123 Main Street, New York, NY 10001

Submit Your BOM

Distortion & Dynamic Range in DACs (THD/SFDR/SNDR)

Distortion & Dynamic Range in DACs (THD/SFDR/SNDR)

← Back to:Digital-to-Analog Converters (DACs)

Distortion and dynamic range are only “real” when measurement conditions are fixed and the worst spur is planned, searched, and verified. This page turns THD/SFDR/SNDR from datasheet numbers into a repeatable workflow: define zones, prevent fake spurs, assign root-cause ownership, and validate performance with production-ready tests.

What this page solves: distortion vs dynamic range for DACs

Distortion metrics and dynamic-range metrics answer different questions. Distortion describes where a clean tone’s energy is pushed into harmonics and spurs (purity and “largest-spur” failure risk), while dynamic range describes how far usable signals sit above the noise floor (weak-signal detail and low-level cleanliness). This page turns THD/SFDR/SNDR/DR into a practical decision framework and a repeatable measurement vocabulary, so datasheet numbers become comparable and test results become actionable.

What this section answers
  • What THD, SFDR, SNDR (SINAD), and DR each penalize in real systems.
  • Why audio and RF prioritize different failure modes even on the same DAC.
  • When a great SNDR can still hide a single catastrophic spur (SFDR failure).
What this page does NOT cover
  • Architecture internals (string, R-2R, CS-DAC, ΣΔ): see the Architecture subpages.
  • JESD timing/alignment and phase coherence: see Interfaces & Synchronization.
  • Clock phase-noise/jitter budgeting math: see Clocking & Phase Noise.
Deliverable: metric → risk → where it matters
Metric What it penalizes What it can hide Most sensitive to Where it matters most Decision trap
THD Harmonic energy caused by nonlinearity (low-order often dominates the “visible” harm). Non-harmonic spurs (images, interference, periodic errors) and an isolated worst spur. Output swing/back-off, driver/load linearity, measurement BW and harmonic limit. Audio tonal purity, clean sine generation, benign narrowband waveforms. Comparing THD without stating harmonic order range and integration BW.
SFDR The largest spur relative to the carrier (or full-scale), often a one-number “mask risk”. Broadband noise floor (a high SNDR can exist while SFDR is failing). Spur taxonomy (harmonic vs non-harmonic), interference coupling, bandwidth of observation. RF transmission, wideband synthesis, any system with “largest spur must be below X”. Mixing dBc and dBFS, or not stating whether DC/harmonics are excluded.
SNDR (SINAD) Total quality: carrier versus (noise + distortion) within a defined bandwidth. The root cause: it does not reveal whether the limit is noise, harmonics, or one spur. Integration BW, window/FFT settings, and whether DC/spurs are excluded. General-purpose waveform quality, audio/IF chain comparisons, ENOB-style ranking. Treating SNDR as a spur guarantee; a single spur can be catastrophic but barely move SNDR.
DR Separation between maximum usable level and the noise floor (often with stated BW/weighting). Largest spurs and harmonic structure (DR can look great while SFDR fails). Bandwidth/weighting, reference noise, and output noise shaping (if used). Audio low-level detail, precision bias/slow waveforms, background cleanliness. DR quoted without bandwidth/weighting; cannot be compared across reports.
Rule of thumb THD explains harmonic cleanliness; SFDR catches “one bad spur”. Use SFDR when any single spur can violate a mask or fall into a sensitive band. Always state the test conditions; otherwise the number is not portable. Audio: harmonic structure + noise floor. RF: worst spur + where it lands. “SNDR-only” ranking can miss a single spur that breaks the system.
Distortion versus dynamic range: where metrics apply in a DAC signal chain Block diagram from digital code to DAC, output network, filter/load and measurement, with a small spectrum view labeling THD, SFDR and SNDR regions. Signal chain and metric “focus areas” Digital code DAC core linearity / glitch Output network driver / load Filter image control Measure FFT / BW Spectrum view (conceptual) Frequency Amplitude Fund. Spur THD: harmonics SFDR: largest spur SNDR: noise + dist BW

Metric definitions that do not lie: THD, SFDR, SNDR, DR (math-lite rules)

A distortion or dynamic-range number is only comparable when its test conditions are explicit. The same label (SFDR, THD, DR) is commonly reported with different reference units and exclusion rules (DC removed or not, harmonics excluded or not, bandwidth integrated or not). This section provides short, engineering-grade definitions and a strict “declare-it-or-ignore-it” checklist, so results can be verified and disputes can be avoided.

Definitions (math-lite) and what each metric really means
THD / THD+N
  • THD: harmonic power relative to the fundamental.
  • THD+N: harmonics plus integrated noise within a stated BW.
  • Must declare harmonic limit (e.g., 2–5th, 2–10th, to Nyquist).
SFDR
  • SFDR: difference between the carrier and the largest spur in the observed band.
  • Spur can be harmonic or non-harmonic depending on stated exclusion rules.
  • Must declare reference unit (dBc or dBFS) and exclusion rules (DC/harmonics).
SNDR (SINAD) / DR
  • SNDR: carrier versus (noise + distortion) integrated over a stated BW.
  • DR: separation between maximum usable level and the noise floor (BW/weighting must be stated).
  • SNDR ranks total quality; SFDR protects against a single worst spur.
Variants & traps that break comparisons
SFDR reporting traps
  • dBc vs dBFS: a unit mismatch can shift meaning without changing the number.
  • Excluded bins: DC removed? harmonics excluded? a “spur-free” claim depends on rules.
  • Observed band: a narrow band can hide images/spurs that appear elsewhere.
THD / SNDR / DR traps
  • THD without a harmonic limit is not reproducible.
  • SNDR without a stated BW/window/FFT rule is not portable.
  • DR without BW/weighting is not comparable across reports.
Must-declare fields (declare it or ignore the number)
  • fOUT (tone frequency) and tone placement (single-tone / multi-tone).
  • AOUT (level): dBFS or Vpp/Irms; state back-off if used.
  • FS / update rate and any interpolation/OSR setting (if applicable).
  • Output network & load: termination, transformer/driver, differential/single-ended.
  • Filtering: analog reconstruction filter + instrument front-end/RBW.
  • Integration BW: start/stop frequency or “to Nyquist”; weighting if used.
  • Reference unit & exclusions: dBc/dBFS; DC removed? harmonics excluded? bins ignored?
Reporting template (copy-ready)

Metric @ fOUT=?, AOUT=?, FS=?, BW=?, load=?, filter/RBW=?, unit=?, exclusions=?.

How to read one spectrum into THD / SFDR / SNDR / DR (repeatable steps)
  1. Lock the reference unit (dBFS or dBc) and document whether DC is removed.
  2. Identify the fundamental bin(s) and confirm no clipping or analyzer overload.
  3. Mark the harmonics within the declared harmonic limit; sum their power for THD.
  4. Scan the observed band for the largest spur using the declared exclusion rules; compute SFDR.
  5. Integrate noise + distortion over the stated BW (excluding the fundamental); compute SNDR.
  6. Determine the noise floor in the same BW and state DR using the same BW/weighting rules.
Standardized spectrum annotation for THD, SFDR, SNDR and DR A simplified spectrum with labeled fundamental, harmonics, largest spur and integrated noise bandwidth to show where each metric is computed. One spectrum, four metrics (standardized annotation) Frequency Amplitude Fund. Spur BW integration region THD: sum harmonics SFDR: worst spur SNDR/DR: set BW Always declare unit (dBc/dBFS) and exclusions (DC/harmonics) before comparing numbers.

Measurement setup: how to test spectrum purity without creating fake spurs

Spectrum purity is often limited by the test chain rather than by the DAC itself. A reliable setup must (1) keep every block in a linear region, (2) prevent leakage and coupling from turning into discrete tones, and (3) use FFT settings that do not manufacture “spurs” through leakage and poor resolution. This section provides a minimum credible chain and a one-page SOP to produce results that can be compared across boards, labs, and vendors.

Minimum credible chain (what must be controlled)
  • Clock / reference: isolate routing, avoid direct feedthrough into analog nodes.
  • DAC output + load: keep driver/load in linear region; avoid analyzer overdrive.
  • Attenuation + simple filtering: protect analyzer linearity; reduce out-of-band energy.
  • Analyzer / FFT: consistent BW, window, points, and averaging rules.
  • Cabling + grounding: eliminate ground loops and EMI pickup before blaming the DAC.
What this section does NOT cover
  • Clock phase-noise/jitter math and budgeting: see Clocking & Phase Noise.
  • Reconstruction filter design details: see Reconstruction / Anti-Image Filter.
  • DAC architecture internals: see the Architecture subpages.
One-page measurement SOP (wire, set, verify, troubleshoot)
  1. Wire the chain: Clock → DAC → Output network → fixed attenuator → (optional simple filter) → Analyzer.
  2. Start with headroom: set output to a safe level and increase only after verifying analyzer linearity.
  3. Lock the reference unit: document dBc vs dBFS and whether DC is removed; keep this constant.
  4. Set observation BW: define the band in which SFDR/SNDR are computed; do not mix bands across comparisons.
  5. Choose FFT settings: use a stable window and point count; verify that “skirts” are not mistaken as spurs.
  6. Verify front-end linearity: add attenuation and re-check; true DUT spurs remain, analyzer IMD changes strongly.
  7. Check coupling: move cables, change grounding, and isolate clock routing to detect pickup/leakage tones.
  8. Record the conditions: fOUT, AOUT, FS, load, filter/RBW, BW, unit, and exclusions.
Fast “real spur vs fake spur” triage (3 checks)
Check If the spur moves / changes like this Most likely class Next action
Move the tone (fOUT) Spur shifts with the tone (harmonic spacing or consistent offset). DUT-related distortion or tone-correlated artifact Sweep amplitude and load; check driver linearity and output headroom.
Change level (AOUT) Spur scales nonlinearly with level, or appears as symmetric side products (IMD-like). Analyzer front-end IMD or overloaded driver/load Add fixed attenuation; verify analyzer linearity; keep DUT level in a safe region.
Change BW / avg Spur magnitude/shape changes dramatically with FFT points, window, RBW, or averaging. FFT leakage / resolution artifact or processing effect Lock a standard window + point count; re-check with coherent sampling conditions.
Minimum credible spectrum measurement chain and spur triage flow Top block diagram from clock to DAC to output network through attenuator and filter into analyzer; bottom flow chart uses three checks to classify spurs as DUT, setup, or coupling. Measure without fake spurs: chain + triage Minimum credible chain Clock DAC Output load / driver Atten linearity Analyzer FFT / BW Filter 3-step spur triage Spur found Move tone? fOUT sweep Change level? AOUT sweep Change BW? FFT / avg Likely DUT distortion / images Likely setup overload / FFT Likely coupling leakage / EMI

Spur taxonomy: harmonics, images, clock feedthrough, and code-related spurs

A single spur can break a mask or contaminate a sensitive band even when SNDR looks strong. The fastest way to debug is to classify a spur by how it lands on the frequency axis and how it moves when fOUT, FS, or level changes. This section provides a clean taxonomy and a feature-to-cause table that narrows the suspect list before any deep architecture analysis.

Four classes (how to recognize them fast)
  • Harmonics: appear at k·fOUT (nonlinearity signature).
  • Images: appear by sampling/mapping rules and move strongly with FS.
  • Feedthrough: fixed to clock/reference/leakage frequencies.
  • Periodic/code-related: fixed spur spacing (Δf) reveals a periodic error source.
What this section does NOT cover
  • Reconstruction filter design: see Reconstruction / Anti-Image Filter.
  • Clock noise derivations: see Clocking & Phase Noise.
  • Architecture-specific mechanisms: see the Architecture subpages.
Deliverable: spectrum feature → likely root cause (fast narrowing)
Observed feature Likely class Most likely path Fast check Next page
Spurs at 2·fOUT, 3·fOUT, 5·fOUT (integer multiples) Harmonics Nonlinearity in DAC/driver/load, compression, transformer core effects Sweep AOUT by 6–12 dB; change load; keep analyzer headroom constant Output Forms & Front-End
Spurs that “move” when FS changes, following sampling/image mapping Images Insufficient filtering, mapping around Nyquist, wideband energy leaking into observed band Change FS; verify spur location changes predictably; add simple reconstruction filtering Reconstruction / Anti-Image Filter
Fixed spur at clock/reference-related frequency (does not follow fOUT) Feedthrough Clock leakage, reference coupling, digital edge injection into analog output/return paths Re-route/disable clock source; change shielding/grounding; watch spur stability Clocking & Phase Noise
Spurs at regular spacing (Δf), forming a “comb” Periodic / code-related Periodic error source: supply ripple, clock divider interaction, update pattern, digital periodicity Identify Δf; look for matching periodic sources; randomize pattern to see if comb collapses Waveform & Spur Control
Spur stays fixed while cables move / grounding changes Coupling Ground loop, EMI pickup, enclosure resonance, external interference Shorten/replace cables; change grounding; add shielding/ferrites; check repeatability Supply & Grounding
Spur taxonomy: four spectrum patterns for fast classification A 2×2 grid of simplified spectra showing harmonics, images, feedthrough/leakage, and periodic comb spurs, each labeled with minimal markers. Four spur patterns (fast recognition) Harmonics Fund. 2nd 3rd Images Fund. Image moves with FS Feedthrough Fund. Clk fixed frequency Periodic / code-related Fund. Δf spacing

Root causes inside and outside the DAC: which error creates which spur

A spectrum is a map of error injection points. The fastest way to improve SFDR/SNDR is not guessing architectures, but recognizing whether a problem looks like harmonics, a single worst spur, wideband energy, or symmetrical sidebands, then checking the most probable paths first. This section provides a practical mapping from “what appears in the FFT” to “what to inspect in the signal chain”.

Four spectrum symptoms (the engineering shortcuts)
  • Harmonics: integer multiples of fOUT (linearity-limited behavior).
  • Worst spur: one tone dominates SFDR (leakage, periodic errors, mixing).
  • Wideband: broadband rise or “grass thickening” (glitch/settling/overload).
  • Sidebands: symmetric tones around fOUT (AM/FM/PM modulation from ripple/leakage).
What this section does NOT cover
  • Architecture circuit implementation: see String / R-2R / CS / ΣΔ pages.
  • Clock phase-noise math: see Clocking & Phase Noise.
  • Filter design details: see Reconstruction / Anti-Image Filter.
Error → spectrum signature mapping (what the FFT is trying to tell)
Error / injection point Spectrum signature Most likely place Fast check Fix direction (no implementation details)
Static nonlinearity (INL-related) Harmonics at k·fOUT; usually stronger low-order (2nd/3rd) DAC core linearity, code transition regions, calibration state Sweep AOUT; harmonics scale strongly with level and stay locked to k·fOUT Operate in a cleaner region (back-off), avoid overload elsewhere, enable available linearity features
Update transient (glitch / major carry) Wideband energy; elevated “grass”; sometimes discrete spurs from repeatable transitions DAC switching moment, output network ringing, measurement BW too wide Compare different code steps; watch wideband changes; add simple post filtering to check sensitivity Reduce the transition stress (pattern/step), shape energy with appropriate bandwidth control
Driver / load enters nonlinearity Harmonics and IMD products; SFDR collapses at higher levels; spur pattern can look “IMD-like” Output buffer, external op-amp, transformer, termination network, heavy capacitive load Add headroom (attenuation/back-off) and re-test; change load/termination; IMD changes strongly Match output mode and impedance; keep driver in linear region; avoid marginal load conditions
Supply / reference ripple (AM modulation) Symmetric sidebands at fOUT ± fM; spur spacing equals ripple/modulation frequency Reference node, analog supply rail, shared regulators, ground return impedance Correlate sideband spacing with known ripple sources; change supply filtering and observe change Reduce ripple at the injection node; isolate noisy loads; improve return paths
Clock feedthrough / leakage Fixed tones tied to clock/reference; may not move with fOUT; can dominate SFDR Clock routing near analog output, shared ground/return, coupling into output network Disable/replace clock source; re-route probe/cable; spur remains at the same frequency Improve isolation and shielding; separate returns; reduce coupling paths
Periodic digital activity / pattern periodicity Comb spurs with constant Δf spacing; worst spur may be non-harmonic Periodic update patterns, digital coupling, PSU beat frequencies, environmental interferers Measure Δf and match it to periodic sources; randomize pattern and re-check spur collapse Break periodicity (randomization/dither); isolate periodic aggressors; verify with A/B measurement
Priority triage: when SFDR drops, check these 3 paths first
Case A — single worst spur dominates
  1. Leakage/coupling (clock/ref/EMI) — fixed-frequency spur that ignores fOUT moves.
  2. Periodic aggressor — comb or stable spacing indicates a periodic source.
  3. Measurement chain IMD — spur reacts strongly to attenuation/headroom changes.
Case B — harmonics are the limiter
  1. Driver/load linearity — most common; changes with load/termination and level.
  2. Amplitude headroom — back-off often improves high-order distortion quickly.
  3. DUT linearity features — confirm calibration/segment options are correctly enabled.
Case C — sidebands around the carrier
  1. Supply/reference ripple — sideband spacing equals the modulation frequency.
  2. Clock path coupling — changes with routing/shielding; often fixed to a clock-related tone.
  3. Ground return impedance — return-path modulation can create symmetric products.
Error injection points along the DAC signal chain and their spectrum signatures Block diagram from digital input through DAC core, update/glitch, driver, filter/load to measurement, with small tags marking likely signatures: harmonics, wideband, IMD, sidebands, spur. Error injection map (where spurs are born) Digital DAC core INL / linearity Update glitch / carry Driver load / IMD Filter load Measurement Harmonics Wideband IMD Supply / Ref Sidebands Clock leak Spur Classify first (harmonics / worst spur / wideband / sidebands), then inspect the highest-probability injection points.

Practical control knobs: how to improve THD, SFDR, and SNDR in real designs

Improvements must be goal-driven. SFDR is usually limited by one worst spur and coupling paths, while SNDR is limited by the total (noise + distortion) within bandwidth. The same change can improve one metric and worsen another, so each knob below is paired with expected benefit, side effects, and a verification method that keeps the measurement conditions consistent.

Deliverable: change → expected improvement → side effect → how to verify
Change (knob) Expected improvement Side effect / trade-off Verification method
Back-off output level (e.g., 3–12 dB) Lower harmonics and IMD; worst spur often drops faster than the noise floor Reduced output power; SNDR can become noise-limited in-band Sweep AOUT with fixed analyzer headroom and fixed BW; compare SFDR vs SNDR trends
Add fixed attenuation before analyzer (e.g., 6–20 dB) Removes analyzer IMD and overload artifacts; stabilizes repeatability May raise displayed noise floor due to reduced signal at the instrument input Repeat the same test with/without attenuation while keeping DUT output constant
Match load / termination to the output mode Improves SFDR by keeping the driver in a linear region and reducing reflections/ringing Output swing may reduce; power and thermal stress can increase in the termination A/B test load values; compare harmonics and IMD; keep the same measurement BW
Reduce coupling paths (clock/ref/ground separation) Improves worst-spur SFDR by removing fixed-frequency leakage tones Layout changes may be required; probing and cabling becomes more sensitive Move tone and re-check fixed spur stability; repeat with different grounding/cable routing
Add simple reconstruction filtering (band control) Reduces images and wideband energy in the observed band; SFDR can improve where it matters Group delay and passband ripple can increase; bandwidth and phase response become constraints Compare SFDR before/after using the same observation band and the same exclusion rules
Dither / randomization (break periodicity) Collapses comb spurs and reduces worst spur magnitude (SFDR), making artifacts less discrete Noise floor can rise; SNDR may trade down even if SFDR improves Track worst spur and integrated noise separately; compare both SFDR and SNDR under identical BW
Update mode selection (RTZ vs NRZ, behavior-level) Can reshape image/spur behavior and change where energy lands in-band Bandwidth and power can change; may require different filtering assumptions A/B compare spectra using the same observation band; confirm changes are not measurement artifacts
Decision flow: choose knobs to improve SFDR or SNDR, then verify A goal-driven flow chart that starts with SFDR vs SNDR, routes to knob groups (amplitude, load, filter, code, mode), and ends with verification steps. Improve the right metric: choose knobs, then verify Define goal SFDR or SNDR Goal: SFDR Goal: SNDR Coupling Code tricks Back-off Load / mode Filtering RTZ / NRZ Verify (keep conditions fixed) FFT compare Sweep AOUT Load A/B Two-tone Report: fOUT, AOUT, FS, BW, load, filter/RBW, unit, and exclusions for reproducible SFDR/SNDR.

Intermodulation & multitone: when THD is irrelevant but SFDR kills you

Many wideband and RF systems are not single-tone. In multitone conditions, intermodulation (IMD) creates in-band products that can dominate the worst spur, even when single-tone THD looks excellent. This section shows how to interpret IMD-driven SFDR, how crest factor (PAPR) changes usable dynamic range, and how to choose tones and spacing to avoid “easy landing spots” for spurs—without turning this into a communications-chain course.

Why THD can mislead in two-tone / multitone
  • THD summarizes harmonics of one tone (k·f0), not mixing between tones.
  • IMD produces m·f1 ± n·f2 products; IMD3 often lands close to carriers.
  • SFDR in-band is often set by the worst IMD product, not by harmonic totals.
What this section does NOT cover
  • DUC/NCO algorithm details: see RF DAC / DUC pages.
  • System-level link metrics and full comms planning (kept to DAC-side rules only).
Deliverable: minimum two-tone test SOP (repeatable and comparable)
  1. Pick tones and define the band: choose f1/f2 and a clear in-band region where SFDR is evaluated.
  2. Set levels with headroom: define per-tone amplitude and total composite level; apply back-off to avoid overload.
  3. Lock the chain: keep output mode, load/termination, attenuation, and any filtering fixed for the entire comparison.
  4. Fix FFT settings: keep BW, RBW/points, window, and averaging consistent; do not change them between A/B runs.
  5. Mark IMD landing points: locate IMD3 (2f1−f2, 2f2−f1) and optionally IMD5; record the worst in-band spur.
  6. Report conditions: f1/f2, spacing, per-tone level, FS, load, observation BW, unit (dBc/dBFS), and exclusions.
Practical tone planning rules (DAC-side only)
Avoiding in-band IMD traps
  • Choose spacing so IMD3 does not land inside the most sensitive in-band region.
  • Do not align spacing with known periodic aggressors (ripple, update cadence, interference).
  • If a fixed leakage spur exists, avoid tone choices that place IMD products on top of it.
Crest factor (PAPR) and usable dynamic range
  • Higher PAPR requires more back-off to stay linear.
  • Back-off lowers carrier levels; noise floor can become the limiter for SNDR.
  • Track worst spur and integrated noise separately to avoid false conclusions.
Two-tone spectrum with IMD3 and IMD5 landing points A simplified spectrum showing two carriers f1 and f2, with IMD3 products 2f1-f2 and 2f2-f1 near the carriers, and IMD5 further out, illustrating how SFDR can be set by IMD not THD. Two-tone: IMD can set SFDR even when THD looks great Frequency Noise floor f1 f2 IMD3 IMD3 IMD5 IMD5 IMD3 (2f1−f2 / 2f2−f1) often lands near the carriers and can dominate SFDR. SFDR

Audio hi-fi focus: noise floor, low-order distortion, idle tones, and weighting

In audio, “good numbers” can still sound wrong if the harmonic structure is unfriendly, if idle tones appear near quiet passages, or if the reported SNDR/DR uses a bandwidth or weighting that hides the real noise profile. This section maps listening risks to measurable checks, and provides an acceptance-field checklist that prevents spec misunderstandings.

What matters most in hi-fi validation
  • Low-order distortion: 2nd/3rd structure can matter more than total THD.
  • Idle tones: discrete lines at low level can be audible even if THD+N is good.
  • Weighting / BW: A-weighting and bandwidth choices change the reported SNDR/DR.
What this section does NOT cover
  • Deep ΣΔ modulator theory: see Delta-Sigma DAC pages.
  • Full audio system topics (speaker/room): kept to DAC-side measurable rules.
Deliverable: audio acceptance fields (must be stated to avoid spec disputes)
Field Why it matters Minimum statement
Test tone & level THD/SNDR depend strongly on frequency and amplitude; near-quiet behavior matters for audibility Frequency (e.g., 1 kHz) and level (dBFS), plus at least one low-level point
Bandwidth integration Noise and DR depend on integration limits; comparisons break if BW differs State the integration band (e.g., 20 Hz–20 kHz) and whether DC is excluded
Weighting (A-weighted or not) Weighting reshapes the reported noise; A-weighting can hide low-frequency hum components Explicitly state A-weighted / unweighted for SNDR/DR and noise measurements
Load and output mode Driver behavior and distortion structure depend on load, termination, and differential/single-ended mode Output mode, load impedance, and any transformer/line driver settings
Harmonic structure Two designs with similar THD can sound different if 2nd vs 3rd dominance differs At minimum, report 2nd and 3rd levels (or indicate dominance) alongside total THD+N
Idle tone check Discrete lines near silence can be audible; not captured by a single THD number Define a low-level or near-zero condition and the maximum allowed discrete tone within 20 Hz–20 kHz
Same noise floor, different harmonic structure: 2nd-dominant vs 3rd-dominant Two simplified spectra with the same noise floor. Left shows a stronger 2nd harmonic and weaker 3rd. Right shows a stronger 3rd harmonic. Illustrates why structure matters beyond total THD. Same noise floor, different harmonic structure 2nd-dominant Noise floor Fund. 2nd 3rd 3rd-dominant Noise floor Fund. 2nd 3rd Structure matters: similar THD totals can hide very different low-order profiles and listening risks.

RF transmission focus: wideband SFDR, images/spurs planning, and mask thinking

For RF transmission and direct-RF sampling designs, the key risk is not a single SFDR number at one tone, but where images and spurs land across a wide observation bandwidth. A clean design comes from defining “zones” (main band, guard, image bands, filterable regions), predicting likely landing points, and knowing when the problem must be handed off to clocking, filtering, or RF-DAC-specific pages.

How to read wideband SFDR (the three pitfalls)
  • Frequency-dependent: worst spur often shifts as fOUT changes.
  • Bandwidth-dependent: wider observation increases the chance of a worse spur.
  • Readout consistency: unit (dBc vs dBFS), search band, and exclusions change results.
When to hand off to a deeper page
  • Clocking & Phase Noise: near-carrier sidebands or strong clock sensitivity.
  • Reconstruction Filter: images land too close to the main/guard zone to be ignored.
  • RF DAC / DUC: artifacts correlate with digital upconversion/interpolation settings.
Deliverable: landing-point table template (rules-level, not heavy math)

Use the template below to define zones first, then mark where images/spurs land. The goal is a fast “Yes/No” decision: does anything land in a forbidden zone, and which page owns the fix (clocking, filtering, RF-DAC features, or layout/coupling).

Inputs (define first) Zones (draw on the spectrum) Predicted landings (rules-level) Forbidden zone? Owner (handoff if needed)
Fs = ____
Fc = ____
Main BW = ____
Guard = ____ 		In-band (main)
Guard zone
Adjacent zone
Filterable region 		Images appear in bands mirrored around Fs (rules-level)
Spur types: harmonics / images / feedthrough / periodic comb 		Yes / No Clocking / Filter / RF DAC / Coupling
fOUT sweep range = ____
AOUT/back-off = ____
Load = ____ 		“Do-not-land” notes:
mask-critical region(s)
known fixed spur region(s) 		Worst spur can move with fOUT
Wideband search reveals hidden worst-case points 		Yes / No Verification / Layout / Reference
Unit = dBc or dBFS
Observation BW = ____
Exclusions = ____ 		Keep readout consistent
for apples-to-apples planning 		If tones are fixed while a spur is fixed, suspect leakage/coupling Yes / No Clocking / Filter / Coupling
Frequency-zone planning for RF transmission: main band, guard, images, and filter regions A band planning diagram on a frequency axis, showing the main band, guard zones, image bands, and filterable regions, with markers indicating spur/image landing points and forbidden zones. Plan zones first, then check where images and spurs land Frequency Main band BW Guard Guard No-land No-land Image band Image band Filter region Filter region Spur Image If an image or spur lands in a no-land zone, the fix is usually owned by clocking, filtering, RF-DAC features, or coupling/layout—not by one “better SFDR” number.

Error budgeting & verification: from specs to guaranteed performance

Guaranteed performance comes from turning datasheet numbers into a testable budget and a minimum verification set. This section provides a practical approach to roll up reference noise/ripple, driver nonlinearity, filtering, coupling/leakage, and clock-related behavior into explainable SFDR/SNDR outcomes. It also defines a verification matrix that makes failures assignable to one owner path.

Budget categories (keep ownership clear)
  • Reference: noise / ripple signatures (floor or sidebands).
  • Driver: harmonics and IMD scaling with amplitude.
  • Filter/load: band shaping and sensitivity to bandwidth/termination.
  • Coupling: fixed-frequency leakage spurs.
  • Clock-related: near-carrier behavior and clock-cleanliness sensitivity.
Minimum verification set (cover the real risks)
  • Single-tone baseline + repeatability checks.
  • Sweep frequency and amplitude to find worst-case points.
  • Two-tone and multitone for IMD and PAPR/back-off behavior.
  • Temperature sweep only after the “worst point” is known.
Deliverable: verification matrix (dimension × metric × threshold × backtrace path)
Test (dimension) Target metric Pass threshold (template) Conditions to freeze If fail → first owner
Single-tone baseline SFDR / SNDR / noise floor SFDR ≥ __ dBc (in-band)
SNDR ≥ __ dB (BW = __) 		fOUT, AOUT, Fs, load, BW/RBW, window, unit, exclusions Coupling (fixed spur) / Reference (sidebands) / Driver (harmonics)
Sweep frequency Worst spur map (wideband SFDR) Worst spur ≤ __ dBFS in forbidden zones
or SFDR ≥ __ dBc across range 		Same observation BW, spur search rules, and reporting unit Filter/load (band shaping) / Clocking (near-carrier) / Coupling
Sweep amplitude Harmonics/IMD scaling vs level Harmonics/IMD remain below __ dBc
at required output levels 		Same load and chain headroom (attenuation), same measurement BW Driver (nonlinearity) / DUT settings / Measurement headroom
Two-tone IMD3/IMD5 and in-band SFDR Worst IMD product ≤ __ dBc
within the defined band 		Tone spacing, per-tone level, BW, unit, and exclusions Driver / Coupling / Clocking sensitivity
Multitone (PAPR/back-off) Usable dynamic range under crest factor No spur in forbidden zones at required back-off
Noise floor within budget 		Same spectrum-search rules; track worst spur and integrated noise separately Driver / Reference (noise) / Coupling (fixed spur)
Temperature sweep Worst-point stability and drift Worst-case remains within threshold across temp range Same worst-point frequency and same measurement conditions Reference / Driver / Layout thermal gradients
Verification matrix and backtrace routing from failures to ownership A simplified grid showing verification tests versus ownership categories (reference, driver, filter/load, clocking, coupling), with arrows indicating how typical failures route to the first suspected owner. Verify, then assign ownership fast (matrix + backtrace) Test Ref Driver Filter Clock Coupling Baseline Sweep f Sweep A Two-tone Multitone Temp Reference Driver Filter / Load Clocking Coupling Use the matrix to freeze conditions, find worst points, then route failures to one first owner for fast debugging.

Production checklist & IC selection notes (ask vendors / de-risk)

This section closes the loop from “nice datasheet numbers” to purchasable, verifiable, and mass-producible spectrum purity. Use the inquiry template to lock test conditions, demand worst-case and distribution data, and build a production screen that finds worst spurs fast.

Inquiry must-have condition fields (SFDR/SNDR cannot be compared without these)

Any SFDR/SNDR claim without the condition fields below is not acceptance-ready. Require vendors to answer each field explicitly and in the same units (dBc vs dBFS), the same observation band, and the same exclusions.

Condition field What must be stated (minimum) Typical pitfalls to prevent
Output form Voltage / Current / Differential, common-mode, termination method Comparing diff vs SE results; hidden CM settings
Tone plan fOUT points and sweep range; single-tone / two-tone / multitone Cherry-picked “best” frequency only
Amplitude & headroom dBFS / Vpp / Vrms, any back-off (e.g., −6 dBFS), crest factor if multitone Hidden back-off inflates SFDR/SNDR
Sampling & modes Fs, interpolation/DUC on/off, RTZ/NRZ if applicable Mode-dependent spurs not disclosed
Observation band Spur search band (wideband vs in-band), and any “forbidden zones” Changing the band changes “worst spur”
Noise integration SNDR/DR integration bandwidth, RBW/VBW (or FFT settings), weighting if audio Different BW makes numbers incomparable
Filter & output network Reconstruction filter (type/cutoff/order), balun/transformer, driver stage and gain Driver dominates distortion but is omitted
Load & termination 50Ω/100Ω/high-Z, SE/diff termination, cabling and attenuation Reflections/termination create fake spurs
Clock condition Clock source type, jitter-cleaning on/off, SYSREF/LMFC if JESD, near-carrier exclusions Clock leakage mistaken as DUT spur
Copy-ready inquiry snippet (paste into vendor email)
Please quote and guarantee SFDR/SNDR with full conditions: 1) Output form: Voltage / Current / Differential (common-mode = ___) 2) Fs = ___ ; modes: interpolation/DUC = on/off ; RTZ/NRZ = ___ 3) Tone(s): single-tone fOUT = ___ (plus sweep range ___ to ___) 4) Level(s): ___ dBFS (and back-off points: ___ dBFS) 5) Observation band for spur search: ___ (wideband or in-band) ; exclusions: ___ 6) SNDR/DR integration BW: ___ ; RBW/VBW or FFT settings: ___ 7) Filter/output network used: ___ ; load/termination: ___ 8) Clock setup (jitter cleaner / SYSREF/LMFC if JESD): ___ 9) Provide min/guaranteed and distribution data (mean/sigma or 3σ) over temp range ___.

Vendor “must-provide” worst-case & consistency data (avoid sample-to-mass surprises)

  • Guaranteed, not typical: SFDR/SNDR min values at defined frequency points and temperatures (not only room temp).
  • Worst-point disclosure: where the worst spur occurs across the sweep (frequency region + mode conditions).
  • Distribution: mean + sigma (or 3σ/Cpk) for SFDR/SNDR and worst spur under the same test conditions.
  • Temperature behavior: worst-point stability across temperature; show whether the worst point migrates with temp.
  • Revision & PCN control: silicon revision impact on spurs; require notification and re-qualification triggers.

Production test checklist (capture worst spurs with minimum time)

A) Fast screening (per-unit, seconds-level when automated)
  1. Fixed single-tone at 2–3 points (low / mid / near-band-edge): run spur search and record worst spur in the defined band.
  2. Two output levels: near full-scale and a back-off point (e.g., −6 dBFS). This separates driver nonlinearity issues from noise-floor limits.
  3. Freeze the chain: fixed filter + fixed load/termination + fixed attenuation. Changing the chain changes the spur map.
  4. Freeze FFT/SA rules: same observation band, RBW (or FFT points/window), same spur exclusions, same reporting unit (dBc or dBFS).
  5. Pass/fail limits: define thresholds per zone (in-band, guard, forbidden/no-land) and fail on zone violations, not on a single headline SFDR.
B) Audit tests (per-batch / per-shift, targeted to worst-point)
  • Two-tone IMD spot-check: one spacing only; fail on worst IMD3 product in the defined band.
  • Limited sweep around the known worst region: a short sweep around the worst-point frequency range, not full-band.
  • Temperature spot-check: test only the worst-point frequency at temperature corners to validate migration risk.
C) Recommended production instrumentation options (choose one)
  • Spectrum analyzer (SA): simplest spur search; best when the band is high and RBW control is needed.
  • FFT tester via sampling ADC: cost-effective for automated screening (fixed filter/load + ADC + FFT + limits).

Example MPN shortlist (selection notes by role)

The list below is intentionally short. Each item maps to a production-relevant role: RF DAC, precision setpoint DAC, driver/buffer, reference/LDO, clocking, and FFT test ADC.

  • RF / wideband SFDR DAC: AD9164; DAC38RF82; DAC38RF89
  • Precision setpoint / bias DAC: AD5791; DAC81416
  • Output driver / buffer (often the real distortion owner): OPA1612; ADA4898-2
  • Reference & clean supply (noise/ripple control): ADR4525; LT3042
  • Clocking / synchronization (JESD systems): AD9528
  • FFT test ADC (screening chain): AD9689; ADC32RF45
Production spectral screening chain: pattern to pass/fail with fixed load and FFT test A production test chain diagram showing pattern or BIST feeding a DAC, followed by fixed filter and fixed load, then FFT test (ADC or spectrum analyzer) and pass/fail limits. Highlights key controllable knobs: fOUT, AOUT/back-off, observation bandwidth, and load. Production spectral screening chain (fixed conditions → repeatable results) Pattern / BIST Tone set DAC Device under test Fixed filter Reconstruction Fixed load 50Ω / term FFT test ADC or SA Pass / Fail Limits by zone Freeze controllable knobs: fOUT AOUT (back-off) BW / RBW Load Fixed filter and fixed load turn spectrum screening into a repeatable pass/fail process; zone-based limits capture worst spurs faster than a single headline number.

Request a Quote

Accepted Formats

pdf, csv, xls, xlsx, zip

Attachment

Drag & drop files here or use the button below.

FAQs: THD / SFDR / SNDR for DAC spectrum purity (with JSON-LD)

These FAQs capture long-tail questions without expanding the main content. Each answer is structured for fast decisions: what it means, what to check first, which conditions must be frozen, and where to jump next on this page.

SFDR is higher but SNDR is worse—what does that mean in practice?
Answer

The output is dominated by a higher noise floor (or integrated noise), while discrete spurs are well controlled. In RF, this often limits usable dynamic range; in audio, it can reduce “black background” even if no strong spurs exist.

Quick checks
  • Confirm SNDR/DR integration bandwidth (noise gets worse with wider BW).
  • Check whether a driver/LDO/reference is raising the wideband floor.
  • Verify SFDR search band is the same as the application “forbidden zone”.
  • Repeat at a back-off level (e.g., −6 dBFS) to see if noise vs nonlinearity is the limiter.
Conditions to freeze

Reporting unit (dBc vs dBFS), observation band, SNDR/DR integration BW, filter/output network, load/termination, and clock setup.

Next section

Jump to H2-10 (Error budgeting & verification) and H2-11 (Production checklist).

Is SFDR reported in dBc or dBFS, and why does it change decisions?
Answer

dBc is referenced to the carrier (tone amplitude), while dBFS is referenced to full-scale. If output level/back-off changes, dBc may stay similar while dBFS shifts, which can change whether a spur violates an absolute mask or a relative requirement.

Quick checks
  • Lock the output level definition (dBFS vs Vpp) before comparing parts.
  • If “forbidden zone” limits are absolute, prefer dBFS-style thinking.
  • Require vendors to state the unit explicitly in quotes and reports.
Conditions to freeze

Output amplitude/back-off point, unit, observation band, and spur exclusion rules.

Next section

Jump to H2-11 (Inquiry fields).

When does looking at SNDR mislead and SFDR becomes the real risk?
Answer

In transmit chains, a single discrete spur in or near a protected region can be a hard failure even if integrated noise is excellent. SNDR averages noise+distortion over a bandwidth; SFDR exposes the worst single offender that can violate zone-based limits.

Quick checks
  • Define the “no-land / forbidden” zones first (guard, adjacent, image bands).
  • Run wideband spur search; record worst spur location, not only the value.
  • Confirm spurs are real (move tone, change BW/RBW, check coupling).
Conditions to freeze

Forbidden zone definition, observation band, spur search rules, and reporting unit.

Next section

Jump to H2-9 (Wideband SFDR planning).

A spur stays at a fixed frequency when fOUT moves—what does it imply?
Answer

A fixed-frequency spur often indicates leakage/coupling (clock, LO, DC-DC, digital activity) or test-chain artifacts rather than DAC nonlinearity. Real harmonic/image products typically move as the tone moves.

Quick checks
  • Change fOUT slightly; confirm the spur frequency does not move.
  • Change cable routing/grounding/attenuation; fixed spurs often change.
  • Turn off nearby switching supplies or change their frequency (if possible).
  • Check if the spur matches known clocks (reference, SYSREF, DC-DC).
Conditions to freeze

Same measurement chain headroom (attenuation), same RBW/FFT settings, same cabling/termination.

Next section

Jump to H2-10 (Backtrace ownership) and H2-11 (Freeze conditions).

A spur moves with the tone—how to tell harmonic vs image quickly?
Answer

Harmonics follow multiples of the tone (2×, 3×, …), while images usually relate to sampling/reconstruction and appear mirrored around sampling bands. The fastest method is to sweep fOUT and observe whether the spur tracks a multiple or reflects around a band edge.

Quick checks
  • Sweep fOUT by a small ratio; check if spur stays at 2×/3× position (harmonic).
  • Change Fs or interpolation mode; image patterns often shift with Fs/modes.
  • Insert/adjust reconstruction filtering; images should change more than true harmonics.
Conditions to freeze

Fs, mode settings (interp/DUC/RTZ), observation band, and filter/output network.

Next section

Jump to H2-9 (Zone planning).

Why do near-carrier sidebands appear and when should clocking be suspected?
Answer

Near-carrier sidebands often indicate modulation of the carrier by a periodic source (clock leakage, reference ripple, supply ripple, or coupling). If sidebands change strongly with clock source/jitter cleaning, clocking is a primary suspect.

Quick checks
  • Swap clock sources or enable/disable jitter cleaning; compare sidebands.
  • Change supply filtering on reference/clock rails; look for sideband changes.
  • Change fOUT; if sideband spacing stays constant, suspect a fixed modulator frequency.
  • Check for correlation with SYSREF/LMFC activity in JESD systems.
Conditions to freeze

Clock configuration, supply rails (reference/clock), observation band near carrier, RBW/FFT windowing.

Next section

Jump to H2-10 (Backtrace) and H2-11 (Freeze clock conditions in inquiries).

How to avoid analyzer IMD and cabling/termination creating fake spurs?
Answer

Fake spurs commonly come from measurement headroom limits, reflections, and unintended mixing. The safest approach is to lock a clean chain (fixed attenuation, fixed termination, fixed filter), then verify a spur is real by changing tone and bandwidth.

Quick checks
  • Add attenuation to improve analyzer linearity; confirm spurs drop as expected.
  • Use correct 50Ω/100Ω termination; avoid adapters that create reflections.
  • Shorten/replace cables; route away from digital clocks and DC-DC inductors.
  • Run a “spur reality check”: move fOUT, change RBW/BW, see if the spur behavior matches physics.
Conditions to freeze

Attenuation/headroom, termination, filter/output network, cabling layout, RBW/FFT settings.

Next section

Jump to H2-11 (Production screening chain).

Which FFT/SA settings matter most for a trustworthy worst-spur search?
Answer

The most important settings are the observation band, RBW (or FFT length), windowing/leakage control, averaging method, and spur exclusion rules. Changes in these can move the reported “worst spur” even when the DUT is unchanged.

Quick checks
  • Fix the observation band and report it in every plot/report.
  • Use consistent RBW/FFT length so noise/spur visibility is comparable.
  • Use a window that controls leakage when coherent sampling is not guaranteed.
  • Document any exclusions (DC bins, carrier leak bins, known clock lines).
Conditions to freeze

Observation band, RBW/FFT length, window/averaging, exclusion rules, and reporting unit.

Next section

Jump to H2-11 (Inquiry fields + production SOP).

Why does SFDR change when RBW/observation bandwidth changes?
Answer

Wider observation bands increase the chance of encountering a worse spur, and narrower/wider RBW changes how energy is captured and displayed. SFDR is a “maximum-of-many” metric, so measurement scope strongly affects the reported worst case.

Quick checks
  • Always report the spur search band together with SFDR.
  • If comparing devices, use the same RBW/FFT length and the same exclusions.
  • For zone-based planning, match the band to the actual “no-land” zones.
Conditions to freeze

Observation band definition, RBW/FFT length, averaging, and exclusion rules.

Next section

Jump to H2-9 (Wideband planning).

Why does back-off often improve SFDR, and when it does not?
Answer

Back-off reduces stress on nonlinear elements (DAC output stage, driver, transformer), often lowering harmonics/IMD and improving SFDR. It may not help if the dominant limit is a fixed-frequency leakage spur or a noise-floor limit.

Quick checks
  • Try two levels (near FS and −6 dBFS) and compare harmonic/IMD changes.
  • If the worst spur does not change with level, suspect coupling/leakage.
  • If SNDR worsens as level drops, the system is noise-limited.
Conditions to freeze

Output level definition (dBFS/Vpp), load/termination, filter/output network, and spur search band.

Next section

Jump to H2-10 (Verification matrix).

Why does changing load/driver change THD/IMD more than the DAC itself?
Answer

The driver, transformer/balun, and termination often operate closer to nonlinear regions than the DAC core, especially at high frequency or high swing. As a result, measured distortion can be owned by the surrounding chain, not the DAC silicon.

Quick checks
  • Hold DAC settings constant; swap driver stage or termination and compare harmonics/IMD.
  • Reduce swing or add back-off; if IMD improves strongly, the driver is likely the owner.
  • Check headroom of the measurement chain (analyzer/ADC) to avoid fake IMD.
Conditions to freeze

Driver topology, load/termination, filter/network, output level, and measurement headroom.

Next section

Jump to H2-10 (Ownership backtrace).

What is the minimum production test to catch worst spurs without long sweeps?
Answer

Use a fixed chain (fixed filter and load) and test a small set of tones at two output levels, then run a wideband spur search inside the defined forbidden zones. Record both the worst spur value and its location, and fail on zone violations.

Quick checks
  • Tones: low / mid / near-band-edge (2–3 points only).
  • Levels: near full-scale + a back-off point (e.g., −6 dBFS).
  • Freeze FFT/SA rules: observation band, RBW/FFT length, averaging, exclusions.
  • Add a small audit test per batch (two-tone IMD at one spacing).
Conditions to freeze

Fixed filter/load, unit, observation band, spur search rules, output level definition, and clock conditions.

Next section

Jump to H2-11 (Production checklist).

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.