feat(rtty): add built-in RTTY (Baudot/ITA2) decoder (#1392)

[motoham88]

Summary

• New RttyDecoder class: mark/space bandpass filters (3× baud BW), fast envelope (tau = 0.3/baud), Schmitt trigger hysteresis (10%), stop-bit validation, full Baudot/ITA2 decode with LTRS/FIGS shift tracking
• RTTY decode panel in PanadapterApplet: appears automatically when a slice enters RTTY or DIGL mode, hides on exit
• Panel controls: fldigi-standard mark frequency selector (Auto/2125/2210/1700/1275/1000/915/850/500 Hz), shift (45–850 Hz), baud rate (45.45–300), REV toggle for high-space/LSB signals
• M%/S%/SNR/LOCKED stats display updated ~2×/second
• All settings persist to AppSettings across sessions
• "Auto" mark frequency follows the radio's rttyMark (RTTY mode) or diglOffset (DIGL mode) live, including mid-session changes
• Confirmed working at 2125 Hz and 915 Hz tone sets, both polarities

Fixes included in this branch

• CR/LF rendered as <br> in the HTML text panel (were collapsed as whitespace)
• Spaces escaped as   (were collapsed by HTML inline rendering)
• refreshRttyDecodeState called on all mode-change paths, not just active-slice change
• Auto mark tracks live rttyMarkChanged / diglOffsetChanged signals
• Decoder not started when panel applet is not yet wired
• FIGS shift state (figsMode) preserved across parameter changes
• AppSettings keys flattened — keys containing / are silently dropped by the XML writer; all four RTTY keys were affected, preventing any settings from persisting

Test plan

• Switch a slice to RTTY mode — panel appears; switch away — panel hides
• Switch to DIGL mode — panel appears with Auto mark tracking diglOffset
• Tune to a 45.45 baud 2125/1955 Hz signal (Rev off) — text decodes with line breaks
• Tune to a reversed-polarity signal — enable REV, confirm decode; relaunch and confirm REV is remembered
• Change mark freq combo to 915 Hz, confirm decode on a 915 Hz signal
• Confirm M%/S%/SNR stats update and LOCKED appears on a live signal
• Change baud rate to 50 — confirm decoder restarts with new rate

🤖 Generated with Claude Code

[aethersdr-agent]

Thanks for the contribution @motoham88 — this is a clean, well-scoped RTTY decoder with thoughtful DSP choices (3× baud bandpass, 0.3/baud envelope, 10% Schmitt hysteresis, 1.5-bit-period sampling alignment) and the user-facing details (fldigi-standard mark/shift/baud presets, REV toggle, Auto-mark tracking rttyMark/diglOffset, stats display, settings persistence) are well thought through. The Baudot tables look correct and FIGS state is preserved across parameter changes as the comment claims.

A few things worth addressing before merge:

1. STREAMDECK_PLUS_SESSION.md is out of scope

The 170-line StreamDeck+ session summary at the repo root has nothing to do with RTTY and looks like a stray commit (fc54bd2a) — it documents a previously-merged feature (PRs #3236 / #3250). Please drop it from this branch. Repo-root markdown is also a maintainer-owned path.

2. DIGL also activates the RTTY decoder

refreshRttyDecodeState() treats RTTY and DIGL identically:

const bool isRtty = s && (s->mode() == "RTTY" || s->mode() == "DIGL");

DIGL is a general LSB-data mode the radio uses for any digital protocol — PSK, JT/FT modes, etc. — not just FSK/RTTY. So tuning a PSK31 or FT8 station in DIGL will pop the RTTY panel and run a Baudot decoder on the audio. The ✕ closes it for that session, but the next slice/mode refresh re-shows it. Is the intent to surface RTTY only for explicit RTTY mode, and treat DIGL as opt-in? Worth at least mentioning in the panel title (e.g. "RTTY decode — DIGL") so the operator knows what they're looking at.

3. RttyDecoder::stop() — wait timeout silently abandons the worker

if (m_workerThread) {
    m_workerThread->wait(2000);
    m_workerThread = nullptr;
}

If wait(2000) returns due to timeout rather than thread completion, the worker is still running but m_workerThread gets nulled. A subsequent start() would then create a second worker, and both would race on m_ringBuf / m_paramsChanged / the envelope state. The decode loop exits within ~10 ms of m_running = false in practice, so this is unlikely to fire — but it's worth checking the return value and either retrying or logging, rather than dropping the pointer. (Same shape as the hid_write return-check fix in PR #3250.)

4. Minor: setters always pulse m_paramsChanged

setMarkFreqHz/setShiftHz/setBaudRate/setReversePolarity set m_paramsChanged = true even when the value is unchanged. Combined with refreshRttyDecodeState() being called from rttyMarkChanged / rttyShiftChanged / diglOffsetChanged / mode-change paths, this means the filter is re-designed and the bit-clock is reset on every spectrum-overlay-triggering signal — even when nothing relevant moved. Not a correctness bug, but it can interrupt decode mid-character on a clean signal whenever an unrelated overlay update fires. Cheap guard: compare-and-skip in the setters.

Nothing blocking apart from the stray STREAMDECK_PLUS_SESSION.md. Nice work.

---

_{🤖 aethersdr-agent · cost: $29.6434 · model: claude-opus-4-7}

[motoham88]

Thanks for the thorough review. Addressed in 6b0c585:

1. STREAMDECK_PLUS_SESSION.md — You're right it shouldn't be here. It was in the branch tree from the reset --hard main starting point; removed in the fix commit. The deletion also cleans it off main on merge, which seems like the right outcome.

2. DIGL auto-activation — Agreed. refreshRttyDecodeState now only fires on mode() == "RTTY". DIGL users who want Baudot decode can be handled as a future opt-in (context menu or a separate mode check). Also removed the vestigial diglOffsetChanged connection and simplified the Auto-mark path to just use s->rttyMark().

3. stop() wait timeout — Return value is now checked; logs a qCWarning on timeout rather than silently nulling the pointer. Worker exits within ~10 ms in practice so this is a safety net, not a likely path.

4. No-op setter pulsing m_paramsChanged — All four setters now compare-and-return-early when the value is unchanged. Overlay-triggered refreshRttyDecodeState calls that don't actually change any parameter no longer reset the bit clock mid-character.