ITU compliance for the AEC: the Rev 5 plan and the Stage 0 requirements matrix

[tap]

Docs-only PR (no code): the plan and the foundation document for the ITU compliance effort.

HANDOFF Rev 5 — the plan

Six stages to prove the AEC chain passes the applicable ITU specifications with asserted margin: requirements matrix → calibrated ITU signal layer → residual-echo post-filter + comfort noise → compliance test suite → proof notebook → external/docs integration. Scope decisions recorded: anchors are the automotive series (P.1100/P.1110/P.1120) + P.340 + a G.168-adapted battery + G.167 as historical tier; the post-filter is approved; margin policy is ≥ 6 dB on level clauses / ≤ ½ time on time clauses, asserted in CI so margin erosion fails, not just violation.

Stage 0 — docs/itu-compliance.md

The requirements matrix, distilled from the recommendation texts themselves (all read end-to-end this session: P.1110/P.1100 03/2017, P.1120 10/2025, P.340 05/2000, P.501 04/2025, G.168 04/2015, G.167 03/1993, G.131 11/2003):

• Tier A — automotive battery + P.340 framework (~30 rows): TCL ≥ 46 dB (target ≥ 52), residual echo < −58 dBm0(A) (target < −64), convergence mask ERL ≥ 40 dB by 1.2 s (target 600 ms), spectral echo masks, switching dynamics, P.340 Category 1 full-duplex double-talk targets (≤ 3 dB talker attenuation, ≥ 27 dB per-band echo loss), comfort-noise tolerances, and the Annex E closed-loop stability sweep (far-end ERL 50→0 dB, no howling).
• Tier B — G.168-adapted battery (12 rows): convergence masks, four double-talk tests, leak rate, infinite-return-loss, tone stability, path swings, comfort noise — labeled adapted (G.168 disclaims acoustic scope); the seven line-only tests excluded by name.
• Tier C — G.167 historical rows: run and reported, not claimed (withdrawn rec, provisional values).

Each row: clause, requirement, required value, margin target, ALGO/MIXED/TERMINAL scope tag, owning test name (row IDs become gtest names). Plus the Stage 1 signal/asset build list (CSS generators from the P.501 sample tables, AM-FM orthogonal double-talk pair, dBm0/dBPa/dBov calibration, cabin and time-varying echo paths) and the Stage 2 design constraints the rows pin.

Grounding for the simulation approach, from the texts: P.340 (10.3.1) and G.167 (5.2.3.1) explicitly permit simulated echo paths; P.1110 clause 8 defines digital insertion interfaces so "no HATS is necessary"; and P.1120 (10/2025) normatively cites P.340 (2000) — the edition in hand — resolving the P.340-2019 procurement gap for every operative table.

Known gaps recorded in the matrix: three tests use ITU real-speech attachment WAVs (Stage 1 will fetch for git-ignored local use or run documented synthetic stand-ins marked method-equivalent); P.502 is not on the free portal (methods reconstructed from the citing test descriptions).

🤖 Generated with Claude Code

https://claude.ai/code/session_01TLE2rvptWxV8catE8vTvgx

---

Generated by Claude Code