor and RMS jitter metrics look excellent. In many systems, the worst discrete line sets the usable dynamic range more directly than a small change in broadband noise.

Typical failure modes driven by one line

ADC / DAC (wideband sampling)

SFDR is frequently limited by the largest discrete spur, not by the broadband floor. If one line lands in-band (or folds in-band), it can cap achievable resolution even with excellent RMS jitter.

RF LO / synthesizers

Spurs create discrete modulation sidebands that can degrade EVM/ACLR, worsen blocking performance, and leak into image regions. A clean noise floor does not compensate for a mask-violating line.

Clock trees & distribution

Periodic (deterministic) timing components often show up as sidebands. These can propagate through fanout, appear at multiple endpoints, and trigger mask failures even when random jitter budgets are met.

Guardrail: treat the failure correctly

  • If a single line fails a mask: continue with spur taxonomy, measurement setup, and root-cause tracing in the next sections.
  • If the noise floor is elevated: use the Phase Noise & Jitter page for density interpretation and jitter-integration workflows.
SFDR limited by one moving spur Two side-by-side conceptual spectra show the same noise floor. Board A has a higher spur, resulting in worse SFDR than Board B. Same noise floor, different max spur → different SFDR Board A Board B SFDR set by max spur SFDR set by max spur Example: max spur = −62 dBc → SFDR = 62 dB Example: max spur = −72 dBc → SFDR = 72 dB Frequency offset → Frequency offset →
A single deterministic line can cap SFDR and violate a mask even when the broadband floor is unchanged. Treat “one-line failures” as spur problems first.

Spur taxonomy & frequency math: where do lines show up?

The fastest way to debug discrete lines is to classify them by “spacing fingerprints” and by how they move when a single parameter changes. This section provides a practical taxonomy and a small set of frequency templates that map observed lines to likely families—without drifting into loop-transfer derivations or phase-noise integration.

Spur families (what the spacing often reveals)

Reference / PFD spur

Lines appear at ±k·fREF or ±k·fPFD offsets around the carrier. The key fingerprint is a stable “comb spacing” that changes when fREF or fPFD is changed.

Fractional spur

Discrete components related to fractional ratio patterns (ΣΔ sequence periodicity) often move with the fractional setting. A quick fingerprint is that changing the fraction shifts spur locations or significantly changes spur heights.

DDS images

Mirror families appear due to sampling/reconstruction; they are often symmetric around fs/2 or a Nyquist boundary and shift predictably when sample rate or output frequency changes (Nyquist zone choice).

Harmonics & mixing (including PSU coupling)

Harmonics (2f, 3f) and intermods show up at combinations of two or more tones. Coupled lines commonly align with fSW (buck frequency) and its multiples, or with digital clocks.

Frequency templates (enough math to predict locations)

Mixing / intermod template

fspur = | m·fA ± n·fB | (m, n are small integers). Use this to test candidates such as fREF, fPFD, fSW, or a nearby digital clock.

DDS image heuristic

Image families are commonly symmetric around fs/2 (or a Nyquist boundary). Changing sample rate or Nyquist zone shifts image positions predictably; use this symmetry as the primary fingerprint.

One-sweep confirmation (shrink the suspect list fast)

Spacing equals fPFD or fREF

Likely family: Reference/PFD spur. Quick confirm: change fPFD (or ref divider) and verify that the spacing changes accordingly.

Line moves with fractional ratio

Likely family: Fractional spur. Quick confirm: keep carrier constant, change only the fraction (or dither profile) and log spur movement/height change.

Mirror symmetry around fs/2

Likely family: DDS images. Quick confirm: change sample rate or Nyquist zone and verify that the image shifts as predicted by symmetry.

Line tracks fSW / digital activity

Likely family: Coupling / mixing. Quick confirm: change switching frequency, disable a noisy IO burst, or swap to a low-noise supply and compare amplitude.

Spur families cheat-sheet One frequency axis shows a carrier and different spur families marked with distinct shapes: REF/PFD comb, fractional spurs, DDS image mirrors around fs/2, and PSU coupling at fSW and harmonics. Spur families on one axis (fingerprints) Carrier REF / PFD FRAC fs/2 DDS IMG PSU Frequency axis (offset or absolute) → Legend: comb=REF/PFD · squares=FRAC · triangles=DDS images · circles=PSU coupling
Use spacing and movement fingerprints to classify lines quickly. Treat the result as a suspect list that will be confirmed by one controlled parameter sweep.

Masks & acceptance criteria: reading system masks without fooling yourself

A spur mask is only actionable when it is bound to a repeatable measurement setup and a clearly stated device state. This section turns “mask compliance” into a pass/fail contract that survives cross-lab comparisons and avoids false failures caused by overload, mismatches, or inconsistent RBW/averaging.

Boundaries (what is evaluated here)

  • In scope: clock/LO purity masks, system spur limits, and fixture/test masks used for production and validation.
  • Out of scope: full EMI compliance workflows and regulatory radiated testing. Avoid using an EMI receiver as a spur-compliance authority unless the setup is explicitly defined for spur measurement.

Common mask expressions (and what must be stated)

Offset-point limit

“Spur ≤ X dBc at offset Y.” Must state: carrier power, offset definition, RBW/VBW, detector/avg, and device state (mode/profile).

Band-limited limit

“Within band A–B, max spur ≤ X dBc” (or spur power ≤ X dBm). Must state: span/trace mode (avg vs peak-hold), input linearity margin, and termination.

State-bound limit

Limits that depend on output swing, dividers, SSC on/off, or configuration profiles. Must state: exact register state, output standard/termination, and temperature.

A pass/fail contract that survives lab-to-lab comparisons

  1. Limit: X dBc (or X dBm).
  2. Where: offset Y, or band A–B.
  3. How measured: RBW/VBW, detector, averaging/trace, input attenuation/coupling, and reference lock mode.
  4. State & conditions: mode/profile, carrier power, termination, temperature, supply condition.
Single-line template

Pass criteria: “Spur ≤ X dBc at offset/band Y, measured with RBW/VBW=…, detector=…, avg/trace=…, input atten=…, coupling=…, state=…, carrier=…, temp=….”

Anti-fooling checklist (common false failures)

Overload check: increase input attenuation (or add an external pad) and verify spur levels scale as expected; overload spurs often change nonlinearly.

RBW consistency: do not compare spur numbers across different RBWs unless the mask explicitly defines the conversion method.

Mismatch/leakage check: swap cables, add a pad, and ensure the measurement reference is stable; small setup changes should not create new “real” lines.

State binding: record the exact configuration (profile, dividers, SSC) when capturing evidence; a mask is invalid without state.

Mask overlay & pass/fail callout A spectrum plot includes a mask limit line. One spur exceeds the mask by 6 dB and is labeled as FAIL. A small setup card lists RBW, averaging, attenuation, and coupling. Mask overlay (concept) → pass/fail is a contract Carrier Mask FAIL spur exceeds mask +6 dB Setup (must be stated): RBW · AVG/Trace · Input Atten · Coupling · State · Temp Frequency offset →
A mask is only meaningful when tied to measurement setup and device state. Use margin-to-mask (e.g., −6 dB fail) as the primary reporting currency.

Fractional-N PLL spurs: root causes and fingerprints

Fractional-N spur behavior becomes debuggable when each discrete line is tied to an injection point and validated with a single-parameter sweep. This section focuses on spur-relevant mechanisms and fingerprints, while loop stability and transfer derivations belong to the Loop Bandwidth page.

Spur-relevant injection points (keep only the spur chain)

PFD/CP periodic error (ref leakage, CP ripple)

Charge-pump pulses and leakage can leave a periodic residue at the loop-filter / Vtune node. Periodic Vtune modulation converts to discrete sidebands at the output.

ΣΔ fractional pattern (repeat sequences)

Some fractional ratios create repeating control patterns. Repetition creates discrete components (lines), often moving or re-weighting when the fractional setting or dither profile changes.

Divider / multi-modulus switching glitches

Multi-modulus switching events can introduce periodic phase disturbance. Certain N / ratio regions show stronger lines because switching alignment becomes more repetitive.

Parasitic coupling into VCO control (digital / supply)

Digital clocks, reference traces, and supply ripple can inject fREF/fPFD-like tones into Vtune or VCO nodes. This often looks like a “PLL spur” but is dominated by board coupling paths.

Fingerprints (spacing → movement → sensitivity)

1) Spacing
  • Spacing ≈ fPFD (or fREF): prioritize PFD/CP leakage or ref injection.
  • Spacing not tied to fPFD and changes with fraction: prioritize ΣΔ fractional patterns.
2) Movement
  • Change fraction: spur frequency/height shifts indicate pattern-driven lines.
  • Change fPFD: spacing shifts indicate ref/PFD-related families.
  • Change N only: if location stays but height changes, suspect injection strength / suppression differences.
3) Sensitivity
  • Change CP current or CP settings: large amplitude change suggests CP ripple dominance.
  • Change loop BW setting: use as a fingerprint (no stability derivation here).
  • Change digital activity / supply: strong dependence suggests parasitic coupling paths.

Quick check SOP (single-variable sweeps + what to record)

Sweeps (one knob at a time)
  • Sweep fraction (keep carrier fixed) to detect ΣΔ-pattern spur movement.
  • Sweep fPFD (ref divider) to validate spacing families.
  • Sweep CP current to test CP ripple dominance.
  • Sweep loop BW setting to fingerprint suppression behavior (no stability analysis here).
Record (minimum fields)
  • State: N, fraction, fPFD, BW profile, CP current, dither profile, SSC on/off.
  • Spur: frequency (absolute/offset), amplitude (dBc), margin vs mask.
  • Setup: RBW/VBW, detector/avg, attenuation/coupling.
  • Trend: Δfrequency/Δsetting and Δamplitude/Δsetting (repeatability across runs).
Frac-N spur injection paths A PLL block diagram shows REF, PFD/CP, loop filter, VCO, and divider feedback. Arrows indicate spur injection paths from CP ripple, sigma-delta pattern, digital coupling, and supply noise into the Vtune/VCO nodes. Fractional-N PLL (concept) — spur injection paths REF PFD / CP fPFD Loop Filter Vtune VCO fOUT Divider N / frac Supply ripple fSW, harmonics Digital coupling ΣΔ pattern CP ripple Fingerprints: spacing ~ fPFD/fREF · moves with fraction · sensitive to CP/BW · reacts to supply/digital activity Validate: sweep fraction · sweep fPFD · sweep CP · sweep BW (single variable)
Treat spurs as injected periodic errors. Map each observed line to an injection path, then confirm with a controlled sweep (one knob at a time).

DDS images & spurs: Nyquist zones, truncation, DAC/RC filtering

DDS spurs become manageable when treated as controllable outcomes of frequency planning (Nyquist zone choice), word-length decisions (truncation behavior), and reconstruction filtering. This section keeps the discussion at the engineering-control level: where images land, what changes move them, and which knobs reduce them.

Control knobs (what can be chosen or configured)

Nyquist zone planning

Images often mirror around fs/2 (or a zone boundary). Selecting fs and the output region can push dominant images into filter stopbands.

Word length & truncation

Phase truncation and LUT quantization can create discrete truncation spurs. Higher effective word length and dither (when available) reduce deterministic lines.

Reconstruction filtering

Analog filtering defines what remains in-band. The goal is to keep the desired tone in the passband and place dominant image families in the stopband.

DAC nonlinearity (recognize, then isolate)

Harmonics and intermods that scale strongly with amplitude indicate nonlinearity-limited behavior. Use amplitude and load changes as fingerprints before deeper distortion analysis.

Typical DDS spur sources (and the fastest confirmations)

Truncation spur

Fingerprint: discrete lines persist near the tone and do not follow simple mirror symmetry. Quick confirm: increase effective word length (or enable dither) and compare spur drop.

Image foldover

Fingerprint: image families mirror around fs/2 and shift predictably when fs or fout changes. Quick confirm: change fs (or zone) and verify mirrored movement.

DAC nonlinearity

Fingerprint: harmonics (2f, 3f) and intermods scale strongly with amplitude. Quick confirm: reduce output level and check whether lines drop faster than the tone.

Engineering strategies (make images land where filters can kill them)

Frequency planning rule

Define the passband first (what must be clean), then choose fs / output region so the dominant image family falls in the stopband. Treat Nyquist zone selection as a placement tool.

Filter goal (spec-level)

State stopband targets at the dominant image locations (not only “generic attenuation”). Report compliance as margin-to-mask at the image frequencies with the measurement setup stated.

Word-length / dither knob

When truncation spurs dominate, prioritize higher effective word length and dither (if supported) before redesigning analog filters.

Clock isolation knob

Maintain clean clocking and isolation so that the DDS chain does not import ref/PSU lines that masquerade as DDS-origin spurs.

DDS spectrum across Nyquist zones A spectrum map shows 0 to fs with fs/2 marked. A main tone appears in the passband, while mirrored image spurs appear symmetrically around fs/2. Shaded regions indicate filter passband and stopband. DDS spectrum map (concept) — images across Nyquist zones Passband Stopband target 0 fs/2 fs Tone Image Trunc Frequency → Level ↑ Planning goal: keep Tone in passband · place Image family in stopband · verify margin-to-mask with stated setup Confirm: change fs/zone → images move symmetrically; change word length/dither → trunc spurs drop
Treat Nyquist zone selection and reconstruction filtering as “image placement tools.” Validate image families by changing fs/zone and observing symmetric movement around fs/2.

Coupling paths that create “mystery spurs”: supplies, grounds, digital IO, fanout

When the datasheet looks clean but a board shows extra lines, the fastest win is to treat each line as an injected periodic tone. This section focuses on injection paths and measurable fingerprints (not full routing rules, which belong to PCB Layout & Routing).

Four injection paths (each with probes + one isolation move)

A) Supplies SMPS ripple → VCO / output buffer rails

Fingerprint: lines align with fSW and harmonics; changing regulator mode/frequency or load changes spur amplitude or frequency.

Probe points
  • VCO / clock buffer rail at local decoupling (short ground spring).
  • Post-LDO output near the sensitive block (same reference point).
Isolation move

Power the same rail from a low-noise source (or shift fSW if supported) and compare spur margin-to-mask.

B) Grounds Return currents / ground bounce → edge modulation

Fingerprint: spurs correlate with bursty current events; measurement grounding changes the observed lines; common-mode shifts show up near clock outputs.

Probe points
  • Local sensitive ground vs system ground (short differential measurement).
  • Clock output common-mode (if differential, compare CM under load/burst).
Isolation move

Pause large-current subsystems or change return paths (single controlled change) and check if the suspect line follows.

C) Digital IO I²C/SPI/SEL/LOCK activity → near-field coupling

Fingerprint: lines align with polling periods or GPIO toggles; stopping bus activity removes the spur quickly; near-field hotspots appear along IO or control traces.

Probe points
  • Near-field scan at Vtune / REF routing region (hotspot mapping).
  • Near-field scan over I²C/SPI/control traces and their return paths.
Isolation move

Disable polling / freeze configuration writes / stop debug toggles and compare spur amplitude at the same analyzer settings.

D) Fanout Additive spurs from mux/crosspoint/fanout stages

Fingerprint: a line appears only after a distribution stage; switching outputs/modes changes the line; input looks cleaner than output at identical measurement setup.

Probe points
  • Measure before and after the stage (same RBW/avg/atten) to isolate additive content.
  • Stage supply rail and output common-mode under different channel enables.
Isolation move

Bypass the stage (or enable only one output) and compare spur margin-to-mask at the same offset frequency.

Fast decision matrix (what the first isolation move should be)

Line aligns with fSW / harmonics → swap/shift the regulator source or mode; confirm frequency/level follows.

Line appears with IO bursts → stop polling/toggling; confirm the spur disappears without changing analyzer settings.

Line changes with grounding → re-measure with controlled grounding/differential probing; confirm correlation with ground delta.

Line only after fanout → measure pre/post stage and bypass/limit outputs; confirm additive content.

Spur coupling “attack map” A board map is divided into Analog, Digital, and PSU islands. A clock chain runs through the analog island: Ref, Cleaner/PLL, Fanout, Endpoints. Arrows show supply ripple, ground bounce, digital IO coupling, and additive fanout spurs. Probe points are marked as TP1 to TP4. Spur coupling “attack map” (board-level concept) Analog island Digital island PSU island Ref Cleaner / PLL Fanout Endpoints (FPGA / ADC / SerDes) mask-sensitive TP1 TP2 TP3 TP4 fSW ripple GND bounce I²C / SPI additive Legend: yellow=supply · red=ground · pink=digital IO · blue=fanout · green dots=probe points First move: change one factor → confirm spur frequency/level follows (margin-to-mask)
The goal is not “perfect theory,” but a fast isolation move that makes a line follow its injection source (frequency or amplitude), proving the dominant coupling path.

Measurement & debug workflow: how to measure spurs without lying to yourself

Spur data becomes useful only when measurement settings, front-end linearity, and DUT state are controlled. This section provides an executable SOP: pick the right tool, state the mandatory parameters, and run a three-step “de-lie” check before calling pass/fail against a mask.

Tool boundaries (what each instrument is trusted for)

Spectrum analyzer

Best for discrete spurs and wide offsets. Always keep the front end in the linear region using attenuation and external pads when needed.

Phase-noise / downconversion

More reliable at very low offsets where analyzer phase noise and leakage can dominate. Use when near-carrier spur claims are contentious.

Oscilloscope FFT

Useful for quick scouting, but results depend on windowing, sampling clock quality, and dynamic range. Treat as directional evidence unless validated.

Must-state parameters (without these, a spur number is not comparable)

Analyzer setup
  • RBW / VBW
  • Detector (peak / RMS / sample) + averaging / trace mode
  • Span / center / marker offsets
  • Input attenuation + preamp on/off
  • AC/DC coupling + external pad / isolator details
  • Reference lock / external reference source state
DUT state
  • Profile / mode / dividers / fractional settings
  • Output standard + termination / load
  • SSC on/off + spread depth (if applicable)
  • Supply condition (rail source, load state) + temperature
  • Fanout enable map (which outputs active)

Three-step “de-lie” SOP (run before pass/fail)

Step 1 — RBW / averaging sanity

Change RBW and averaging in controlled increments and observe whether the line remains a stable discrete component. If the “spur” drifts with settings or disappears unpredictably, treat it as a measurement artifact until proven otherwise.

Step 2 — cable / coupling / pad swap

Swap coupling method (AC/DC), insert a known external pad, and change cables to rule out reflections, leakage, and ground loops. A real DUT spur should not vanish from small front-end configuration changes that preserve linearity.

Step 3 — linearity / overload test

Change input attenuation (or carrier level) and verify spur levels scale consistently. Nonlinear scaling or new lines appearing with reduced attenuation indicates analyzer overload or intermod products.

Repro template (minimum loop for a defensible conclusion)

  1. Lock DUT state (mode/profile/termination/SSC) and log it.
  2. Lock analyzer state (RBW/VBW/avg/atten/coupling/ref lock) and log it.
  3. Capture baseline marker table (frequency, amplitude, margin-to-mask).
  4. Change one knob (DUT or environment) and capture again.
  5. Call root cause only when frequency/level changes follow the knob predictably and repeatably.
Spur measurement setup (do / don’t) Left shows a recommended setup with DUT, external pad/attenuator, and spectrum analyzer with clear labels. Right shows a risky setup with warnings: overload, leakage, and ground loop icons. Spur measurement setup — do / don’t (concept) DO DUT External pad ATTEN Spectrum analyzer RBW / AVG / Ref lock Linear front end pad/atten · avoid overload · consistent settings DON’T DUT Analyzer no pad Overload false spurs / intermods Leakage setup-dependent lines Ground loop cable/earth artifacts Rule: state setup + run de-lie SOP → then compare to the mask (margin-to-mask)
A spur claim is only defensible when the analyzer front end is linear and the setup is fully stated. Run the de-lie SOP before pass/fail decisions.

Mitigation playbook: fixes for Frac-N, DDS, and coupling (decision-tree style)

Spur fixes become repeatable only when each action is selected by a fingerprint and closed by a pass criterion. This playbook starts with fast classification, then applies low-risk changes first, and finally validates improvement using a fixed measurement setup and a margin-to-mask target.

Start here (30-second classification)

Spacing ≈ fPFD / fREF → prioritize Frac-N / ref-leakage actions (fPFD, CP/LF sensitivity, dither profile).

Tracks with N / fractional ratio → prioritize fraction planning + dither (reduce periodic sequences).

Mirror around fs/2 or changes with zone → prioritize DDS image actions (zone, reconstruction filter, word length).

Aligns with fSW / IO burst → prioritize coupling isolation (swap rail, stop IO, bypass fanout, ground sanity).

Only after fanout/mux → verify additive spurs (pre/post measurement, bypass, enable-map).

Actions by category (apply low-risk changes first)

Frac-N ref / PFD families, fractional sequences
Low-risk knobs
  • Change fPFD (shift spur spacing; reduce sensitivity at target offsets).
  • Fraction planning (avoid ratios that create short-period repeat patterns).
  • Enable / tune dither (spread discrete energy; re-check mask margins).
Higher impact knobs
  • Charge-pump / loop-filter sensitivity tuning to reduce periodic injection into Vtune.
  • Loop bandwidth direction: widen to track better / narrow to attenuate certain injected families (no stability derivation here).
DDS images, truncation spurs, reconstruction
Low-risk knobs
  • Change Nyquist zone / fs to move images into stopband regions.
  • Avoid sensitive fOUT where images or harmonics fall into protected bands.
  • Enable dither (reduce discrete truncation lines, verify noise-floor trade).
Higher impact knobs
  • Increase word length (reduce truncation families).
  • Reconstruction filtering (RC/LC/active) to suppress image bands while preserving the passband.
  • Reduce digital coupling into DAC reference/clock domains (treat as coupling branch below).
Coupling supplies, grounds, IO, distribution stages
First prove the path
  • Swap rail source or shift regulator mode/fSW and check if the line follows.
  • Stop IO bursts (disable polling/toggles) and confirm spur removal.
  • Bypass fanout/mux or enable only one output and compare pre/post stage.
Then implement fixes
  • LDO / LC filtering at the injection-sensitive rail.
  • Domain isolation (separate rails/returns for noisy blocks).
  • Reduce toggling and keep control lines away from Vtune/REF/outputs.
  • Shielding / spacing where near-field scans show hotspots.

Verification template (attach to every mitigation action)

  • Measurement lock: RBW/VBW/Avg/Atten/Coupling/Ref lock unchanged.
  • Target line: frequency/offset + amplitude (dBc) + margin-to-mask.
  • Pass: spur drops ≥ X dB (or margin improves ≥ X dB) with the same setup.
  • Fingerprint cleared: the line no longer tracks the original driver (N/fPFD/fSW/IO tick).
  • Repeatability: reproduce the result at least 3 times under identical conditions.

Note: dither changes can convert discrete energy into a smoother floor. A “pass” requires improvement against the system mask, not just a prettier spectrum.

Spur mitigation decision flow Decision flow from observed spur fingerprints (spacing near fPFD, tracks with N, mirror around fs/2, aligns with fSW, only after fanout) into mitigation branches (Frac-N, DDS, Coupling) and ends with a pass criteria box (margin-to-mask improved and no tracking). Spur mitigation decision flow (fingerprint → action → pass) Input fingerprint spacing ~ fPFD (ref family) tracks with N/frac (sequence) mirror ~ fs/2 (image) aligns with fSW (supply/IO) only after fanout (additive) Branch Frac-N fPFD / dither / plan DDS zone / filter / bits Coupling rail / IO / fanout Actions (examples) change fPFD ratio planning enable dither change zone add filter swap rail / stop IO bypass fanout PASS: margin-to-mask improves ≥ X dB, line no longer tracks the driver, repeatable with fixed setup
The fastest route is fingerprint-first: classify the line, apply the lowest-risk knob, then declare “pass” only by margin-to-mask improvement and cleared tracking behavior.

Spur budgeting: from system mask to per-block allocation

A system can fail even when every individual device looks “within spec.” The reason is simple: the mask is enforced at the system output, where multiple blocks can contribute lines. Budgeting converts a top-level mask into per-block allocations and a closed-loop table that exposes which block has the least margin.

Step 1 — Bucket the system mask (turn curves into actionable checkpoints)

  • Define mask buckets by offset ranges or by the spec’s explicit offset points.
  • Attach measurement conditions to every bucket: RBW, averaging, detector, coupling, and reference state.
  • Reserve guardband for integration and uncertainty: target an internal margin of X dB below the system limit.

Step 2 — Allocate per block (source → cleaner → fanout → endpoint)

Budget objects: source / cleaner / fanout / end-device, plus an explicit bucket for board-level coupling. Allocate more margin to the stage most likely to generate the target spur family.

Engineering rules (conservative)
  • Same-frequency spur: assume worst-case coherent addition unless proven otherwise; keep guardband.
  • Different-frequency spurs: focus on the worst single line vs the strictest part of the mask.
  • Allocation is invalid without conditions: RBW/state/temperature must match measurement.

Step 3 — Close the loop with an allocation table (template fields)

Use the following fields as a fixed “table header.” Implement the table as card rows to avoid mobile horizontal overflow.

  • Mask bucket (offset / band)
  • Limit (dBc or dBm) @ RBW (+ instrument setup)
  • Block (source / cleaner / fanout / endpoint / coupling)
  • Allocated (dBc)
  • Measured (dBc)
  • Margin to system mask (dB)
  • Notes (fingerprint, same-frequency risk, state)
Minimum workflow
  1. Fill system mask buckets and guardband targets.
  2. Allocate per block (initial guess is fine).
  3. Measure each stage (pre/post) with fixed setup and backfill “Measured.”
  4. Attack the smallest-margin row using the H2-9 playbook.
Mask → allocation table → system spectrum Left: a mask curve with buckets. Middle: an allocation table for source, cleaner, fanout, endpoint. Right: a system spectrum with a worst spur approaching the mask. Arrows connect mask to allocations and allocations to system result. Budgeting flow: mask → allocations → system result System mask offset → limit near mid far Allocation (per block) near mid far source cleaner fanout endpoint System spectrum worst line mask Budgeting closes the loop: allocate → measure per stage → fix the smallest margin row (H2-9)
Convert the system mask into buckets, allocate per stage, then measure pre/post blocks with identical setup. The smallest-margin row tells exactly where mitigation effort should go.

Engineering checklist: bring-up, production test, and field logs

This section freezes a repeatable evidence package for spurs and masks: a bring-up baseline, a production gate recipe, and a field log bundle. Every failure should immediately map back to fixed measurement conditions and the mitigation decision flow.

A) Bring-up baseline package (must be captured once and kept unchanged)

1) Configuration snapshot (state that defines the spectrum)
  • PLL / Synth: N, fractional ratio, fPFD, reference source, dither on/off, profile ID.
  • Clock chain: cleaner profile, fanout enable map, mux/crosspoint route (pre/post points).
  • Output: standard (LVCMOS/LVDS/HCSL/LVPECL), swing/drive, termination mode, load state.
  • SSC: on/off, depth, rate (if applicable).
  • Thermal: ambient + board sensor location and value.
2) Measurement lock (conditions that must be identical for comparisons)
  • Instrument: RBW/VBW, detector, averaging, span, input attenuation, preamp state.
  • Interface: coupling (AC/DC), reference lock state, cable + attenuator identity.
  • Power state: rail source type, regulator mode, and any spread/skip modes.
3) Baseline captures (evidence that anchors future debug)
  • Wide span: system overview + obvious families.
  • Mask-critical bands: buckets where limits are tightest.
  • Worst line zoom: frequency/offset + amplitude + margin-to-mask.
  • Repeatability: same setup, 3 runs, consistent line height/frequency.
Bring-up measurement accessory BOM (examples; verify impedance, power rating, and frequency range)
Attenuation / isolation
  • Mini-Circuits VAT-10+, VAT-20+ (fixed attenuators)
  • Mini-Circuits BW-S10W2+ (DC block, where applicable)
  • Mini-Circuits ZFSC-2-1+ (2-way splitter/combiner, where applicable)
Low-noise rails (to eliminate rail-driven spurs)
  • Analog Devices (Linear Tech) LT3042, LT3045 (ultralow-noise LDO)
  • Analog Devices ADM7150 (low-noise RF LDO)
  • Texas Instruments TPS7A4700, TPS7A94 (low-noise LDO families)
Quick coupling checks
  • TekBox TBPS01 (near-field probe set, board hotspot scans)
  • Mini-Circuits RCDAT-18G-63 (RF attenuator example; select per band/power)
  • Mini-Circuits ZX60-83LN12+ (LNAs can help in some setups; ensure no overload)

Part numbers above are starting points to standardize setups; selection must be driven by frequency range, impedance, and linearity requirements of the test chain.

B) Production test gate (fixed recipe + golden board + automatic fallback)

Fixed recipe (do not let operators “tune the spectrum”)
  • Lock RBW / averaging / detector and store them as the station recipe.
  • Use margin-to-mask as the pass metric (not “looks clean”).
  • If a unit fails, run a short three-step anti-fake check (RBW change / overload sanity / cabling swap) before debug escalation.
  • Escalation path is fixed: measurement sanity → decision flow → budget row.
Golden board (station-to-station alignment)
  • Define a golden spectrum package: baseline captures + configuration hash.
  • Run golden board at every station shift change to detect drift in cables, attenuators, and instrument settings.
  • Reject results if the station cannot reproduce the golden margins within a fixed tolerance.
Production fixture building blocks (examples; select per band/connector standard)
  • Mini-Circuits VAT-10+ / VAT-20+ (standardize input level & protect analyzers)
  • Mini-Circuits ZFSC-2-1+ (split/route signals reproducibly)
  • Mini-Circuits BW-S10W2+ (DC block where needed)
  • Analog Devices ADM7150 or LT LT3045 (quiet supply option for A/B isolation tests)
  • TekBox TBPS01 (quick near-field hotspot confirmation)

Fixture BOMs should be frozen and versioned; swapping one attenuator or cable often changes spur visibility and invalidates comparisons.

C) Field logs bundle (collect evidence first, infer cause later)

Trend log (time series)
  • Temperature, input voltage, regulator mode, load state.
  • Worst spur per mask bucket: frequency/offset + amplitude + margin-to-mask.
  • Event markers: mode switches, reference switches, IO burst windows.
State dump (reproducibility)
  • Register snapshots for PLL/DDS/cleaner/fanout/mux.
  • Firmware/bitstream/config file version identifiers.
  • Clock chain routing map (which blocks are active).
Correlation tags (fingerprints)
  • Does the line track fPFD, N/frac, fs/2, fSW, or IO tick rate?
  • Does it appear only after a distribution stage (fanout/mux)?
  • Does it change with supply source / grounding / cabling?
Spur test gate checklist Stage-based checklist for Spur work: Bring-up, EVT, DVT, PVT, Field. Each stage requires a snapshot, locked measurement settings, and saved captures/logs. Failures loop back to measurement sanity, decision flow, and budget table. Spur test gate checklist (save evidence at every stage) Bring-up EVT DVT PVT Field Snapshot Measure lock Captures / logs ✅ state ✅ RBW ✅ pics ✅ hash ✅ fixed ✅ rpt ✅ sweep ✅ same ✅ data ✅ station ✅ gate ✅ csv FAIL loop: sanity check → decision flow → budget row update → re-test with fixed setup
Each stage must save the same three artifacts: a configuration snapshot, locked measurement settings, and captured evidence. Failures must follow the same loop to stay reproducible.

Applications & IC selection notes (mask-driven, not part-number-driven)

Selection should start from the system mask and the sensitive frequency map, then translate that into frequency planning, spur budgeting, and only then device capability requirements (fPFD range, dither/profile controls, output standards, and isolation hooks).

A) Endpoint priority map (what “matters first” depends on the endpoint)

RF LO / wideband RF
  • Constraint: the spur mask is often the strictest; a single discrete line can dominate performance.
  • Selection focus: low spurs across operating modes, controllable dither/profile, ref isolation, clean supply integration.
  • Verification: confirm worst-case line vs mask across temperature, power states, and output power levels.
Wideband ADC/DAC sampling
  • Constraint: SFDR can be set by one spur landing near a blocker or folding into the band of interest.
  • Selection focus: predictable spur families, frequency planning freedom (fPFD/fs), and low additive spurs in fanout stages.
  • Verification: map spur frequencies into sensitive FFT bins and worst-case mixing products.
SerDes / PCIe / high-speed clocking
  • Constraint: periodic components can interact with tolerance and SSC compatibility.
  • Selection focus: output standard compliance, controllable SSC behavior, and avoidance of deterministic lines near sensitive offsets.
  • Verification: validate with the protocol’s compliance expectations (details belong to the interface page).

B) Datasheet fields checklist (what to extract before comparing devices)

  • Spur families: reference spurs (±k·fREF / ±k·fPFD), fractional spurs, DDS images/harmonics (if applicable).
  • Operating state: which modes produce which lines (profiles, lock states, dividers, output format).
  • Configurability: fPFD range, dither controls, spur-reduction options, profile memory, bypass/holdover behavior.
  • Integration hooks: additive spurs in fanout/mux stages, supply isolation guidance, status/telemetry registers.

C) Mask-driven selection flow (the only reliable order)

  1. Collect system mask buckets (include measurement conditions).
  2. Plan frequencies to avoid spur landings in protected bands.
  3. Allocate spur budget per block (source/cleaner/fanout/endpoint + coupling bucket).
  4. Select architecture by capability (fPFD/dither/profile/output standard/isolation hooks).
  5. Verify margin-to-mask across worst-case states and temperature, with fixed setup and cleared fingerprints.

D) Reference part numbers (starting points only; verify package, suffix, and availability)

The list below exists to speed up datasheet lookup during architecture selection. Final selection must be driven by mask buckets, landing avoidance, and budget closure.

Clock cleaners / jitter attenuators (mask-driven profiles)
  • Analog Devices: AD9545, AD9548, AD9528, HMC7044
  • Silicon Labs: Si5345, Si5341, Si5338
  • Texas Instruments: LMK04832, LMK05318, LMK03328
  • Microchip (Microsemi/Zarlink families): ZL30739 (example class; confirm exact variant)
RF PLL synthesizers (Frac-N spurs are often the first limiter)
  • Analog Devices: ADF4351, ADF4371 (example families; validate band/output needs)
  • Analog Devices: HMC833 (example family; validate frequency plan)
DDS generators (images/truncation must be controlled by zone + filtering)
  • Analog Devices: AD9910, AD9959 (DDS families; confirm clocking and reconstruction needs)
Fanout / buffers (additive spurs and output standard compliance)
  • Texas Instruments: CDCLVP1216, LMK1C1104 (examples; confirm output standard)
  • Analog Devices: ADCLK954 (example; confirm jitter/additive specs)
Ultra-low-noise rails (spur visibility often tracks supply integrity)
  • Analog Devices (Linear Tech): LT3042, LT3045
  • Analog Devices: ADM7150
  • Texas Instruments: TPS7A4700, TPS7A94
Selection flow: mask-driven architecture choice Flow diagram: mask buckets feed frequency planning, which feeds budget allocation, which feeds device capability requirements (fPFD/dither/profile/output/isolation), then verification (margin-to-mask and cleared fingerprint). Selection flow (mask → plan → budget → capability → verify) Mask input near mid far Frequency plan avoid landing fPFD / fs Budget source cleaner fanout Device capability requirements fPFD range dither/profile output std isolation hooks additive spur control Verify margin-to-mask ≥ X dB cleared fingerprint fixed setup, repeatable
Use the same order every time: mask buckets first, then landing avoidance and budgeting, then capability requirements, and finally verification against margin-to-mask with cleared fingerprints.

Request a Quote

Accepted Formats

pdf, csv, xls, xlsx, zip

Attachment

Drag & drop files here or use the button below.

FAQs: Spurs & Masks (debug without expanding scope)

Each answer is intentionally short and executable. Use a single-variable check first, then apply the smallest fix, and close with a measurable pass criterion.

RMS jitter passes, but one spur fails the mask—what’s the first measurement sanity check?
Likely cause: Analyzer overload, leakage/mismatch, or RBW/averaging settings are creating an inflated spur number.
Quick check: Change only one knob: add +10 dB input attenuation or change RBW by 10×; re-read the same marker.
Fix: Freeze input level (attenuation + preamp off if needed), lock RBW/VBW/detector/averaging, and re-run with the same cable/attenuator chain.
Pass criteria: Spur amplitude is stable across sanity changes (typically ≤ 1–2 dB) and margin-to-mask ≥ X dB under the fixed recipe.
A spur spacing equals fREF—how to confirm reference leakage vs PSU coupling?
Likely cause: Either reference/PFD energy is leaking into the VCO/clock path, or a supply rail is AM/PM-modulating the output at a periodic rate that coincidentally matches.
Quick check: Single-variable test: change fREF (or switch reference source) while keeping output frequency constant; watch whether spur spacing moves 1:1 with fREF.
Fix: If it tracks fREF, improve reference isolation (routing/grounding) and reduce CP/ripple injection; if it tracks rail activity, isolate the supply (quiet LDO/LC) and clean the return path for the sensitive node.
Pass criteria: The suspect line no longer tracks fREF (or rail signature) and stays below mask with ≥ X dB margin across normal operating states.
Fractional spur moves when the fractional ratio changes—what single sweep proves it fastest?
Likely cause: ΣΔ fractional sequence periodicity (or divider/phase detector pattern) is producing deterministic spurs tied to the fractional ratio.
Quick check: Keep fPFD and output frequency fixed, sweep only the fractional ratio (small set of values) and log spur frequency + amplitude at each point.
Fix: Enable/adjust dithering, pick a ratio plan that avoids short repeating patterns, and/or adjust fPFD to move fractional spur families out of protected buckets.
Pass criteria: After the change, the worst fractional spur is reduced by ≥ Y dB and all mask buckets show ≥ X dB margin at worst-case states.
Spur drops when loop BW is widened—hiding it or truly reducing it? What’s the pass criterion?
Likely cause: Wider BW can attenuate certain injected components, but it can also raise noise floor and visually “bury” a line without improving true margin-to-mask.
Quick check: Change only BW/profile, then re-measure the same spur using a narrow RBW zoom; compare spur line height and local noise floor separately.
Fix: Accept BW change only if the line itself decreases; otherwise treat it as a measurement illusion and target the injection path (CP ripple, ref leakage, coupling).
Pass criteria: Spur line amplitude decreases by ≥ Y dB under the same RBW/recipe, and margin-to-mask ≥ X dB across temperature and power states.
Spurs appear only when I²C traffic runs—how to isolate IO coupling in 5 minutes?
Likely cause: Digital edge bursts are coupling (E-field/H-field or return-path) into a sensitive node (VCO control, ref pins, output buffer supply).
Quick check: Single-variable: stop I²C for 30–60 s (or reduce I²C speed by 10×) while holding all other states constant; watch the line appear/disappear or scale.
Fix: Add series damping on SCL/SDA (edge control), route IO away from sensitive islands, and ensure return paths do not cross analog reference regions; gate traffic during spur-critical measurements if allowed.
Pass criteria: Under worst-case traffic (highest toggle density), the spur remains below mask with ≥ X dB margin and remains repeatable (≤ 1–2 dB run-to-run).
DDS output shows images near fs/2—how to pick a Nyquist zone to push images out of band?
Likely cause: Sampling images fold around fs/2; truncation/DAC reconstruction effects place discrete components in predictable mirror locations.
Quick check: Single-variable: move the DDS output frequency within the same configuration; verify the image mirrors around fs/2 (symmetry check).
Fix: Choose a Nyquist zone such that dominant images land in the reconstruction filter stopband; adjust the analog low-pass/band-pass reconstruction filter accordingly (passband/stopband must match the chosen zone).
Pass criteria: All images/spurs in protected bands remain below mask with ≥ X dB margin, and the stopband attenuation meets the required suppression at the image frequencies.
A “mystery” spur tracks the buck switching frequency—what filter/ground test confirms root cause?
Likely cause: Rail ripple or ground bounce is modulating a sensitive node, imprinting fSW (and harmonics) as discrete spurs.
Quick check: Single-variable A/B: power the sensitive rail from a quiet source (temporary low-noise LDO feed) or change buck mode/frequency; confirm spur follows the rail signature.
Fix: Add targeted LC/LDO isolation for the sensitive rail, keep high di/dt loops away, and correct the return path so ripple current does not share reference with the sensitive node.
Pass criteria: Spur at fSW (and key harmonics) drops by ≥ Y dB and remains below mask with ≥ X dB margin across load states.
Two boards show different spur levels with same BOM—what layout/return-path measurement catches it?
Likely cause: Return-path discontinuity (slots/splits), unintended coupling proximity, or different via/stackup parasitics are changing injection gain.
Quick check: Single-variable: run a near-field scan over the sensitive island and compare hotspots; verify whether the spur changes when probing/shorting a suspected return bridge location (controlled experiment).
Fix: Restore continuous reference planes under sensitive routes, add explicit return stitching where needed, and increase separation from aggressors (IO, buck nodes, fanout digital blocks).
Pass criteria: Board-to-board spur difference shrinks to ≤ 1–2 dB under identical recipe, and both boards meet mask with ≥ X dB margin.
Why does adding an attenuator at the analyzer input change spur levels—overload or mismatch?
Likely cause: The analyzer is operating in a non-linear region (overload) or reflections/mismatch are creating frequency-dependent ripple that looks like spurs.
Quick check: Add attenuation in steps (e.g., +10 dB then +20 dB) while keeping all else constant; if spurs collapse disproportionately vs carrier, overload is likely.
Fix: Keep analyzer input comfortably below compression, standardize the attenuator/cable chain, and use proper impedance control (DC block/isolator only where appropriate).
Pass criteria: With the standardized chain, spur dBc remains stable (≤ 1–2 dB) across small input-level shifts and meets mask with ≥ X dB margin.
How to set RBW/VBW so the spur number is comparable across labs?
Likely cause: RBW/VBW, detector type, and averaging change the reported peak value and noise contribution, making “the spur number” non-portable.
Quick check: Align recipes: use the same RBW/VBW, detector, averaging, span, and input level; export markers and screenshots with the full setup line.
Fix: Define a lab-to-lab spur recipe: “dBc @ RBW = ___, VBW = ___, detector = ___, avg = ___, attenuation = ___”; version-control it as a test spec.
Pass criteria: Two labs measuring the same DUT match spur amplitude within ≤ 2 dB (or the project-defined tolerance) and reach the same pass/fail call against the mask.
Spur is within spec at room temp but fails hot—what to log (registers + power + temperature) first?
Likely cause: Temperature changes shift coupling gain, VCO control sensitivity, rail ripple, or internal mode decisions, pushing a deterministic line above mask.
Quick check: Two-point sweep: room vs hot, with an identical measurement recipe; capture (1) register snapshot, (2) rail voltage + ripple signature, (3) board temp at the same sensor point.
Fix: Identify the tracking variable (ratio, fPFD, fSW, IO bursts) and apply the smallest mitigation (dither/profile, isolation, routing/return fix), then re-validate across temperature.
Pass criteria: At the worst-case temperature, margin-to-mask ≥ X dB and the logged evidence is sufficient to reproduce the failure on the bench.
When a fanout buffer adds discrete spurs—how to tell buffer-intrinsic spurs from upstream leakage?
Likely cause: The buffer/mux/crosspoint contributes additive discrete components, or it is amplifying/leaking upstream reference/PLL spurs into the outputs.
Quick check: Single-variable segmentation: bypass the buffer (or route around it) while keeping the source identical; compare spur presence and amplitude at the same measurement point.
Fix: If buffer-intrinsic, improve its supply isolation and select a lower-additive-spur stage; if upstream leakage, fix the source injection (ref/CP ripple/coupling) before distribution.
Pass criteria: The “insert/remove buffer” A/B test cleanly assigns ownership, and the final chain meets the system mask with ≥ X dB margin in worst-case routing modes.

Structured data (FAQ JSON-LD) is included below for SEO.

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.