Iridium: prove + harden the wideband path at 6 MS/s

[kevinelliott]

Summary

Triggered by the live Iridium session decoding nothing. Conclusion: the iridium core works correctly at the station's 6 MS/s — the live silence is the antenna/signal (separately confirmed: SAWbird+IR powered and passing the band, but a full-band capture shows zero burst energy; the Maxtena PN100/M1621HCT-EXT1 passive antenna needs clear sky view).

What this PR adds:

• decodes_real_burst_at_station_rates — the wideband decoder had no test above 2 MS/s, yet the station runs it at 6 MS/s (decim 24); that gap is exactly why 6 MS/s correctness was unknown. The test upsamples the real gr-iridium reference burst to 2 and 6 MS/s at several offsets (including near the band edge) and asserts detect + demod with clean PRBS15. It passes — 6 MS/s is correct.
• Noise-floor warmup — the per-bin floor was seeded from a single frame, so many bins started near a noise null and read tens of dB "hot" for the EMA's whole settling time: a startup burst-detection storm, worse at the larger FFTs of higher rates (wasted CPU — CRC rejects them, so no false decodes). Now averages the floor over the first 64 frames (~33 ms) before detecting; real-burst detection is unaffected.

Note for future testing: modulate() is the wrong way to synthesize high-rate test signal — its RRC is a fixed 81 taps (~half a symbol at 6 MS/s sps=240), so it produces garbage above ~2 MS/s. Upsample real bursts instead (as this test and the existing 2 MS/s test do).

Bench gates unchanged.

🤖 Generated with Claude Code

Summary by CodeRabbit

• Bug Fixes

  • Improved wideband burst detection accuracy by implementing a noise floor warmup phase that ensures stable threshold comparisons before detection begins.

• Tests

  • Added comprehensive integration test validating burst detection across multiple sampling rates, frequency offsets, and noise conditions with payload verification.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro

Run ID: 4db2e3e4-4196-4053-a346-061908b71109

📥 Commits

Reviewing files that changed from the base of the PR and between d3d93f8 and c29467b.

📒 Files selected for processing (2)

• crates/xng-mode-iridium/src/wideband.rs
• crates/xng-mode-iridium/tests/wideband.rs

---

📝 Walkthrough

Walkthrough

This PR introduces a noise-floor warmup phase to the wideband FFT burst detector. A new WARMUP_FRAMES constant controls how many initial frames are used to settle the per-bin noise floor via incremental mean before threshold-based detection resumes. The detector tracks warmup progress with floor_frames and suppresses hot-bin detection until warmup completes, then switches to the existing EMA behavior. An integration test validates the behavior with resampled real burst data across multiple station rates and frequency offsets.

Changes

Wideband Burst Detection Warmup Phase

Layer / File(s)	Summary
Warmup phase implementation crates/xng-mode-iridium/src/wideband.rs	Adds WARMUP_FRAMES constant to define the warmup period, introduces floor_frames: u64 field to IridiumWideband to track frames during the settling phase, initializes the field on struct construction, and replaces the original noise-floor initialization and continuous EMA with a two-phase algorithm: frames are accumulated into an incremental mean during warmup (hot bin detection suppressed), then the code switches to symmetric EMA for the normal phase once the floor has stabilized.
Integration test validation crates/xng-mode-iridium/tests/wideband.rs	Adds decodes_real_burst_at_station_rates test that loads the PRBS15 reference fixture, resamples to multiple station rates with frequency offsets, adds pseudorandom noise, runs the wideband detector over chunked buffers, and asserts decoded bursts match the access preamble and have zero PRBS15 bit violations.

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~12 minutes

Poem

A rabbit hops through FFT frames with glee,
Warmup's begun, the noise floor to see!
Let the bins settle 'fore hot ones ignite,
Then burst detection shines clear and bright. ✨📡

🚥 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 clearly and specifically summarizes the main change: adding a test and fixing noise-floor warmup to validate the wideband detection path at the 6 MS/s station sampling rate.
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 iridium-wideband-rates

---

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.}

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: c29467ba03

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]

Preserve bursts during the warmup window

When a stream or file begins with an Iridium burst, or a burst starts within the first 64 FFT frames, this branch updates the floor and continues before setting any hot bins, so no ActiveBurst is created. At the current 6 MS/s FFT size this suppresses detection for about 44 ms, long enough to drop shorter bursts entirely or miss the preamble of longer ones, after which process can drain the samples because there is no active burst; this regresses captures that do not have an initial quiet lead-in.

Useful? React with 👍 / 👎.