Pull request overview

Adds a clean-room, receive-only KiwiSDR integration to AetherSDR, exposing Kiwi endpoints as client-owned virtual RX antennas that can be selected from existing antenna controls, routed through the RX DSP/speaker path, and rendered via the existing spectrum/waterfall surfaces.

Changes:

• Introduces KiwiSdrProtocol, KiwiSdrClient, and KiwiSdrManager to encapsulate KiwiSDR receive behavior, persistence, reconnection, and per-slice assignment.
• Wires Kiwi virtual antennas into existing UI flows (VFO/RX applet/pan overlay), adds a Kiwi status applet, and adds waterfall/meter handling when Kiwi is active.
• Fixes an NR2 (SpectralNR) block-size regression (packet-burst vs hop cadence) and adds protocol + NR2 regression tests.

Reviewed changes

Copilot reviewed 37 out of 37 changed files in this pull request and generated 1 comment.

Thanks for this, @rfoust — this is a genuinely careful, well-structured contribution. The clean-room boundary is clearly documented, the protocol parsers are defensive about frame sizes, the audio/DSP integration follows the existing source-FIFO pattern, and the conventions I'd normally flag are all respected: receive-only RX antenna profiles persist under a single feature-owned JSON key via AppSettings (no flat keys), sockets are owned through deleteLater()/RAII with disconnect(this) before teardown, cross-thread flags use std::atomic, and auth passwords are redacted in command logs. Nice work.

A few small notes, none blocking:

1. Magic number — confirming the Copilot finding (valid).
src/core/KiwiSdrProtocol.cpp:50 hardcodes the literal 1034:

const bool observedExtendedFrame = frame.size() == 1034;

The constant kObservedExtendedSoundFrameBytes = 1034 is defined at line 11 and already used by extractMeterFromSndVerifiedLayout at line 155. Using the constant here keeps the header parser and the meter extractor from drifting if the observed frame size ever changes. (Note KiwiSdrClient.cpp:31 independently defines its own kObservedSoundFrameBytes = 1034 — three copies of the same observed value across two TUs; worth at least the one-line fix Copilot suggests, and ideally a shared constant.)

While there: the && frame.size() >= 8 on line 51 is dead — observedExtendedFrame already implies size == 1034. Harmless, but it reads as if a non-1034 path could reach the branch.

2. Minor consistency in the multi-source feed.
In AudioEngine::feedKiwiSdrAudioData(sourceId, ...) (AudioEngine.cpp:1917), the non-NR2 path does source->rxBuffer.append(...) directly, whereas the legacy single-source variant routes through processRxAudioData(). The buffer-append-without-lock matches the established Flex/legacy non-NR2 pattern (processRxAudioData:2147), so this isn't a new race — just calling out that the two Kiwi feed overloads take different routes to the same end, which is easy to miss when maintaining one and not the other.

On CI: the test plan leaves "Existing tests pass in CI" unchecked. I wasn't able to pull the check-run status for the head SHA through tooling, so I can't confirm it either way here — worth a glance before merge given the size of the AudioEngine/SpectrumWidget surface this touches.

Scope looks clean — the files not named in the description (VfoWidget, SliceModel, RadioModel, SMeterWidget) all line up with selecting Kiwi profiles from the slice/panadapter antenna controls and routing the meter/waterfall, so nothing reads as out-of-scope.

Overall this is in good shape. The only thing I'd actually ask you to change is the 1034 literal.

_{🤖 aethersdr-agent · cost: $5.1299 · model: claude-opus-4-8}

✅ Approved

Reviewed across Principle I (Flex protocol authority), Principle V (settings), audio-path thread-safety, and clean-room provenance — all pass. CI was green; the clean-room record in docs/kiwisdr-cleanroom-design.md is credible (no AGPL/copied snippets; constants framed as observed/experimental).

Highlights

• Principle I — PASS: Kiwi profiles are namespaced KIWI: and guarded at every antenna-selection entry point (VfoWidget / RxApplet / SpectrumOverlayMenu) before any rxant= command. No path leaks a Kiwi name to the radio; the real Flex antenna value is preserved while a Kiwi antenna is active.
• Audio — thread-safe & mixing-correct: single-thread producers → QueuedConnection → single-thread audio consumer; sums to float, scales by 1/N, clamps; per-source NR2/resampler/FIFO isolation. The SpectralNR overlap-add fix (with its block-size-invariance regression test) is a genuine correctness win.
• Principle V — PASS: one nested-JSON key KiwiSdrRxAntennas, robust to malformed JSON, fields clamped.

Governance: covered by the now-approved RFC #3613 (broadened to cover KiwiSDR; the "virtual RX antenna" surface is the approved go-forward design for remote receivers).

Review fixes folded in (commit a82d38d, signed):

1. Audio-source leak — removeKiwiSdrAudioSource() had no caller, so profile removal only disabled the source and leaked its DSP state. Added KiwiSdrManager::audioSourceRemoved → wired on the audio thread to free the entry.
2. Stripped a checked-in developer-specific absolute path from the design doc.

Both Kiwi/NR2 tests pass locally; clean build. Excellent work @rfoust.

Re-approving after the Copilot review fix (6918050): parseSoundFrameHeader now uses kObservedExtendedSoundFrameBytes instead of the repeated 1034 literal, so the parser and meter-extraction size checks share one constant. Build + kiwi protocol test pass locally. Substance unchanged from the prior approval (Principle I guard, Principle V settings, thread-safe audio, clean-room provenance all verified).