123 Main Street, New York, NY 10001

Submit Your BOM

High-Speed ADC & Clocking (GS/s)

High-Speed ADC & Clocking (GS/s)

← Back to: Test & Measurement / Instrumentation

In GS/s measurement systems, sampling-clock quality (phase noise, RMS jitter, spurs) sets the real ceiling for SNR/ENOB and the repeatability of multi-ADC alignment and triggering. The reliable workflow is: build a jitter/phase-noise budget first, then choose the clock architecture and distribution, and finally validate deterministic latency and field logs so timing stays provably stable in real use.

H2-1 · What problem clocking solves in GS/s ADCs (jitter is the ceiling)

In GS/s acquisition, the sampling clock is not “just a frequency.” Its time uncertainty (RMS jitter) directly sets the best achievable high-frequency SNR/ENOB, and it also determines whether multi-channel timing alignment and trigger repeatability are truly stable.

Decision rule (use this before picking parts)
First compute the jitter budget from the target SNR/ENOB at the highest effective input frequency, then choose the ADC + clock chain that can meet it with margin.
Jitter-limited SNR relationship (no long derivation)
SNRjitter ≈ −20·log10(2π·fin·σt) where fin is the effective highest input frequency of interest, and σt is the total RMS time jitter seen at sampling (clock + aperture).
Why “higher frequency is more sensitive”
  • The term 2π·fin·σt grows linearly with input frequency, so the same jitter produces a larger phase error at higher fin.
  • Practical impact: a clock chain that looks “fine” at 50–100 MHz may cap performance at 300–800 MHz, even if the ADC has more bits on paper.
Two quick numeric anchors (order-of-magnitude)
Target: 60 dB at 200 MHz
Required σt ≈ 0.80 ps (≈ 800 fs) total RMS.
Target: 70 dB at 500 MHz
Required σt ≈ 0.10 ps (≈ 100 fs) total RMS.
These anchors explain why “just choosing a higher-resolution ADC” rarely helps at high frequencies unless the clock chain is already in the right jitter class.
What clocking must guarantee beyond SNR (instrument-grade expectations)
  • Multi-channel alignment: low skew and stable phase relationships across channels over temperature.
  • Trigger repeatability: capture starts at the same effective sampling phase/time, not just “a trigger happened.”
  • Restart reproducibility: deterministic latency / alignment so results do not shift after reboot or relock.
System loop overview: clock chain and sync chain for GS/s ADC capture Block diagram with two parallel paths. Left path shows reference oscillator to jitter cleaner/PLL to clock tree to multi-channel ADC. Right path shows trigger and SYSREF to synchronization/alignment to FPGA capture. Arrows label RMS jitter, skew and deterministic latency. F1 · Clock quality sets the ceiling Ref / OCXO phase noise Jitter Cleaner / PLL residual jitter spurs Clock Tree fanout skew Multi-ADC ENOB ceiling Trigger / SYSREF timing events Sync / Align deterministic latency repeatability FPGA Capture lane deskew · timestamps RMS jitter skew deterministic latency Clock chain quality caps SNR/ENOB; sync chain guarantees repeatable timing across channels and reboots.

H2-2 · Build a jitter & phase-noise budget (from target ENOB back to clock)

A usable clock plan starts with a numeric ceiling. This section turns the target ENOB/SNR into a maximum allowable total RMS time jitter, then allocates that jitter across aperture, PLL/jitter cleaner residual, reference, clock distribution, and coupling paths.

Workflow (target → budget → allocation)
  1. Pick the worst-case frequency point: choose the highest effective input frequency of interest (edge of the usable analog bandwidth or the highest tone that must meet the SNR/ENOB spec). Jitter penalties scale with this frequency.
  2. Convert ENOB to SNR if needed: SNR ≈ 6.02·ENOB + 1.76. Use the resulting SNR as the ceiling for the jitter-only limit (leave margin for other error sources).
  3. Solve for the maximum total RMS jitter: σt,max ≈ (1 / (2π·fin)) · 10−SNR/20. This gives the total time jitter allowed at sampling for the chosen fin.
  4. Separate what cannot be “cleaned”: the ADC has an aperture jitter term that remains even with an ideal external clock. Treat the remaining allowance as the clock-chain jitter budget.
  5. Allocate the clock-chain budget by controllable blocks: jitter cleaner residual, reference contribution, fanout/additive jitter, divider/serializer noise (if used), and PCB/power coupling. Then verify each block with a matching measurement method.
How the jitter terms combine (concept-level, but actionable)
σtotal ≈ √(σaperture2 + σclock2) and the clock-chain term is typically allocated as an RSS of residual PLL/cleaner jitter, reference contribution, additive fanout/buffer jitter, and coupling/noise-induced timing modulation.
Budget template (use “order-of-magnitude” ranges, not vendor promises)
Budget item Typical order (fs RMS) How to validate Pass criterion
ADC aperture jitter ~50–300 fs (architecture-dependent) SNR vs fin slope test / vendor characterization correlation Does not consume the full σt,max alone
PLL / jitter cleaner residual ~50–300 fs (depends on loop/VCO) Integrated phase noise → RMS jitter (defined band) Meets allocated RSS slice with margin
Reference contribution ~10–200 fs (band-dependent) Phase noise + spur scan; lock/holdover events Spurs below system limit; stable in temperature
Clock tree / fanout additive ~20–150 fs per stage (can accumulate) Additive jitter measurement; skew vs temperature Skew + jitter stay inside allocated window
PCB / power coupling into clock From negligible → dominant (layout-driven) Sideband/spur changes under load, rail noise injection tests No “mystery” spurs; stable jitter across operating states
Keep the table honest: use frequency-band-defined jitter numbers (the same integration limits used for acceptance testing).
Common failure patterns this budget prevents
  • Wrong worst-case fin: budgeting at 100 MHz then failing at 400–800 MHz.
  • Ignoring aperture jitter: chasing an “impossible” clock number that cannot beat the ADC’s floor.
  • One-number jitter claims: mixing different integration bands so comparisons are meaningless.
Jitter budget stack: total allocation across aperture, PLL residual, reference, fanout and coupling A budget stack diagram showing total RMS jitter at the top and a breakdown into major contributors: ADC aperture jitter, PLL/jitter-cleaner residual, reference contribution, clock tree additive jitter, and PCB/power coupling. Includes RSS formula reminder. F2 · Target → Budget → Allocation Total σt,max from SNR target @ fin ADC aperture ~50–300 fs PLL residual ~50–300 fs Reference ~10–200 fs Clock tree additive fs PCB coupling layout-driven How to use the stack 1) Choose worst fin highest effective input frequency 2) Solve σt,max σt,max ≈ (1/(2πfin))·10−SNR/20 3) Allocate by RSS σtotal ≈ √(σap2 + σclk2) keep the same integration band 4) Validate each block phase noise → jitter, spur scan, skew drift then verify SNR vs fin at the ADC output Budget first, then implement: aperture sets the floor; the clock chain must stay inside the remaining RSS window across modes and temperature.

H2-3 · Phase noise 101 for instrument clocks (what parts of the plot matter)

A phase-noise plot is only useful when it is read in the same way the instrument will “feel” timing uncertainty: close-in noise, far-out floor, the chosen integration band, and discrete spurs must be interpreted together. Comparing jitter numbers without a matching integration band is not reproducible.

How to read the plot (the parts that actually matter)
  • Close-in region: dominated by 1/f behavior (slow phase wander). This region often drives jitter when the integration starts very near the carrier.
  • Far-out region: the noise floor sets the tail of the integral. A wider upper limit makes the floor contribute more to RMS jitter.
  • Integration band: the jitter number only has meaning when the offset range [f1, f2] matches the system’s sampling structure and acceptance test.
  • Spurs (reference spur): narrow “spikes” can create deterministic sidebands and false tones; they do not average away like broadband noise.
Three “must name” features (each has a different failure mode)
1/f corner
The offset where the slope transitions toward a flatter region; it indicates whether close-in noise dominates the chosen integration band.
Noise floor
The far-out baseline; if the upper integration limit is wide, the floor can dominate RMS jitter even when close-in looks excellent.
Reference spur
A discrete tone from reference/PLL mechanisms; it produces deterministic sidebands and can appear as “ghost lines” in FFT measurements.
Concept chain: integrated phase noise → RMS jitter (no long derivation)
  1. Choose an offset integration band [f1, f2] that matches how the instrument validates timing performance.
  2. Integrate the phase-noise density over that band to obtain a total phase variance (broadband noise plus separately tracked spur power).
  3. Convert phase variance to time-domain RMS jitter using the output clock frequency (a higher carrier changes the phase-to-time scaling).
  4. Report jitter with the integration limits and spur handling method; otherwise, comparisons across designs are not meaningful.
Practical pitfalls (why “nice plots” still fail in real instruments)
  • Band mismatch: integrating to a different f2 than the acceptance test hides floor-dominated jitter.
  • Ignoring spurs: a single reference spur can violate spur limits even when broadband jitter is low.
  • Close-in overfocus: optimizing only the near-carrier slope while the far-out floor dominates the total RMS.
Phase noise plot anatomy: close-in, noise floor, integration band and reference spur Simplified log-log phase noise curve with labeled close-in region, far-out noise floor, an integration band highlighted, a 1/f corner marker, and a discrete reference spur spike. F3 · What to read on a phase-noise plot Phase noise (dBc/Hz) Offset frequency (log) Close-in Crossover Far-out floor low high 1/f corner Integration band [f1..f2] Reference spur Rules Same band for jitter comparisons Track spurs as discrete limits Floor matters when f2 is wide Always report RMS jitter with the integration band and spur handling method; otherwise, the number cannot be compared or reproduced.

H2-4 · Choosing the clock architecture: integer-N vs fractional-N vs DDS-assisted

Architecture choice should be driven by instrument behavior: frequency switching and sweep needs, spur tolerance, lock/settling time, and whether phase continuity or multi-channel coherence must be preserved. This section compares three common approaches using those criteria.

Translate instrument requirements into selection levers
  • Step size / resolution: how fine the frequency grid must be for sweeps and calibration points.
  • Spur tolerance: whether discrete sidebands will create false tones or compromise SFDR limits.
  • Lock/settling time: how quickly the instrument can change frequency and be “measurement ready.”
  • Phase continuity / coherence: whether phase must remain continuous across tuning steps and across channels.
Integer-N PLL
  • Spurs: typically easier to keep low.
  • Step size: limited by reference/divider constraints.
  • Best fit: fixed or coarse grids where spur limits dominate.
Fractional-N PLL
  • Step size: fine grid and flexible planning.
  • Spurs: modulation/quantization products must be managed.
  • Best fit: sweeps and multi-point tuning with controlled spur strategy.
DDS-assisted
  • Step size: ultra-fine with phase control.
  • Spurs: images/aliases require filtering and planning.
  • Best fit: phase continuity and very fine frequency grids.
Selection matrix (use with the jitter budget and spur limits)
Criterion Integer-N Fractional-N DDS-assisted
Spur risk Low Medium (needs strategy) Medium (images/plan)
Step size Limited Fine Ultra-fine
Complexity Low Medium High
Phase continuity Medium Medium High
Typical best use fixed tones / tight spur limits sweeps / multi-point tuning coherence / phase-controlled stepping
Clock architecture comparison: Integer-N, Fractional-N, and DDS-assisted Three-column block diagram comparing Ref→PLL→VCO, Ref→FracN PLL→VCO, and Ref→PLL plus DDS feeding a mixer/divider. Each column includes bottom tags for spur risk, step size, and complexity. F4 · Choosing a clock architecture Integer-N PLL Fractional-N PLL DDS-assisted Ref PLL VCO Divider / Out Ref Frac-N PLL VCO Divider / Out Ref PLL DDS Mixer / Divider Out Spur: Low · Step: Limited · Complexity: Low Spur: Med · Step: Fine · Complexity: Med Spur: Med · Step: Ultra · Complexity: High Pick the architecture that meets the jitter budget and spur limits while matching sweep behavior, settling time, and phase continuity needs.

H2-5 · PLL loop bandwidth & dynamics (how to trade lock time vs noise)

A PLL is a noise-shaping control loop. Its loop bandwidth determines which noise dominates the output across offset frequency: reference-side noise inside the loop versus VCO-side noise outside the loop. The same bandwidth also sets lock time and how fast the clock recovers after frequency steps or disturbances.

Practical rule
Loop BW is a trade knob: BW ↑ usually locks faster but injects more reference noise/spurs; BW ↓ can improve close-in noise only if the VCO is cleaner, but it locks and recovers more slowly.
Wide loop bandwidth (BW ↑)
  • Lock time: faster relock and quicker settling after frequency steps.
  • Noise impact: more reference/divider noise can pass to the output inside the loop.
  • Best fit: sweep modes and fast “measurement-ready” requirements.
Narrow loop bandwidth (BW ↓)
  • Lock time: slower lock and slower recovery from disturbances.
  • Noise impact: VCO close-in noise can dominate; improvement depends on VCO quality.
  • Best fit: fixed tones where clean close-in performance matters more than agility.
Key blocks that set BW and dynamics (concept level, but actionable)
PFD
Detects phase/frequency error; dead-zone or nonlinearities can worsen close-in spurs and settling behavior.
Charge Pump
Converts error to current; mismatch/leakage can create reference-related spurs and bias the control voltage.
Loop Filter
Sets bandwidth and damping; it defines where ref noise hands over to VCO noise and how quickly the loop settles.
VCO
Dominates out-of-loop noise; control-voltage and supply coupling can convert rail noise into phase noise.
Tie-in to phase-noise integration
If the jitter integration band crosses the loop BW, the measured RMS jitter contains both a reference-dominated region and a VCO-dominated region. Always report the integration limits together with loop BW so the result is reproducible.
Noise crossover and PLL loop bandwidth Two contribution curves (reference noise and VCO noise) versus offset frequency and a combined output curve. A vertical marker shows loop bandwidth and indicates the reference-dominated and VCO-dominated regions. F5 · Loop BW sets noise crossover and lock dynamics Relative phase-noise contribution Offset frequency (log) Ref-dominated VCO-dominated Reference noise VCO noise Output (combined) Loop BW Trade BW ↑ lock fast ref in BW ↓ cleaner if VCO locks slow Loop bandwidth defines the crossover between reference-dominated and VCO-dominated offsets and strongly affects settling and lock behavior.

H2-6 · Jitter cleaners & holdover (cleaning, redundancy, and what can go wrong)

In instruments, a jitter cleaner is a reliability component as much as a performance component. It can reject a noisy external reference, provide holdover when references disappear, and support reference redundancy so timing remains controlled and observable.

Two behavior classes (described by what they do)
Narrowband cleaning
Tracks the external reference slowly and prioritizes short-term stability. It rejects reference wander/noise in a wider offset region, but it settles more slowly after reference changes.
Wideband tracking
Follows the reference quickly and is better for agile synchronization, but it can pass more reference noise/spurs to the output. Reference quality requirements become stricter.
Holdover (reference missing)
  • Trigger: ref loss or ref quality below threshold.
  • Behavior: maintain output frequency/phase predictably using the internal oscillator model.
  • Observability: raise status and timestamped alarms so measurements remain traceable.
Redundancy (Ref A/B switching)
  • Goal: avoid uncontrolled phase hits when switching references.
  • Strategy: either minimize phase discontinuity or enforce a bounded, logged phase step.
  • System rule: switching events must be visible to the capture system and field logs.
Failure symptoms (instrument-facing, not lab-only)
  • Spurs rise: reference spur injection or tracking mode too wide.
  • Wander increases: low-frequency disturbances dominate or frequent mode switching occurs.
  • Lost lock: loop dynamics/quality thresholds not met; reacquire loops too often.
  • Phase hit: uncontrolled A/B switching or reacquire step not bounded and logged.
Reference A/B selection with jitter cleaner and holdover state machine Diagram showing Ref A and Ref B feeding a selector into a jitter cleaner and clock output. A side state machine shows transitions between LOCK, HOLDOVER, REACQUIRE, and ALARM. F6 · Ref A/B + cleaner + holdover state machine Ref A quality monitor Ref B quality monitor Selector A / B switch Cleaner track / clean Clock Out Quality thresholds + logs ref loss · quality low · switch event timestamp State machine LOCK HOLDOVER REACQUIRE ALARM ref loss lock ok quality low bounded step Holdover and redundancy keep timing controlled during reference faults; switching and reacquire events must be observable and logged.

H2-7 · Clock tree & distribution (fanout, skew, additive jitter, isolation boundaries)

In multi-channel and multi-board instruments, the clock tree does more than deliver frequency. It determines channel-to-channel skew and adds additive jitter at every distribution stage. If the tree is not planned, the distribution network can consume the jitter budget and create alignment drift that cannot be fixed later.

Distribution strategy (practical)
Choose a topology first (star / balanced tree / cascade), then decide where to divide or buffer. Control additive jitter per stage and keep skew measurable, stable, and calibratable.
Star / hub fanout
  • Skew: easier to control by matching paths and loads.
  • Jitter: multiple fanout stages add jitter if overused.
  • Best fit: many channels with tight phase alignment needs.
Balanced tree
  • Skew: kept small by symmetric level count and similar loading.
  • Jitter: additive points exist, but can be planned per level.
  • Best fit: multi-board systems with structured clock domains.
Daisy-chain / cascade
  • Skew: accumulates stage by stage; drift risk is higher.
  • Jitter: each stage adds jitter; total can grow quickly.
  • Best fit: limited channels where topology simplicity is critical.
What grows in the tree (and why it matters)
Fanout additive jitter
Every buffer/fanout stage contributes its own timing noise. When stages are cascaded, additive jitter becomes a real part of the total RMS jitter budget and can limit achievable SNR/ENOB.
Skew sources
Skew comes from path differences (routing/connectors), device delay variation (channel-to-channel), and temperature drift across boards. In instruments, drift and repeatability often matter as much as initial skew.
Termination and amplitude (keep timing points consistent)
Use consistent termination and controlled amplitude across branches so edge shape and threshold crossing remain stable. Inconsistent loading can change edge timing and turn a “distribution” issue into a repeatability problem.
Clock tree topology comparison: star versus daisy-chain Side-by-side diagrams comparing a star fanout topology and a cascaded daisy-chain topology. Labels show skew accumulation, additive jitter points, and an optional isolation boundary. F7 · Clock tree topologies and error growth Star fanout Clock Hub Additive jitter points Source ADC 1 ADC 2 ADC 3 ADC 4 Skew small (if paths matched) · Jitter points concentrated Daisy-chain / cascaded Source Buffer 1 Buffer 2 Buffer 3 ADC A ADC B ADC C Skew accumulates · Additive jitter at each buffer Isolation boundary (if needed) Topology choice sets how skew and additive jitter grow; keep distribution consistent and measurable across channels and boards.

H2-8 · Multi-ADC synchronization: JESD204 SYSREF, deterministic latency, alignment

Multi-ADC synchronization in instruments is not just “same frequency.” The goal is deterministic latency: after a restart or link re-initialization, the sampling point and alignment return to a repeatable, known position. This repeatability enables coherent processing, stable calibration, and reliable multi-channel comparisons.

What deterministic latency means (instrument-facing)
When the system reboots or the JESD link is rebuilt, channel timing returns to the same alignment grid. Without deterministic latency, phase relationships and calibration offsets can change run-to-run.
Alignment flow (concept + steps)
  1. Clock stable: sampling clock distribution is stable across ADCs and the FPGA.
  2. Link bring-up: JESD lanes establish and reach an operational state.
  3. SYSREF distribution: SYSREF is delivered consistently to all devices.
  4. LMFC alignment: alignment grid locks using SYSREF within the valid timing window.
  5. Verify repeatability: restart/rebuild and confirm alignment returns predictably.
SYSREF window (concept)
SYSREF must be captured inside a defined timing window. If SYSREF arrival differs across devices, each device can lock to a different grid point and deterministic latency breaks.
Deskew (digital compensation)
Residual skew remains after hardware matching. Lane and channel deskew in the digital domain compensates remaining errors, but it depends on repeatable alignment to be stable across restarts.
SYSREF distribution and multi-ADC alignment with deterministic latency SYSREF fanout and sampling clock fanout feed multiple ADCs connected to an FPGA with multiple lanes. FPGA capture blocks show LMFC alignment, lane deskew, and channel alignment, with a repeatability check loop. F8 · SYSREF + multi-lane capture alignment (repeatable after restart) Sampling Clock fanout matched paths SYSREF windowed capture ADC 1 ADC 2 ADC 3 FPGA Capture LMFC align Lane deskew Channel align Repeatability check restart → align → same result lanes lanes lanes LMFC deskew align SYSREF window Deterministic latency depends on consistent SYSREF distribution, alignment to the LMFC grid, and repeatable deskew across restarts.

H2-9 · Trigger & time alignment (external trigger, time-of-arrival, jitter on trigger paths)

In GS/s capture systems, trigger is a timing chain. Any trigger jitter or path delay drift moves the effective start time of acquisition and directly shows up as measurement instability. A robust implementation treats trigger as a distributed, observable, and calibratable path.

Why trigger quality matters
Trigger timing error becomes acquisition timing error. Slow edges, noisy thresholds, temperature drift, and distribution skew can all shift the trigger decision time and reduce repeatability.
Conditioning & threshold
  • Edge shaping: make the decision edge clean and consistent.
  • Threshold strategy: keep decision level controlled and stable.
  • Noise tolerance: avoid multi-trigger or time-walk on noisy inputs.
Delay, fanout, distribution
  • Skew: channel-to-channel trigger arrival differences.
  • Additive jitter: each stage can add timing noise.
  • Calibratable: measurable delays enable stable alignment.
Time-of-arrival (TOA) / TIE
  • TOA stamp: record event time for alignment.
  • TIE concept: track timing error versus an ideal grid.
  • Repeatability: make trigger timing observable run-to-run.
Practical acceptance
A trigger path is “good” when inter-channel skew is controlled, arrival time is repeatable across restarts, and exceptions can be detected by timestamps rather than guessed from waveforms.
External trigger distribution and time alignment External trigger enters a conditioner and threshold block, then goes through delay and fanout to ADC/FPGA capture sync. Labels show trigger jitter, path delay, and a time-of-arrival timestamp output used for alignment. F9 · Trigger distribution, jitter, and time alignment Ext Trigger Conditioner threshold · shaping Delay / Fanout ADC / FPGA capture sync Trigger jitter Path Δt TOA stamp timestamp / TIE Align by sync + stamps Trigger paths have their own jitter and delay drift; timestamps make timing observable and enable repeatable alignment.

H2-10 · Practical PCB & power rules for low-jitter clocks (layout that actually matters)

Low-jitter clocks are often limited by layout, not by the clock IC datasheet. The goal is to prevent supply and digital switching noise from converting into phase modulation (sidebands) and added jitter, while keeping routing and thermal conditions stable for repeatable timing.

Supply isolation purpose (PLL / VCO)
Isolate sensitive clock rails so wideband switching noise and transients do not modulate phase. A quiet rail and a continuous return path reduce sidebands and improve run-to-run jitter stability.
Clock routing
  • Differential: keep pairs together with consistent spacing.
  • Controlled impedance: stable edge timing across branches.
  • Length match: match to the required alignment level.
  • Avoid split returns: do not cross plane splits.
Partitioning
  • Ref / PLL / tree zone: keep compact and quiet.
  • Keepout: separate from DC/DC and fast digital edges.
  • Measurable: plan test access for verification.
Thermal / mechanical
  • Thermal gradients: reduce drift around the reference.
  • Hot spots: keep heat sources away from Ref/PLL.
  • Stability: avoid stress and airflow surprises near the timebase.
Minimal “do not miss” checklist
Keep Ref/PLL/clock tree together in a quiet zone, isolate sensitive rails, route clocks as controlled differential pairs, match lengths where alignment requires it, never cross plane splits, and avoid placing fast switching noise next to the clock network.
Clock layout good versus bad: partitioning, returns, and noise coupling Side-by-side board sketches. Left shows a clean Ref/PLL/Clock Tree zone separated from noisy digital and DC/DC areas with matched differential routing. Right shows a bad example with switching noise near the clock, routing crossing a split return, and coupling indicated by noise icons. F10 · PCB rules that keep clocks low-jitter (Good vs Bad) Good Ref / PLL quiet rail Clock Tree matched routes FPGA / DSP kept away DC/DC noise zone Keepout Diff pair routes Bad Ref / PLL DC/DC too close Switching coupling Split return Clock crosses split (avoid) Keep clock-sensitive zones quiet and compact, isolate rails, avoid crossing plane splits, and keep switching noise away from Ref/PLL and clock routes.

H2-11 · Validation checklist (how to prove jitter, sync, and latency are under control)

“Pass once in the lab” is not enough for GS/s instruments. A complete proof is layered: R&D validation (measure the true performance), production tests (fast Go/No-Go), and field self-checks (continuous evidence and trend detection).

Evidence pack (what “done” looks like)
A complete record includes: phase-noise plot with integration band, integrated RMS jitter result, spur scan summary, lock/relock timing distribution, temperature sweep drift summary, and deterministic-latency repeatability checks.
R&D validation (measure the ceiling)
  • Phase noise + integration: plot + integrated jitter over the system-relevant offset band.
  • Spur scan: sweep modes/frequencies and record worst spur level + location.
  • Frequency hop & lock time: distribution (min/typ/max) and phase-hit checks.
  • Temperature sweep: drift summary (min/avg/max) and unlock/holdover events.
  • Deterministic latency: repeated cold-starts produce the same alignment grid point.
  • Multi-ADC skew: residual skew after deskew is stable across reboots.
Example clock-chain ICs (reference designs): TI LMK04828, ADI AD9528, ADI HMC7044, ADI AD9545, Skyworks Si5345/Si5395, TI LMX2594.
Production test (fast Go/No-Go)
  • Lock within window: lock must assert within a defined time limit.
  • Frequency sanity: main clock and SYSREF frequencies within tolerance.
  • Spur threshold spot-check: a few critical offsets must stay below a limit.
  • SYNC repeat: multiple power cycles produce the same sync status result.
  • Deskew table applied: post-deskew alignment meets the manufacturing limit.
Example distribution IC: TI CDCLVP1216 fanout (low additive jitter, multi-output).
Field self-check (continuous evidence)
  • Lock state: lock/unlock counters and time-in-state.
  • Holdover tracking: entry time and reacquire time distribution.
  • Sync health: SYSREF missing and LMFC slip/alignment reset counts.
  • Trigger drift stats: timestamp drift/skew estimates (trend-based).
  • Temp-binned drift: phase drift summary by temperature bins.
Example low-noise references: Crystek CCHD-957, NDK NZ2520SDA, Abracon AOCJY-10.000MHZ (OCXO).
Acceptance (instrument-grade)
Jitter is controlled when integrated jitter meets the budget and spurs remain below limits across modes. Sync is controlled when deterministic latency is repeatable after restarts and multi-ADC alignment remains stable across temperature and time.
Validation flow swimlanes: R&D, Production, Field Swimlane diagram with three lanes (R&D, Production, Field). Each lane lists short validation checkpoints for jitter, spurs, lock time, deterministic latency, and field evidence. F11 · “Done” proof: R&D → Production → Field Swimlanes Short checkpoints · consistent evidence R&D measure truth Production fast Go/No-Go Field evidence logs PN + Integrate Spur scan Lock time Temp sweep Det latency repeat Multi-ADC skew Lock OK Freq check Spur limit SYNC repeat Deskew applied Lock state Holdover events SYSREF missing LMFC slip Temp-binned drift Keep each lane short and evidence-driven: plots and stats in R&D, thresholds in production, counters and trends in the field.

H2-12 · Event logs & field evidence (catch latent clock faults before users do)

The most costly clock faults are intermittent: reference quality changes, temperature-dependent drift, occasional sync slips, and rare trigger timing anomalies. A clock/sync subsystem should be observable: log what matters, convert counters into threshold rules, and export evidence for service.

What to log (field-ready telemetry fields)
PLL / cleaner state
  • pll_lock_count / pll_unlock_count
  • holdover_entry_time
  • reacquire_time_ms (min/avg/max)
Why: frequent unlock/holdover signals reference, supply, or coupling issues.
Sync & alignment health
  • sysref_missing_count
  • lmfc_slip_count
  • alignment_reset_count
Why: any slip/reset means deterministic latency is no longer trusted.
Trigger timing evidence
  • trigger_toa_drift (trend)
  • path_skew_est_ps (stats)
  • trigger_jitter_rms (stats)
Why: trigger drift directly appears as acquisition start-time uncertainty.
Environment & drift summary
  • temp_bin_low/nom/high
  • phase_drift_ps (min/avg/max)
  • ref_valid_fail_count
Why: temperature-binned summaries expose latent drift and aging trends.
Threshold rules (turn counters into actions)
  • Clock reference unstable: holdover entries exceed a time-window threshold → raise CLOCK_REF_UNSTABLE.
  • Sync invalid: lmfc_slip_count > 0 or alignment reset occurs → raise SYNC_INVALID.
  • Drift trend: phase drift stats worsen across temperature bins → schedule recalibration or service alert.
  • Trigger timing risk: trigger drift/skew stats exceed a trend threshold → flag TRIGGER_TIMING_DRIFT.
Concrete IC examples (commonly used building blocks)
References: Crystek CCHD-957 (low-noise XO), NDK NZ2520SDA (low-phase-noise XO), Abracon AOCJY-10.000MHZ (OCXO).
Jitter cleaners / clock generators: TI LMK04828, ADI AD9528, ADI HMC7044, ADI AD9545, Skyworks Si5345/Si5395.
Synthesizer (wide tuning): TI LMX2594.
Distribution: TI CDCLVP1216.
Clock and sync telemetry closed loop: counters, rules, alarms, export Block diagram from Clock/Sync subsystem to Telemetry counters, Threshold rules, Alarm handling, and Service log export. Includes short tags for what to log and why it matters. F12 · Field evidence closed loop (log → rules → alarm → export) Clock / Sync subsystem Telemetry counters + stats Threshold rules Alarm actions Service log export evidence package What to log • lock/unlock counters • holdover entry + reacquire time • SYSREF missing / LMFC slip • trigger drift + skew stats Why it matters • catches intermittent ref quality loss • proves sync repeatability over time • exposes temperature-linked drift trends • creates serviceable evidence counters · stats · bins Treat clocks like a monitored subsystem: log the right fields, apply clear rules, and export evidence before users notice failures.

Request a Quote

Accepted Formats

pdf, csv, xls, xlsx, zip

Attachment

Drag & drop files here or use the button below.

H2-13 · FAQs × 12 (High-Speed ADC & Clocking)

These FAQs focus on practical decisions and evidence: jitter budgeting, phase-noise interpretation, spur control, deterministic latency, trigger timing stability, and field-ready validation/logging.

1) When is clock jitter the true ENOB/SNR limit (not the ADC)?
Clock jitter becomes the ceiling when the measured SNR drops mainly with higher input frequency while the ADC looks “fine” on low-frequency tests. Use the jitter-limited SNR relationship to back-calculate the maximum allowed RMS jitter at the highest meaningful input frequency. If the budget is already consumed by the clock chain (reference + PLL + fanout + coupling), upgrading the ADC alone will not recover ENOB.
2) How should the integration band be chosen for phase-noise → RMS jitter?
The integration band must match what the sampling system “sees.” Near-offset noise often dominates sampling aperture uncertainty and close-in spurs, while far-offset noise can matter for wideband clocks and aggressive filtering. A good rule is to align the integration range with the clock-conditioning chain and the effective sampling/processing bandwidth. Integrating too narrowly hides real jitter; integrating too broadly can punish irrelevant noise and mislead comparisons.
3) Which spurs are most harmful in GS/s capture, and why?
Spurs are harmful when they create stable, repeatable artifacts that survive averaging and show up as tones or sidebands in FFT-based measurements. Common offenders include reference-related spurs, fractional-N modulation spurs, and supply-coupled spurs that track load or switching states. Unlike broadband jitter, a single strong spur can dominate a narrowband measurement. The most practical approach is a spur sweep across modes, offsets, and temperatures, then enforce limits at the worst cases.
4) Integer-N vs fractional-N vs DDS-assisted: what instrument need drives the choice?
Integer-N is often preferred when the lowest spur risk matters more than fine frequency step size. Fractional-N is chosen when fine steps or fast tuning is required, but it demands strong spur management and careful modulation settings. DDS-assisted architectures are useful when phase control and fine steps are needed without sacrificing a clean main PLL, but images and filtering must be handled. The safest decision sequence is: meet jitter/spur limits first, then optimize step size and tuning behavior.
5) How does PLL loop bandwidth trade lock time versus noise?
Loop bandwidth sets where the output transitions from being reference-dominated to VCO-dominated in phase noise. A wider loop tends to lock faster and track reference changes more tightly, but it can inject more reference noise into the output. A narrower loop can suppress reference noise transfer and rely more on the VCO close-in performance, but it may lock slower and react poorly to disturbances. The practical target is the bandwidth that minimizes integrated jitter while meeting lock-time and mode-switch requirements.
6) What is the practical difference between “narrow cleaning” and “wide tracking” jitter cleaners?
Narrow cleaning prioritizes suppressing incoming reference noise and delivers a very clean output, but it may track reference wander slowly and can increase reacquire time after reference loss. Wide tracking follows the reference more actively and supports fast reacquisition, but it may pass more reference noise through. In instruments, the choice is driven by whether external references are noisy or unstable, and whether fast, predictable recovery is required. Common building blocks include jitter-cleaner clock generators such as LMK04828 or AD9528-class devices.
7) Holdover: what should be monitored, and what is a “good” reacquire?
Holdover should be treated as a controlled state with evidence. Log entry time, duration, and a reacquire-time distribution (min/avg/max). Track whether reacquire introduces a phase hit or any sync reset, because those events can invalidate deterministic timing assumptions. A “good” reacquire is predictable (tight distribution), recoverable without repeated relocks, and produces the same alignment outcome after restart procedures. Devices like AD9545-class DPLLs are commonly used when robust reference translation and holdover behavior are required.
8) Clock tree design: where do additive jitter and skew really come from?
Additive jitter accumulates at every fanout or conversion stage, so the “best” source clock can be degraded by distribution choices. Skew arises from trace length mismatch, device propagation delay differences, temperature drift, and unequal termination conditions. Star distribution reduces cascaded accumulation but needs careful matching; daisy-chain/cascade can simplify routing but may amplify skew and jitter across stages. Keep swing and termination consistent, match where alignment requires it, and isolate noisy return paths that modulate edges.
9) JESD204 synchronization: what does deterministic latency mean and how can it be proven?
Deterministic latency means that after power cycles or link resets, the sample alignment and system timing return to the same repeatable point, not merely the same frequency. Prove it by repeated cold-start runs, capturing alignment indicators and measuring phase/sample-point repeatability across trials. SYSREF distribution must be consistent and arrive inside the valid timing window for each device, and digital deskew should only correct residual, bounded errors. A valid proof includes both “repeatability stats” and clear pass/fail rules.
10) What are common signs of SYSREF distribution problems (missing SYSREF, LMFC slip)?
A missing SYSREF event typically indicates distribution integrity or timing-window issues and often correlates with occasional misalignment after mode changes. LMFC slip or alignment reset counts indicate that deterministic latency is compromised and that time correlation across channels may be invalid. The practical response is to flag data validity, capture a timestamped event record, and optionally force a controlled re-alignment sequence. Reliable systems treat these as “hard evidence events,” not as waveform anomalies to be guessed later.
11) Why do trigger paths need their own jitter/drift statistics in high-speed capture?
Trigger timing is part of the measurement timing chain. Jitter or drift on the trigger conditioner, threshold decision, delay element, or fanout directly shifts acquisition start time. That shift appears as poor repeatability even if the sampling clock itself is clean. Log time-of-arrival drift statistics, skew estimates, and trigger-event jitter estimates, then apply thresholds and trends to trigger alarms. This approach makes timing faults observable and serviceable instead of intermittent “ghost” failures.
12) Which PCB and power mistakes most often create phase-noise sidebands and added jitter?
The most common failures are supply and return-path coupling into the PLL/VCO region, and routing practices that convert switching noise into edge timing modulation. Keep the reference/PLL/clock-tree zone compact and quiet, isolate sensitive rails, and avoid crossing plane splits that break return continuity. Route clocks as controlled differential pairs with consistent impedance and matching where alignment requires it. Also avoid thermal gradients and mechanical stress near the timebase, since temperature-driven drift can masquerade as “random” timing error.
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.