Programmable EQ / De-Emphasis for High-Speed Signal Conditioning
← Back to:Active Filters & Signal Conditioning
Programmable EQ / de-emphasis improves eye margin by reshaping frequency and step response to counter predictable channel loss and ISI—while keeping noise, headroom, stability, and production repeatability inside measurable limits.
The right solution starts with separating loss from reflections, then selecting corners/peaking or de-emphasis strength that maximizes eye score without triggering noise peaking, compression, or fixture-dependent “false optima.”
Definition & Scope: Programmable EQ vs De-Emphasis (and where they sit)
Programmable equalization and de-emphasis are often mixed up because both “improve the waveform.” They act at different points in the chain and modify different signal properties. Locking the terminology up front prevents incorrect tuning, wrong measurements, and “preset hunting” that cannot converge.
Key takeaways (terminology locked)
- RX EQ (often CTLE / shelving / peaking) shapes the frequency response H(f) to compensate channel frequency-dependent loss (especially high-frequency attenuation).
- TX Pre/De-Emphasis changes time-domain weighting within a UI (step/impulse emphasis) to reduce ISI by boosting transitions relative to steady levels.
- Phase / group-delay equalization primarily targets phase (τg), not magnitude; it is a different knob class from amplitude EQ and is evaluated with different pass criteria.
Engineer’s mapping: concept → measurable → programmable knob
RX EQ
Measure: S21/IL slope, eye opening vs. frequency content
Program: peaking (dB), corner(s) fz/fp, shelf gain, bypass
Program: peaking (dB), corner(s) fz/fp, shelf gain, bypass
TX De-Emphasis
Measure: step response, UI-level ISI, overshoot/ringing sensitivity
Program: de-emph ratio, transition boost width (≈ UI), preset tables
Program: de-emph ratio, transition boost width (≈ UI), preset tables
“Programmable” levels
Presets (repeatable) → continuous tuning (fine control) → closed-loop calibration (temperature/lot stability)
Scope guardrails (avoid wrong “fixes”)
- Covered here: analog/mixed-signal programmable shaping (CTLE-like EQ, shelf/peaking, de-emph presets), plus measurement, tuning flow, and production hooks.
- Not covered: digital FFE/DFE, CDR/PLL behavior, protocol compliance rules, and full filter-topology derivations.
- Practical rule: TX changes time weighting; RX changes frequency shaping. Both must be judged by a single measurable score (eye opening / margin), not by “looking better” on one instrument setting.
When to use: Symptoms, goals, and what “eye opening” really means
“Eye opening” is not a single knob. It is the outcome of a chain where loss, ISI, reflections, noise, and overload compete. The correct approach is: (1) classify the symptom, (2) translate it into measurable goals, then (3) touch the right knob. Skipping the translation step is the most common reason presets look “random.”
Symptom classes (observable features, not guesses)
- Low eye height: amplitude loss, elevated vertical noise, or compression/limiting in a stage.
- Narrow eye width: dominant ISI (frequency-selective loss) or time uncertainty (jitter), often aggravated by reflections.
- Overshoot / ringing: impedance discontinuity, poor termination, or stability issues; aggressive peaking can amplify it.
- “More peaking makes it worse”: noise peaking, overload, or reflection amplification—do not keep turning the same knob.
Goals (define “better” before tuning)
- Eye height / eye width: direct readouts from the same measurement setup (fixture/probe must remain constant).
- Margin / score function: a single numeric score across presets (mask margin, eye area, or a consistent BER proxy).
- Jitter tolerance (relative): comparable only when test chain and bandwidth are held constant; treat it as a tuning score, not a spec claim.
- Compensation bound: channel insertion-loss in the band-of-interest sets a practical upper limit—beyond it, noise and overload dominate.
When EQ/de-emphasis is limited (stop and fix the real problem)
- Noise-floor limited: peaking lifts high-frequency noise along with signal, shrinking margin even if edges look sharper.
- Reflection-dominated: discontinuities cause multi-path ringing; termination and routing fixes usually beat more peaking.
- Overload/compression: if any stage clips, EQ magnifies distortion; restore headroom before re-tuning.
- Non-repeatable setup: probe/fixture changes shift the optimum; lock the setup before concluding “best preset.”
Channel model essentials: loss, skin effect, dielectric, reflections (minimal but sufficient)
The channel does not need a full electromagnetic textbook to tune EQ correctly. A “sufficient” model only needs to answer three questions: how loss grows with frequency, where the slope changes (the corner), and whether reflections dominate. If reflections dominate, EQ will not converge; impedance continuity and termination must be fixed first.
Minimal loss model (enough to pick corners)
- Insertion loss IL(f) increases with frequency. Practical channels often show a “gentle-to-steep” transition: the corner marks where ISI becomes the dominant limiter for eye width.
- Conductor loss (skin effect) typically drives a steadily increasing slope as frequency rises. Dielectric loss increases the high-frequency penalty further; the combined effect is what EQ tries to counter.
- Use the corner to anchor tuning: place EQ turn-on near where IL(f) starts to accelerate, then cap the high-frequency boost to avoid noise peaking.
Reflections and discontinuities (do not treat them as “more EQ”)
- Reflections create time-domain echoes: overshoot, ringing, and multiple threshold crossings that shrink eye width. These are caused by impedance discontinuities (connectors, vias, stubs, poor return paths, incomplete termination).
- EQ shapes the overall response but cannot “target-cancel” a specific echo delay. Aggressive peaking often amplifies ringing and makes the result worse even if edges look sharper.
- Practical rule: if the waveform shows distinct ringing cycles or repeated zero crossings, prioritize termination and routing fixes before any preset tuning.
Fix order (what converges fast)
Priority 1 — impedance first
Termination, connector/via discontinuities, and return-path continuity determine whether ringing dominates.
Priority 2 — then loss/ISI
Once reflections are controlled, IL(f) slope and corner determine where EQ helps most.
Priority 3 — confirm repeatability
Lock the fixture/probe/loading before declaring a “best preset”; measurement drift can mimic channel changes.
Transfer functions you actually use: shelving, peaking, zero/pole CTLE, and phase side-effects
EQ becomes actionable when it is described as a small set of shapes controlled by a small set of knobs. In practice, most “programmable EQ” options map to three families: HF shelving, peaking, and zero/pole CTLE-like shaping. Each family improves certain channel patterns and carries predictable side-effects that must be budgeted.
Three practical shapes (what they fix best)
HF shelf
Best for smooth loss slopes; simple and stable. Budget noise rise because the entire high-frequency region is lifted.
Peaking
Best near the IL corner where ISI starts to dominate. Risk: noise peaking, overshoot, and stronger phase distortion.
CTLE (zero/pole)
Controlled turn-on (fz) and cap (fp) make it easier to match loss without lifting the far-high-frequency noise too much.
Knob language: fz / fp / gain (no derivation, correct direction)
- fz (turn-on): frequency where boost begins. Align it near where IL(f) becomes steeper.
- fp (cap): frequency where boost stops rising. Use it to prevent excessive far-high-frequency noise peaking.
- Gain / peaking (dB): compensation magnitude. Limit it by headroom and by measured margin, not by “sharpest edge.”
- Direction rules: earlier corner → lower fz; noise gets worse → lower gain or introduce fp sooner; ringing appears → reduce peaking and fix termination.
Side-effects that must be budgeted (why “more boost” can fail)
- Group-delay ripple: phase shaping can reduce eye width even if magnitude looks improved.
- Noise shaping: peaking lifts high-frequency noise along with signal; the margin score can drop.
- Stability and loading: certain presets can excite oscillation or ringing with real fixtures and capacitive loads.
- Overload risk: boosting transitions increases peak swing and can push stages into compression; restore headroom first.
De-Emphasis / Pre-Emphasis: step response shaping and ISI control
TX de-emphasis and pre-emphasis are easiest to tune when treated as measurable step shaping, not as a vague “high-frequency boost.” The intent is to redistribute energy within one unit interval (UI) so the receiver sees less data-dependent tailing (ISI). Correct settings are constrained by channel loss, reflections, headroom, and noise.
What it is (UI energy re-allocation)
- Pre/De-emphasis changes the relative weight of transition content versus steady-level content inside a UI. That reduces the “memory” that previous bits leave on the current bit (ISI).
- On loss-dominated links, the primary benefit is cleaner threshold crossings and less data-dependent distortion, which expands usable eye width.
- On reflection-dominated links, the same boost can amplify ringing; impedance and termination must be addressed before stronger presets are attempted.
Parameters to tune (ratio, width, cursor intent)
- De-emph ratio (dB or linear): transition amplitude/energy relative to the steady level. Higher ratio can reduce ISI but raises peaks.
- Transition boost width (≈ UI scale): how long the boost “holds.” Too short under-compensates long tails; too long can distort baseline and worsen margin.
- Cursor meaning: the main transition is strengthened while adjacent UI content is suppressed/enhanced to counter the channel tail (ISI). Treat cursor choices as time-domain weighting aligned to the channel’s memory length.
Match to channel (where it fails)
Good match
Loss/ISI dominates: edge is slow and tails are long. De-emph reduces tail impact and improves consistent crossing.
Common failure
Reflection/ringing visible: stronger transition boost increases overshoot and repeated crossings; fix termination first.
Hard constraints
Peak swing and noise set the ceiling: excessive boost can overload stages and reduce margin even if edges appear sharper.
Programmability mechanisms: switches, digipots, DAC-gm tuning, varactors, SC options (trade-offs)
“Programmable” does not only mean “adjustable.” The mechanism determines bandwidth, linearity, noise, repeatability, and calibration burden. Fast links tend to be limited by parasitics and voltage-dependent nonlinearity, while production systems are limited by drift and repeatability across temperature and lots.
Selection axes (use the same yardstick)
Bandwidth & parasitics
Switch capacitance, routing, and element Q determine whether the intended response survives at high frequency.
Linearity & noise
Voltage-dependent resistance/capacitance and bias-dependent gm can convert “tuning range” into distortion and margin loss.
Repeatability & calibration
Temperature drift and code dependence decide whether the “best setting” remains best without frequent re-calibration.
Mechanisms (what to watch in high-speed chains)
- Switchable R/C arrays: repeatable presets; watch switch parasitics, Ron nonlinearity, and injection.
- Digipots: convenient and code-driven; watch bandwidth, wiper behavior, noise, and temperature/code dependence.
- DAC-controlled gm: continuous tuning and closed-loop search; watch gm drift, bias-point nonlinearity, DAC noise injection.
- Varactors / variable elements: wide tuning range; watch strong voltage dependence and distortion in large-swing paths.
- Switched-capacitor options: accurate ratios and repeatability; watch clock feedthrough, folding, and clock coupling.
Production hooks (what makes “best preset” repeatable)
- Provide bypass/loopback paths so response and margin can be measured without probe-dependent ambiguity.
- Store coefficients/presets in EEPROM/flash and validate at key temperatures to ensure drift does not flip the optimum.
- Define pass criteria as score distribution (margin consistency) rather than “pretty curves” on one bench capture.
Architectures: CTLE blocks, differential EQ, cascaded sections, and “do not accidentally build an oscillator”
Practical programmable EQ is usually implemented as CTLE-style continuous-time shaping, most often in a differential chain. The architecture decision is rarely “one best filter”; it is about where gain/peaking is distributed, how common-mode is controlled, and how stability is protected under real loads.
CTLE block basics (interfaces first)
- Input conditions matter: source impedance, termination, and biasing change the realized response and peaking.
- Output conditions matter: real loads (ADC/limiter/next stage) and stray capacitance can reduce phase margin and trigger ringing.
- Common-mode control in differential CTLEs sets linear range and symmetry; unstable or drifting CM can shrink eye height.
- A bypass path is a design requirement: it enables A/B verification, production screening, and safe fallback.
Cascading strategy (many small boosts usually win)
Single large peaking
Fast to try, but sensitive: higher peak swing, stronger noise peaking, and more exposure to parasitics and load-dependent phase loss.
Cascaded small peaking
Better headroom control: each stage stays closer to linear operation while the overall response matches the channel more smoothly.
Cost to budget
Noise and mismatch accumulate across stages; guardband is required across load, temperature, and production variation.
Stability guardrails (avoid “accidental oscillators”)
- Treat capacitance at the output (routing, probes, ESD, ADC input) as a stability risk; some presets will only fail under certain loads.
- Keep per-stage peaking bounded; if large boost is required, prefer multiple stages and cap far-high-frequency gain.
- Ensure CM loop and differential loop do not fight each other; drifting CM can look like “random EQ instability.”
- If ringing appears only in some presets, treat it as phase margin collapse before blaming the channel model.
Design flow: from target eye margin → choose EQ shape → set corner(s) → allocate noise/headroom
A programmable EQ setting is only “best” when it improves margin under real constraints. The practical flow is: classify the dominant problem, choose a response shape, align corners to the channel, then budget noise and headroom with stability and temperature guardband. The output should be a repeatable parameter set and a verification plan.
Step 1 — Measure and classify (loss/ISI vs reflection vs overload)
- Loss/ISI: slow edges and long tails; IL(f) grows steeply near a corner.
- Reflection: visible ringing cycles or repeated crossings; fixing termination outranks more EQ.
- Overload/noise: clipped tops, compressed transitions, or margin limited by noise floor.
Step 2 — Choose shape and cap the boost
- Shelf for smooth loss slopes.
- Peaking for a strong corner (but cap far-high-frequency gain).
- CTLE (fz/fp) when controlled turn-on and capping are needed.
- Set a maximum allowed boost from headroom and noise budget before tuning details.
Step 3 — Place corners and decide stage count
- Align fz near where IL(f) begins to steepen.
- Use fp to cap the boost and reduce noise peaking sensitivity.
- If required peaking is large, prefer cascaded small boosts with per-stage limits.
Step 4 — Allocate noise and headroom (budget, do not guess)
- EQ raises high-frequency signal and high-frequency noise; measure margin impact, not just edge speed.
- Check peak swing against linear range; compression can erase eye height.
- Confirm stability across expected load capacitance and probe/fixture variations.
Step 5 — Guardband and verify (repeatable outputs)
- Lock the parameter table: presets, fz/fp, per-stage limits, and bypass mode.
- Lock the budget table: noise and headroom margins reserved for drift and variation.
- Lock the verification plan: A/B with bypass, temperature points, and stability checks under worst-case loads.
Measurement & tuning: eye diagram, S-parameters, TDR, swept presets, and closed-loop calibration hooks
Tuning programmable EQ becomes repeatable only when measurement, scoring, and configuration management are treated as a loop. The goal is to classify the dominant limitation (loss/ISI vs reflection vs overload/noise), sweep presets efficiently, and lock results into production-friendly artifacts.
Measurement ladder: with VNA vs without VNA
With VNA
- Use S21 to map loss slope and corner regions.
- Use S11/S22 to flag reflection-dominant cases.
- Output: IL(f) corner + reflection status label.
Without VNA
- Use eye/step captures to compare relative high-frequency loss.
- Use TDR/step reflections to locate discontinuities by delay.
- Output: classify loss/ISI vs reflection, then set safe boost limits.
Eye metrics: convert “looks better” into a score
- Eye height tracks vertical noise margin and compression effects.
- Eye width (or a jitter proxy) tracks timing margin and ringing-driven crossings.
- Overstress flags mark overload, multi-crossing ringing, or unstable presets.
- Use a weighted score to compare presets consistently across benches and temperatures.
Tuning strategy: coarse sweep → local refine → temperature re-check
- Coarse sweep: scan a limited set of presets to find the best score cluster quickly.
- Local refine: adjust corners/peaking around the Top-N candidates without exceeding boost caps.
- Temperature re-check: confirm the best setting remains stable or build a small temp map.
Closed-loop hooks: bypass, loopback, EEPROM, and config versioning
- Bypass / loopback enables A/B isolation between channel effects and EQ settings.
- EEPROM configuration should store preset ID, corner codes, and guardband limits.
- Version fields prevent silent regressions after firmware or production changes.
Required artifacts (production-ready outputs)
Preset score table
One row per preset: height, width, flags, score, temperature bin.
Final parameter table
Selected preset, corner codes, per-stage caps, bypass mode, CM mode.
Verification plan
A/B with bypass, worst-case load, probe variation, and temperature points.
Pitfalls: noise peaking, saturation/overload, EMI/ESD interaction, impedance mismatch, and layout parasitics
Most “EQ made it worse” failures are predictable: the same boost that compensates loss can also amplify noise, push stages into compression, expose reflections, or magnify protection and layout parasitics. Use symptom-driven triage to converge quickly.
Noise peaking (faster edges, worse margin)
Mechanism
Peaking boosts high-frequency signal and high-frequency noise; strong boosts can create a narrow noise peak that erases eye height.
Quick check
Edges look steeper but eye height drops; jitter proxy worsens; results change strongly with measurement bandwidth.
Fix direction
Cap far-HF gain (fp), distribute boost across stages, and increase score penalties for noise/jitter.
Saturation / overload (transient overswing causes compression)
Mechanism
Large peaking increases transition energy; the front-end may clip or compress briefly, shrinking eye height and slowing recovery.
Quick check
Only some presets show flattened tops, sticky recovery, or extra crossings after overshoot.
Fix direction
Lower per-stage boost, increase stage count, and enforce headroom caps before refining corners.
Impedance mismatch / reflections (EQ cannot “erase echoes”)
Mechanism
Reflections are time-domain echoes from discontinuities; EQ shapes spectrum but cannot remove echo timing and may amplify ringing.
Quick check
TDR/step shows clear reflections; eye exhibits periodic ringing and repeated crossings.
Fix direction
Fix termination/connectors/trace transitions first; re-enable EQ only after ringing is reduced.
Protection interaction (TVS/RC adds poles and nonlinearity)
Mechanism
TVS capacitance and clamp nonlinearity reshape amplitude/phase; strong EQ can magnify the extra pole and distortion.
Quick check
“Best preset” shifts after changing TVS/RC parts, voltage, or temperature; some settings become unstable only with protection populated.
Fix direction
Minimize effective capacitance, keep current paths controlled, and validate response with protection installed (details belong to the clamp/ESD page).
Layout parasitics (asymmetry and return paths)
Mechanism
Differential imbalance and discontinuous return paths create unintended filtering and coupling that distort phase and symmetry.
Quick check
Behavior follows the PCB (not the channel cable); probe placement changes results; polarity swap exposes asymmetry.
Fix direction
Enforce symmetry, shorten connector-to-EQ distance, and keep return paths continuous (deep layout rules belong to the layout page).
Engineering checklist: schematic, layout, bring-up tests, production consistency
This checklist turns programmable EQ tuning into a reviewable and repeatable engineering process. It focuses on bypassability, measurability, controlled parasitics, predictable presets, and production-grade configuration traceability.
A) Schematic hooks (bypass, test points, safe defaults)
- Hard bypass path across the EQ chain (A/B isolation must be possible).
- Test points at input, stage outputs, limiter/driver input, and decision/sampling node.
- Power-up preset defined as a conservative “safe” state; recovery path defined for field service.
B) Programmability integrity (codes must map to physics)
- Code-to-response map documented (preset ID → peaking / shelf / corner codes).
- Repeatability plan across temperature and units (guardband for drift and parasitics).
- Config versioning built-in (prevents silent regressions after firmware or BOM changes).
C) Termination and protection placement (avoid “EQ fights echoes”)
- Termination placed to control reflection at the correct boundary (connector/fixture rules).
- Protection networks treated as poles/nonlinearity; validate response with protection populated.
- Return paths for clamps/TVS are reviewed as “signal path,” not an afterthought.
D) PCB review (symmetry, planes, via strategy)
- Differential symmetry: matched topology, consistent via count, minimal imbalance.
- Reference planes: continuous return; avoid slot crossings and long detours.
- Connector-to-EQ distance: keep short; parasitics scale quickly at high edge rates.
E) Bring-up tests (sequence prevents blind tuning)
- Validate the channel first: basic eye/step/TDR, confirm reflections are controlled.
- Validate linearity next: cap boost, check overload signs before hunting best presets.
- Tune last: sweep presets with a score; refine locally; re-check temperature points.
F) Measurement setup control (fixtures and probes are part of the system)
- Probe/fixture identity tracked (revision, insertion loss, reflection behavior).
- Bandwidth and loading controlled; capture settings documented with each tuning record.
- A/B bypass always run with the same setup to isolate EQ impact reliably.
G) Production consistency (temp bins, guardbands, config locking)
- Temp strategy: single preset across temp if stable; otherwise a small temp map with hysteresis.
- Guardbands: caps on peaking and corner shifts to avoid noise peaking and overload.
- Config locking: store preset + limits + version fields; verify after programming.
H) Serviceability (field-safe fallback and traceability)
- Fallback mode: conservative preset always available for recovery and diagnostics.
- Event logging: flags for overload/ringing/noise peaking, plus temperature and supply state.
- Reproducibility: field reports must include config version and fixture/probe identity.
Review checklist table (use for design reviews)
| Item | Why it matters | Quick check | Pass criteria |
|---|---|---|---|
| Bypass path | Separates channel issues from EQ issues quickly. | Run A/B bypass with identical setup and stimulus. | A/B delta matches expectation; no hidden saturation or ringing introduced. |
| Test points | Enables stage-by-stage isolation and root-cause ownership. | Probe input + stage outputs + decision node; compare to bypass. | All nodes measurable without disturbing the system (no “probe-only” failures). |
| Preset safe default | Prevents power-up overload and unstable settings. | Power-cycle and confirm safe state before tuning. | No clipping signs; score above baseline; tuning starts from repeatable state. |
| Diff symmetry + planes | Avoids unintended filtering/coupling and phase imbalance. | Swap polarity / swap lanes; verify behavior follows the channel, not the PCB. | No strong lane-specific artifacts; ringing/jitter not dominated by layout asymmetry. |
| Config versioning | Ensures traceability across firmware, BOM, and fixture revisions. | Log config_id + build_rev + temp_bin with each tuning result. | Any field issue can be reproduced by reloading the recorded config. |
Production data schema (minimal fields that enable traceability)
| Field | Type | Example | Used for |
|---|---|---|---|
| serial_id | string | SN-2026-000123 | Unit tracking |
| fixture_rev | string | FX-R3 | Measurement consistency |
| temp_bin | enum | -20C / 25C / 85C | Temp mapping |
| config_id | string | EQ-CTLE-07 | Traceability + rollback |
| preset_id | int | 12 | Re-load tuning |
| score | float | 0.83 | Preset ranking |
| flags | bitmask | OVERLOAD|RINGING | Fast triage |
Applications: high-speed test, cable/backplane interfaces, ADC capture front-ends, and compliance fixtures
Applications differ mainly by what “best” means: reproducibility for fixtures, BER margin for long channels, and linear SNR/headroom for ADC capture. This section focuses on system block combinations and verification priorities (not protocol details).
A) High-speed test & compliance fixtures
Typical chain
Pattern/Source → Fixture/Probe → Termination → EQ preset → Limiter/Driver → Scope/Comparator/ADC
- Primary goal: stable and comparable measurements across fixture revisions and operators.
- Key risk: fixture loss and group-delay ripple dominate, so “more boost” can reduce repeatability.
- Verification: fixture identity + A/B bypass + score table stored with setup bandwidth and loading.
B) Cable / backplane interfaces (long channels)
Typical chain
TX pre/de-emph → Connector → Cable/Backplane → Termination → CTLE/EQ → Limiter/Decision
- Primary goal: maximize margin without triggering noise peaking or overload.
- Priority rule: if reflections dominate, fix discontinuities/termination before pushing peaking.
- Verification: TDR-based echo check + preset sweep (coarse→refine) + temperature stability re-check.
C) ADC capture front-ends (capture fidelity first)
Typical chain
Source/Link → Termination → EQ/CTLE → Driver/FDA → ADC (S/H)
- Primary goal: linearity, SNR, and headroom; eye opening alone is not a sufficient objective.
- Key warning: EQ can amplify high-frequency noise and push the driver/ADC front-end into compression.
- Boundary: EQ compensates the channel; anti-alias/reconstruction targets sampling-system requirements (handled in the anti-alias/reconstruction section).
What changes across applications
Best = reproducible
Fixtures prioritize repeatability and traceability over maximum boost.
Best = margin
Long channels prioritize margin while keeping reflections controlled.
Best = fidelity
ADC capture prioritizes linear headroom and noise budgets; boost caps are essential.
H2-13. IC selection logic: vendor questions + specs mapped to eye-opening targets
This section turns “programmable EQ / de-emphasis” selection into a repeatable RFQ checklist. The goal is not “best typical curves,” but measurable eye-margin improvement with repeatable presets, known noise/linearity costs, and production-friendly test hooks.
A) Target input template (fill this before comparing parts)
Vendors can only answer accurately if the operating point is explicit. Use the table below as the fixed “conditions header” for every question.
| Item | Unit | Your value | Notes (keep protocol-agnostic) |
|---|---|---|---|
| Data rate / edge content | Gbps (or rise time) | X | Use UI time or 20–80% rise time for “edge difficulty”. |
| Channel loss reference | dB @ f | X dB @ X GHz | Use S21 if available; otherwise quote insertion-loss at a single anchor frequency. |
| Target output swing | mVpp (diff) | X | Linearity specs must be tied to this swing and load. |
| Load / termination | Ω / pF | X Ω, X pF | Include fixture/probe capacitance and any series isolation resistor. |
| Common-mode plan (diff chains) | V | VOCM = X | Ask for VOCM accuracy and range under boost conditions. |
| Environment & production | °C / lots | X to X °C | Repeatability across temp & lot matters more than typical peak performance. |
B) Vendor question rules (to prevent “good typicals, bad reality”)
Always bind every spec to conditions: supply, swing, load (Ω + pF), temperature, and boost/de-emphasis setting.
Ask for “boosted-mode” noise/linearity: peaking improves the eye only if it does not push the chain into noise peaking or overload.
Demand repeatability: step size, code-to-code variation, temp drift, and lot-to-lot spread for key corners (fz/fp/peaking/VOCM).
Require test hooks: bypass/loopback, register readback, config lock, and a way to log the final preset in production.
C) Target → datasheet spec mapping (what to ask, and why)
Treat EQ as a controlled trade: eye margin gained vs noise/linearity/stability cost. Use the mapping below to keep every vendor answer tied to a target.
| Target | Key specs to request | Ask vendor (exact wording) | Pass criteria (placeholders) |
|---|---|---|---|
| Eye height ↑ | peaking range (dB), fz/fp range, output swing linearity, integrated noise (boosted) | “Provide eye/score improvement vs peaking at Vpp=X, RL=XΩ, Temp=X. Include output compression limit and boosted-mode noise.” | Eye height margin ≥ X (or score ≥ X), no compression at target swing. |
| Eye width ↑ | jitter transfer / residual DJ (as available), bandwidth under boost, group-delay ripple (if provided) | “Share jitter/eye width impact for each preset across Temp and lot. Provide recommended guardband against noise peaking.” | Eye width margin ≥ X UI and stable across temp/lot. |
| IL compensation | max HF boost, tuning resolution, corner coverage (fz/fp), stability with real load | “For IL=X dB@X GHz, propose presets and the fz/fp used. Confirm stability for RL=XΩ and CL=X pF.” | Meets IL target with ≤ X dB noise penalty; no oscillation/ringing growth. |
| Production repeatability | step size, code repeatability, temp drift of corners, config retention/lock | “Provide worst-case drift/spread of peaking and corner vs temperature and lot. Provide readback/lock mechanism and test mode.” | Preset-to-preset variation within ±X; drift within ±X% across temp. |
D) RFQ field list (copy/paste table)
Ask vendors to answer in this format. Empty cells are a red flag. All results must be tied to the target conditions from section A.
| Group | Field | Condition to bind | Vendor answer | Pass criteria |
|---|---|---|---|---|
| EQ capability | Max peaking / shelf (dB) + usable range | Data rate, Vpp, RL/CL | [ ] | Peaking ≥ X dB without overload |
| EQ capability | Corner ranges (fz / fp) + step size | Temp range, coding mode | [ ] | Covers IL breakpoints with guardband |
| Linearity | THD / SFDR under boost | Vpp, RL/CL, freq point(s) | [ ] | SFDR ≥ X dBc @ target swing |
| Noise | Input-referred noise + integrated noise | Bandwidth, boost setting | [ ] | Noise penalty ≤ X dB for needed boost |
| Diff chain | VOCM range/accuracy; output CM compliance | Supply, Vpp, RL | [ ] | CM stays within X mV across presets |
| Stability | Load limits and isolation recommendations | CL (fixture), series R | [ ] | No ringing growth; no oscillation |
| Repeatability | Preset drift vs temperature + lot spread | Temp min/max, lots | [ ] | Drift ≤ ±X% (fz/fp/peaking) |
| Testability | Bypass/loopback, register readback, config lock | Production flow | [ ] | Preset traceable + stable after power cycle |
E) Reference examples (specific part numbers; starting points only)
These part numbers are listed to speed up datasheet lookup and vendor Q&A. Final selection must be driven by the RFQ fields above (worst-case, conditions, and guardband), not by protocol labels in marketing names.
Linear redrivers with RX EQ (CTLE) and/or TX de-emphasis
- TI DS100BR410 — quad redriver; receive EQ (CTLE) + adjustable transmit de-emphasis; pin/SMBus programmability.
- TI DS100BR111 — 2-ch repeater; input equalization + output de-emphasis (ultra-low power family).
- TI SN75LVPE4410 — quad linear redriver; CTLE input equalization with linear output driver.
Programmable equalizer/redriver (multi-vendor)
- Analog Devices / Maxim MAX14950 — quad PCIe equalizer/redriver with programmable input equalization and output redrive.
- Analog Devices / Maxim MAX14954 — quad PCIe equalizer/redriver (same concept family; verify the exact variant fit).
- Diodes / Pericom PI3UPI1608 — 8-diff-channel linear redriver; programmable CTLE/flat gain/output swing.
- Diodes / Pericom PI3EQX8904 — 4-channel linear redriver; programmable linear equalization/output swing/gain.
- NXP PTN3944 — multi-channel (x4) linear equalizer optimized for very high-speed interfaces.
Interface-focused “EQ boxes” useful for fixtures and lab setups
- Microchip EQCO510 — USB 3.2 reclocker/redriver (useful as a programmable SI conditioner in compliant fixtures).
- Diodes PI2DPX1263 — DisplayPort redriver; I²C programmable CTLE/flat gain/linearity controls per channel.
Selection reminder
Prefer parts that publish or can provide: (1) boosted-mode noise/linearity, (2) load/stability limits, (3) preset drift vs temperature and lot, (4) bypass/readback/lock for production traceability.
Request a Quote
H2-13. FAQs (10–12) — troubleshooting & production closure
These FAQs close long-tail issues without expanding the main content. Each answer is structured for execution: Likely cause → Quick check → Fix → Pass criteria (with measurable placeholders).
