Pull request overview

This PR mutes KiwiSDR audio output whenever the Flex radio is transmitting and full duplex is disabled, mirroring SmartSDR’s RX-audio behavior while keeping Kiwi streams connected. It also extends the automation bridge to expose relevant audio-engine state needed to assert on the new mute gate.

Changes:

• Add a transmit-driven KiwiSDR mute predicate in MainWindow and synchronize it to AudioEngine (including RADE EOO holdoff integration).
• Implement an AudioEngine-level “KiwiSDR transmit mute” gate that suppresses mixing/output and clears/prebuffers appropriately.
• Extend the automation bridge (get audio) and documentation to surface kiwiSdrTransmitMuted plus audio buffer/stream state.

Reviewed changes

Copilot reviewed 9 out of 9 changed files in this pull request and generated 1 comment.

Reviewed the diff. This is clean, well-scoped work — thanks @rfoust. The mute gate is correctly driven from radio-authoritative full-duplex + TX/interlock state, the signal wiring covers the relevant paths, and the new automation audio snapshot is a nice testability win.

Copilot's data-race concern is valid in premise but already addressed. AudioEngine is moved to a dedicated worker thread (MainWindow.cpp:985), so building the snapshot on the AutomationServer (GUI) thread would indeed read m_audioSink/m_audioSource cross-thread. But the second commit (619be5de) routes through audioSnapshotOnObjectThread(), which formats the JSON on the AudioEngine's own thread via Qt::BlockingQueuedConnection and only calls inline when already on that thread (correctly avoiding self-deadlock). So the current head resolves it — Copilot appears to have flagged the first commit. ✅

A few things I verified that hold up well:

• Full-duplex toggle coverage — RadioModel::setFullDuplex() emits infoChanged(), which is wired to syncKiwiSdrTransmitMute(), so the gate releases if FDX is enabled mid-TX. Good.
• Liveness watchdog interaction — feedKiwiSdrAudioData() updates m_lastAudioFeedTime before the transmit-mute early return, so the #1411 RX-restart watchdog won't spuriously fire during a long transmit while Kiwi packets are being dropped. Correct ordering.
• Gate state — m_kiwiSdrAudioTransmitMuted atomic + m_dspMutex-guarded buffer flush in setKiwiSdrAudioTransmitMuted(), invoked via QueuedConnection onto the audio thread. Thread-safe.

One minor, non-blocking observation: in the multi-source feedKiwiSdrAudioData(sourceId, ...), the source->muted check moved to after m_lastAudioFeedTime.start(), so an enabled-but-muted source now bumps the liveness timer where it previously didn't. Given the watchdog's purpose this is harmless (arguably better), just calling it out as an intentional-looking behavior change in case it wasn't.

Conventions all look good — atomics with explicit memory order, no flat-key AppSettings, RAII, Q_INVOKABLE/queued-invoke pattern consistent with the existing Kiwi code, and the docs table updated for the new audio model + fullDuplex field. Nice work.

_{🤖 aethersdr-agent · cost: $3.3854 · model: claude-opus-4-8}

Approved. Reviewed the mute gate end-to-end:

• Radio-authoritative predicate (Principle II) — kiwiSdrTransmitMuteRequired() = not-full-duplex AND (isTransmitting ∨ isTuning ∨ isRadioTransmitting ∨ RADE-EOO). isRadioTransmitting fires from the interlock handler, so hardware PTT / DAX / TCI / VOX are all covered alongside client MOX/Tune. Not a keying path — Principle VI N/A.
• Reliable unmute — every TX-on path has a matching off-signal (moxChanged/tuneChanged/radioTransmittingChanged + RADE transitions), and the state-change guard prevents redundant calls. FDX toggled mid-TX releases the gate via infoChanged.
• Mute, not pause/disconnect — atomic flag gates externalKiwiSourceAudible(); the Kiwi stream stays connected, no churn.
• Thread-safe — atomic + m_dspMutex-guarded flush via QueuedConnection onto the audio thread; the new audio bridge snapshot reads cross-thread state via BlockingQueuedConnection on the audio thread (Copilot's race resolved in 619be5d).
• #1411 watchdog ordering preserved — liveness timer bumped before the mute early-return.

Live-verified on a real radio; CI green on all 6. Thanks @rfoust.