VDL2: gated noise estimator + burst-length sanity cap (16-17 → 19)

[kevinelliott]

The session's persistence pays: re-auditing the dumpvdl2 oracle showed 40/41 of its frames at 24-30 dB SNR — the missing frames were strong, not weak, overturning three investigations' framing. Instrumentation then found the real faults:

1. False-lock monsters: bogus headers passing the 25-bit FEC with lengths up to 112,241 bits (3.4 s of "collection"). Capped at 16k bits.
2. Noise-floor inflation: the energy gate's EMA learned from burst power during post-rewind rescans, shadowing rapid back-to-back bursts (XID/ack sequences — exactly the capture's pattern). Now a gated estimator: learns only below the gate, tiny up-creep for re-convergence.

rate	before	after
50 kS/s	16	19
100 kS/s	17	19
600 kS/s	16	18

…and the count is now flat across EF 8–20 / threshold 0.4–0.6 where the old estimator wobbled 17–20. Full suite green. Next step documented: time-aligned burst diff against the oracle for the remaining 19-vs-41.

🤖 Generated with Claude Code

Summary by CodeRabbit

• Bug Fixes

  • Improved VDL2 demodulator noise floor estimation to prevent degradation from high-energy signals.
  • Added filtering for invalid transmission lengths during burst collection.
  • Added optional debug logging for demodulation troubleshooting.

• Documentation

  • Added investigation notes documenting signal handling improvements and frame detection results.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 79e8f626ce

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Raise the TL cap to the protocol limit

For any standards-valid VDL2 burst whose decoded transmission length is above 16,000 bits, this new filter turns the burst into a header failure before collection/deinterleaving even though the local PHY code still defines the 17-bit header maximum as 131,071 bits (header::MAX_TL) and interleave::layout accepts lengths up to that value. That makes the demod silently drop larger AVLC/ATN transfers; if the goal is to reject false-lock monsters, the sanity check should not be below the protocol-supported range or should be based on a configurable/validated operational limit.

Useful? React with 👍 / 👎.

[coderabbitai]

Caution

Review failed

Pull request was closed or merged during review

📝 Walkthrough

Walkthrough

The UW-hunting demodulator's noise-floor estimator was gated to prevent learning from high-energy burst samples. Header decoding now rejects transmission lengths above 16,000. Debug logging was added for RS decoding outcomes. Investigation documentation reports updated frame counts and root causes.

Changes

Noise-floor gating and false-lock prevention

Layer / File(s)	Summary
Noise-floor estimator gating crates/xng-mode-vdl2/src/demod.rs	ENERGY_FACTOR increased from 6.0 to 12.0; noise estimator only EWMA-updates when sample power is below the noise-derived threshold; otherwise performs a small up-creep to prevent inflation from burst energy.
Header transmission length validation crates/xng-mode-vdl2/src/demod.rs	Decoded transmission lengths are filtered to reject values above 16,000, causing invalid headers to be treated as decode failures.
Debug logging for RS decoding crates/xng-mode-vdl2/src/demod.rs	VDL2_DEBUG-gated logging reports RS success and failure outcomes with transmission-length bits, symbol count, and soft metric.
Investigation documentation docs/notes/VDL2-DEMOD-V2.md	New section explains false-lock root causes (false-lock length handling and gated energy-EMA learning), reports revised frame counts (19 frames), and identifies next steps as timestamp-based burst alignment.

Estimated code review effort

🎯 3 (Moderate) | ⏱️ ~20 minutes

Poem

In bursts so bright, the gates did fail,
Learning noise from the signal's tale,
We tuned the threshold, raised the bar,
And filtered false-locks near and far.
Now logs can peek behind the veil—
A cleaner demod, our debugging tale. 🐰✨

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title directly and specifically describes the two main changes: a gated noise estimator and a burst-length sanity cap, with measured frame count improvements (16-17 → 19).
Docstring Coverage	✅ Passed	Docstring coverage is 100.00% which is sufficient. The required threshold is 80.00%.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch vdl2-skew

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}