Detach kernel USB-audio driver so Android sounds can't kill a USB-direct TX

[patrickrb]

Problem

USB-direct TX intermittently goes out as dead air: the rig keys, debug.log shows
libusb native write FAILED (rc=5 TRANSFER_NO_DEVICE) after ~280ms, no usbDetach,
and RX capture on the same device keeps running. Reported by users as "TX stops
working until I unplug/replug" (the replug is a coincidence — see below).

Root cause (2026-08-25 bench log, Pixel 11 Pro XL + CM108 0d8c:0012)

claimInterface(force) on the UAC AudioStreaming interfaces is a silent no-op for
the kernel's snd-usb-audio driver: it binds the card at the AudioControl interface
and marks the streaming interfaces owned-but-unused. So the ALSA card survived our
claim and Android kept the rig's sound card registered as a usb_headset sink +
source (dumpsys audio → setWiredDeviceConnectionState … from UsbAlsaDevice).

Every sound Android routed there made the kernel driver flip interface 1's
alt-setting under our in-flight iso URBs; the kernel completes those with
-ESHUTDOWN → libusb TRANSFER_NO_DEVICE. Correlation over the whole log:

write result	alert ≤6 s before	no alert
OK	1	302
FAIL (rc=5, device still on bus)	20 (all at dt ≈ 1.9 s)	2 (BT car-kit connect at 09:08, nav prompt in car)
FAIL (rc=-4, whole bus dropped)	0	2 (electrical, see below)

The dominant trigger is the app's own QSO-complete DX-alert notification ding
(ALERT fire … QSO complete 1.9 s before the TX slot). rc=-1 IO at submit is the
same thing when the alt-setting was already flipped back before our submit.

Iso OUT has no device-side handshake, so only host-side actions can fail an OUT URB
while the device stays attached; RX surviving pins it to an interface-1-only action,
i.e. the kernel class driver.

Fix

UsbAudioDevice.open() now calls detachKernelAudioDriver(), which force-claims
the AudioControl interface. That runs the real usb_audio_disconnect, retiring the
ALSA card, so nothing Android plays can reach the endpoint. The per-cycle TX open
skips the claim when the session-long RX capture already holds it on the same
device. No AudioControl interface / refused claim → logged, behaviour unchanged.

Side effect, by design: phone sounds Android would have routed into the rig's mic
input are dropped instead (they were inaudible to the operator anyway and could have
been keyed on air). The kernel doesn't rebind the driver on release; the card
returns on the next replug.

Also: the UsbAudioWriteErrorTest Javadoc blamed rc=5 on RF knocking the device off
the bus — corrected. CLAUDE.md gains TX-pipeline gotcha #4 with the log signatures.

Not fixed here (electrical)

Two failures today were a real bus drop mid-TX: usbDetach for hub + CP2105 + C-Media
together, serial.send: port not open!, and dumpsys batterystats --history shows
the phone lost hub power (-plugged) at the same instant (08:37:02 at 35 % drive on
40 m, 09:00:12). That's the hub resetting under RF — ferrites / powered hub / shorter
cable, not software.

Tests

• New UsbAudioControlInterfaceTest (7 tests) for audioControlInterfaceIndex().
• testDebugUnitTest --tests com.k1af.ft8af.wave.*: 19 suites green.

Hardware verification — TODO before merge

Not yet run on hardware (the bench phone carries the non-debuggable Play build, so a
debug install would wipe its QSO log). To verify: plug in the rig interface, check
debug.log for UsbAudioDevice: kernel audio driver detach via AudioControl iface 0: OK,
then trigger a QSO-complete alert (or any notification sound) ~2 s before a TX slot
and confirm the next write logs libusb native write OK. Also confirm
dumpsys media.audio_flinger doesn't misbehave once the ALSA card is gone.

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 42.36%. Comparing base (551b0a2) to head (36d469f).

Additional details and impacted files

@@             Coverage Diff              @@
##                dev     #777      +/-   ##
============================================
+ Coverage     42.29%   42.36%   +0.07%     
  Complexity      226      226              
============================================
  Files           267      267              
  Lines         31730    31819      +89     
  Branches       3634     3650      +16     
============================================
+ Hits          13419    13479      +60     
- Misses        18054    18082      +28     
- Partials        257      258       +1     

Flag	Coverage Δ
android	16.91% <ø> (+0.25%)	⬆️
native	9.93% <ø> (ø)

Flags with carried forward coverage won't be shown. Click here to find out more.
see 2 files with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[Copilot]

🟡 Changes recommended

Failed opens can leak a USB connection and control-interface claim, while the core claim behavior lacks regression coverage.

Once you've addressed the issues Copilot identified, you can request another Copilot review.

Pull request overview

Detaches Android’s kernel USB-audio driver to prevent routed sounds from interrupting direct USB FT8 transmission.

Changes:

• Force-claims the UAC AudioControl interface.
• Adds interface-selection unit tests.
• Updates USB failure diagnostics and documentation.

File summaries

File	Description
UsbAudioDevice.java	Implements driver detachment and claim lifecycle.
UsbAudioControlInterfaceTest.java	Tests AudioControl interface selection.
UsbAudioWriteErrorTest.java	Corrects failure-cause documentation.
CLAUDE.md	Documents the USB-direct TX failure mode.

Review details

• Files reviewed: 4/4 changed files
• Comments generated: 3
• Review effort level: Balanced

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.