• Main link: TMDS/FRL must lock and stay locked under stress (temperature, cable swaps, hot-plug).
• Sideband: EDID/DDC/HPD/CEC failures can look like “SI problems” but are not.
• Protection: HDCP success is a system state, not a single pin or register.
• Usable link: verify in layers (PHY → Link → Video → HDCP), with measurable pass criteria.

Roles & responsibility boundaries: Source / Sink / Capture

“Link usable” definition: 4-lock model + stability metrics

Scope guardrails (anti-overlap)

• HDCP required? determines authentication paths, test cases, and some capture restrictions.
• FRL required? drives the margin strategy (channel loss/jitter tolerance) and the bring-up plan.
• Compatibility target? (wide display list vs controlled environment) drives EDID/HPD robustness requirements.
• Latency target? impacts buffering, re-sync behavior, and capture pipeline tolerance.

Translate requirements into constraints (method, not a datasheet dump)

Capture-specific constraints (why “locked” can still fail)

Requirements → test plan (prevent surprises)

1. Identify the active mode: TMDS or FRL (from status/logs).
2. Locate the pipeline stage: capability, selection, training/lock, or steady-state.
3. Match the failure fingerprint: snow/sparkles, flicker, black screen, periodic drops.
4. Apply the smallest fix: change one dimension and re-validate pass criteria (X/Y placeholders).

TMDS: practical envelope + failure fingerprints

FRL: training vs steady-state (why failures look like black screen)

Stage classifier: capability → selection → training/lock → steady-state

• Data path: pixel/format → packet/encode → TMDS/FRL → PHY → connector.
• Clock path: reference → PLL → link clocks → PHY clocks (jitter margin).
• Control plane: EDID policy, HPD events, DDC/CEC behavior, HDCP policy and retries.

Data path: what to validate (without turning this into a register map)

• Format coherence: color format, depth, HDR flags, and timing must be consistent across pipeline stages.
• Rate coherence: payload tier must match the negotiated mode (TMDS/FRL) and selected rate ladder.
• Backpressure visibility: detect underrun/overrun events and treat them as first-class root causes.

Clock path: stability beats “pretty waveforms”

• Measurement consistency: verify the same node and the same accounting method before comparing jitter numbers.
• Drift sensitivity: stress with temperature and PSU disturbances to reveal marginal clock paths.
• Pass criteria: mode holds with re-train=0 for ≥ X hours under worst-case stress profile.

Control plane: the hidden cause of “random” re-training

• Mode policy: deterministic selection rules and safe fallbacks (no oscillation between modes).
• HPD behavior: debounce and event logging; correlate HPD edges with link resets.
• DDC reliability: treat EDID read failures as first-order control-plane faults.
• HDCP retries: distinguish authentication failures from main-link margin issues by state and counters.

Bring-up tiers (increase pressure step-by-step)

1. Baseline: known-stable payload, no protection; prove video lock + 4-lock stability.
2. Higher payload: enable HDR/greater depth; confirm deterministic mode selection.
3. FRL tier (if required): reach target rate without training retry storms.
4. Protected content (if required): HDCP state stable under content switching and hot-plug cycles.
5. Stress: temperature sweep + PSU disturbances + cable variation; re-train=0 for ≥ X hours.

• Lock chain: PHY lock → link lock (TMDS/FRL) → video lock → HDCP state.
• Recover chain: soft re-sync → link recovery → session restart (event-driven, rate-limited).
• Output chain: recovered video → buffering/DDR → output interface (CDC/FIFO/backpressure).

Rx front-end: CDR + EQ/CTLE (receiver-only knobs)

Keep tuning local to the receiver: prioritize lock acquisition and steady-state margin. Treat presets as diagnostic tools: a preset that changes the symptom strongly indicates a front-end margin issue.

• Acquisition knobs: CDR mode, lock window, lane lock policy.
• Margin knobs: CTLE level, EQ presets, adaptive enable/disable (if supported).
• Observability: lock flags, retry counters, and time-aligned event logs.

Four-level lock model (define, fingerprint, first check)

Recovery pipeline: controlled, rate-limited, and measurable

Recovery should be a staged state machine to avoid oscillation. Escalate only when lower tiers fail within defined thresholds.

Output chain: CDC/FIFO/backpressure (common “locked but still broken” root cause)

• CDC risk: crossing from recovered video clocks into memory/output domain can create bursts or drift.
• FIFO risk: underrun/overflow produces drop/repeat artifacts that resemble link instability.
• Backpressure risk: downstream throttling forces frame dropping unless explicitly handled.

Rx bring-up tiers (increase pressure step-by-step)

1. Baseline: stable payload; confirm PHY/link/video locks all stable for ≥ X minutes.
2. Higher payload: enable HDR/greater depth; artifacts=0 within Y minutes.
3. FRL (if required): training success ≥ X% over N cycles; no retry storms.
4. HDCP (if required): re-auth=0 over X hours; stable across N content switches.
5. Stress: temperature + PSU disturbance + cable variation; re-train=0 for ≥ X hours.

DDC/EDID: treat it as a reliability channel, not a “background wire”

• Primary symptom: EDID read retry/fail causes unstable negotiation and oscillating mode choices.
• First-check quad: EDID retries, correlation vs hot-plug, correlation vs HPD edges, correlation vs +5V detect edges.
• Pass criteria: EDID read success ≥ X% over N hot-plug cycles; retry ≤ X/hour.

EDID instability fingerprints (symptom → first check → fix direction)

HPD behavior: debounce and rate-limit to prevent re-training loops

• Fingerprint: periodic blanking or repeated mode re-entry without visible cable changes.
• Quick check: HPD edge counter aligned to link re-train or HDCP re-auth timestamps.
• Fix direction: apply debounce window and event cooldown; log every HPD edge with cause tags.
• Pass criteria: HPD toggles ≤ X/hour (non-plug events) and re-train=0 for ≥ X hours.

+5V supply & detect: the hot-plug boundary that can mis-trigger sessions

• Fingerprint: false plug/unplug behavior during load changes or cable movement.
• Quick check: detect-edge counter aligned to EDID/HPD events and session rebuild.
• Fix direction: stabilize detect threshold behavior and avoid noisy boundaries during hot-plug.
• Pass criteria: false-detect events = 0 across N hot-plug cycles and stress profile.

• Layered checks: stop at the first failing layer and record the exact transition point.
• Steady-state checks: measure re-auth frequency and success across content switches.
• Repeater awareness: detect repeater/topology early to avoid misdiagnosis as “link/SI.”

Layer A · Capability (version, roles, repeater declaration)

Start by confirming what the sink declares and what the source attempts. Capability mismatch frequently creates “successful UI output but protected content fails.”

• Observability: negotiated HDCP version, role, and any declared repeater/topology fields.
• Fingerprint: EDID appears normal but protected playback remains black.
• First check: verify the same capability snapshot is used for the entire session (no oscillation).
• Pass criteria: capability snapshot stable for ≥ X minutes (no flips), across N reads.

Layer B · Authentication (handshake, timeouts, retry storms)

• Key idea: the critical information is the breakpoint: which state transition times out or loops.
• Observability: auth state, timeout counters, retry counters, and a timestamped state trace.
• Failure fingerprint: repeated auth attempts with periodic blackouts or mode re-entry.
• Pass criteria: auth completes within ≤ X ms and retries ≤ X per session.

Layer C · Session/Keys (session establishment and continuity)

The most common bring-up trap is “auth appears done,” but the session is rebuilt repeatedly. Confirm whether the session remains stable long enough to enter a clean encrypted steady state.

• Observability: session-id lifecycle, re-auth triggers, and session rebuild reasons.
• Failure fingerprint: playback starts then collapses after mode changes or sideband events.
• Pass criteria: session remains unchanged for ≥ X minutes under normal content switching.

Layer D · Encrypted steady-state (protected playback acceptance)

• Acceptance focus: re-auth rate and success across stress patterns, not one “green” run.
• Measure: re-auth count, protected/unprotected transitions, blackout duration, and failure rate.
• Pass criteria: re-auth ≤ X/hour; blackout duration ≤ X ms; N content switches succeed at ≥ X%.

Repeater/topology: identify early (do not expand switch/matrix details)

• Identification: sink declares repeater role; topology fields (depth/device count) are present.
• Risk fingerprint: downstream changes trigger upstream re-auth, presenting as random blackouts.
• Pass criteria: topology stable (no depth/device-count changes) for ≥ X minutes; re-auth ≤ X/hour.

Symptom → first localization (fast triage)

1. Clock tree: Ref → PLL → divider/mux → PHY clock → serializer.
2. Budget table: contributors + measurement uncertainty → total → margin vs target.
3. Decision path: fix measurement → fix config → fix PDN coupling → trigger retiming (only as a condition).

Clock tree sensitivity: identify the dominant nodes

• Reference: quality and SSC policy influence the entire chain.
• PLL node: loop behavior and rail coupling can convert PDN noise into phase noise.
• PHY clock: the last distribution stage is often the most correlated with link stability.

Measurement definition: different points and windows produce different “truth”

• Point consistency: Ref vs PLL vs PHY measurements are not interchangeable.
• Bandwidth/window consistency: use the same window for steady-state and for training/mode transitions.
• Time alignment: correlate jitter snapshots with training failures, retries, or link drops.

Budget table (card-list form to avoid mobile overflow)

Conditions that justify re-timing (do not expand implementation here)

• Persistent negative margin: budget margin stays < 0 and is highly sensitive to cable length/temperature.
• Rate dependence: failures disappear when stepping down link rate, indicating insufficient timing margin.
• No improvement after low-cost fixes: measurement/config/PDN actions do not change the failure signature.

• Control discontinuities: keep geometry and reference environment consistent.
• Protect return integrity: avoid plane splits and forced return detours.
• Avoid over-tuning: skew matching must not degrade impedance/crosstalk.
• Leave observability: reserve test/repair options near the connector and critical transitions.

Module A · Impedance control (geometry + reference consistency)

HDMI signal quality is dominated by how often impedance “changes.” The priority is fewer geometry changes and fewer reference changes, not only a single nominal value.

• Control points: layer transitions, connector pads, breakout regions, and any width/spacing edits.
• Failure fingerprint: rate-dependent sparkles/flicker that worsens with longer cables.
• Pass criteria: impedance tolerance within ±X% across all control points; discontinuities ≤ X critical spots.

Module B · Return-path continuity (plane splits are a red-line)

Differential routing still needs a nearby reference path. Plane splits force return detours, expanding loop area and creating simultaneous SI and EMI failures.

• Hard rule: do not cross a reference-plane split or void under the pair.
• Layer change rule: provide a short return via path (ground via fence) adjacent to the transition.
• Pass criteria: continuous reference under the full route; all transitions have return support within ≤ X mm.

Module C · Skew & length-match (the “good-enough” principle)

Skew control is a sampling-margin problem. Over-aggressive serpentine routing often increases coupling and discontinuities, worsening the true margin.

• Placement rule: any serpentine should be in a quiet region with stable reference and spacing.
• Pass criteria: skew within X (unit/limit defined by the project), without introducing new discontinuities.

Module D · Vias / connector / cable transitions (reflection-aware design)

Vias and connector breakouts are predictable reflection sources. Designs should be built for measurement and repair, not only for “best-case routing.”

• Via strategy: minimize stubs and keep pair symmetry through transitions.
• Connector breakout: avoid abrupt pad/anti-pad changes; keep reference consistent.
• Observability: reserve differential test access and optional tuning footprints near critical transitions.
• Pass criteria: transition symmetry verified; test/tuning options present for at least X critical areas.

Layout sign-off checklist (fill project thresholds)

• All diff pairs have continuous reference under-route; no plane-split crossings.
• All layer transitions include nearby return support; via symmetry preserved.
• Serpentine only in quiet zones; no forced “length-match at all costs.”
• Critical transitions have measurement and tuning options (targets defined as X).

1. Sideband stable? HPD toggles / DDC errors point to H2-6.
2. Mode domain? TMDS vs FRL, training vs steady-state point to H2-3.
3. Protected-only failure? black only on protected content points to H2-7.
4. Margin-driven? strong dependence on cable/temperature points to H2-8 and H2-9.

Symptom tree A · Black screen (shortest path)

• EDID/DDC unstable → go to H2-6 (DDC pull-ups, noise, HPD behavior).
• EDID OK but no video lock → go to H2-3 (TMDS/FRL selection + training/lock).
• Only protected content black → go to H2-7 (HDCP layered bring-up).
• Strong cable/temperature sensitivity → go to H2-8 then H2-9 (jitter budget + layout discontinuities).

Symptom tree B · Flicker / intermittent drops

• HPD toggles or DDC errors rise → H2-6.
• Periodic re-training signatures → H2-3.
• Thermal or load correlation → H2-8 (clock/PDN coupling).
• Improves with different cable/connector → H2-9 (discontinuities and return integrity).

Symptom tree C · Sparkles / snow (margin-driven errors)

• Confirm mode domain → H2-3 (TMDS vs FRL and training stability).
• If rate downshift reduces errors → H2-8 (jitter budget and margin).
• If cable/connector swap changes behavior → H2-9 (reflection and return-path issues).

Minimum observability set (log and counter checklist)

• Mode/training: TMDS/FRL mode, training state, retry count, lock status.
• Sideband: HPD toggle count, DDC read errors / NACK rate / bus-busy flags.
• HDCP: auth retry count, session rebuild reason, encrypted steady-state indicator.
• Stability: error counters available in the platform, plus temperature/rail event timestamps for correlation.

Bring-up acceptance targets (fill thresholds)

• Training success ≥ X% across N cycles and worst-case conditions.
• HPD toggles ≤ X/hour; DDC errors ≤ X/hour.
• HDCP re-auth ≤ X/hour; protected content switch success ≥ X%.
• Continuous playback ≥ X minutes with no visible artifacts or resets.

• Data path: TMDS/FRL selection, training/lock, steady-state stability (maps to H2-3/H2-8/H2-9).
• Sideband: EDID/DDC/HPD/5V behavior and reliability (maps to H2-6).
• HDCP: authentication, session maintenance, repeater/topology cases (maps to H2-7).
• Stress: temperature/PSU noise/cable variations/compatibility matrix (maps to H2-8/H2-9; sideband triggers map to H2-6).

Design-in test hooks (reserve in schematic + layout)

Hooks are not “extra features.” They enable isolation (Tx vs channel vs Rx) and make CTS failures reproducible. The focus here is what to reserve, not how to implement vendor-specific registers.

Regression strategy (matrix validation, mobile-friendly)

Regression must be matrix-based: swap cable, display, and source, then add temperature and rail disturbance. Large tables are avoided here; list the dimensions and define a minimum coverage set.

• Dimensions: Source × Sink/Display × Cable × Mode (TMDS/FRL) × Temperature × Rail noise events.
• Minimum coverage set: worst cable + most demanding display + highest mode + hot condition + rail perturbation.
• Pass criteria: training success ≥ X%, HPD toggles ≤ X/hour, DDC errors ≤ X/hour, protected switch success ≥ X%.

Gate A · Design Gate (sign-off before layout freeze)

Gate B · Bring-up Gate (must pass before CTS submission)

• B1 · Minimum link: stable video ≥ X minutes with no visible artifacts.
• B2 · Mode coverage: TMDS/FRL and rate ladder transitions succeed ≥ X% over N cycles.
• B3 · Sideband stability: HPD toggles ≤ X/hour, DDC errors ≤ X/hour (logged).
• B4 · HDCP readiness: auth success ≥ X%, protected content switch success ≥ X%, re-auth ≤ X/hour.
• B5 · Long-run stability: continuous run ≥ X hours under normal conditions.
• B6 · Thermal/rail stress: stability maintained through defined thermal sweep and rail events (thresholds X).
• B7 · Failure is classifiable: any failure can be triaged to a domain within ≤ X minutes using logs/counters.

Gate C · Production Gate (consistency + fast test)

• C1 · Batch consistency: multi-board / multi-lot pass rate ≥ X% on the minimum coverage set.
• C2 · Alternate materials: critical alternates validated through the same regression set (no “single-cable only” validation).
• C3 · Cable/connector tolerance: defined cable class set passes with margin; failures are actionable and classifiable.
• C4 · Fast-test definition: production test outputs a clear pass/fail using fixed counters and thresholds (X).
• C5 · Traceability: configuration/version binding and log export path are fixed for field returns.
• C6 · Change control: any BOM/layout/firmware change triggers the locked regression set automatically.