A client that calls bladerf_connect and never disconnects kept the USB
claim until the server exited, blocking every other tool on the shared
bladeRF. A watchdog thread now releases the device after
MCBLADERF_IDLE_TIMEOUT seconds without a tool call (default 600, 0
disables). Middleware counts tool calls in flight, so a release never
lands mid-call, and it also skips while a capture runs. The RX/TX
settings are saved first; the next device call reconnects and restores
them, sample rate included, since the radio resets that to 4 MHz on
open. bladerf_status reports auto_released in between.
Version 2026.11.15.
- CS16 streams read into an int16 buffer; files hold raw SC16 Q11 (full
scale 2048) instead of all zeros
- bladeRF-cli backed tools release the USB claim while they run and
restore RX/TX settings afterwards; loopback uses the plugin setting
- calibrate_dc runs "calibrate lms" + "calibrate dc rxtx"; a bare "cal dc"
only printed help and reported success
- report the span of all SoapySDR ranges (rate was capped at 10 Msps)
- parse "Firmware version" from bladeRF-cli 1.10
- startup banner goes to stderr, off the stdio JSON-RPC channel
- shared RX chunk reader: settle discard, overflow and stall handling,
stream always closed; captures report overflows, clipping, peak dBFS
- spectrum power normalized to dBFS
- unit tests, README troubleshooting, version 2026.10.25