Please find below a patch as per suggestion of Martin Ling: call libusb_interrupt_event_handler in rtlsdr_cancel_async to trigger the event loop handler to look at the async_cancel variable. (https://lists.osmocom.org/hyperkitty/list/osmocom-sdr@lists.osmocom.org/thre...) I have tested timings on a windows 11 machine with and without the patch when running a async transfer at two different sampling rates: sample rate with without theory 1536K 0 - 2 ms 0 - 84 ms 85 ms 288K 0 - 4 ms 0 - 457 ms 455 ms The table shows the range of timings to cancel the async transfer as measured over a couple of runs. The "theory" column shows the theoretical time to complete one transfer at the particular sampling rate (i.e. driven by the size of a transfer buffer). And indeed, on some machines it appears that cancelling the async transfer, if unlucky, can take as much time as waiting for a full transfer to complete. The call to libsub_intertrupt_event_handler fixes that. For large transfer buffers and/or low sample rates this can make a difference. Thanks to Martin for sharing the insight. Jasper Author: jvde.github <jvde.github@gmail.com> Date: Wed Jan 26 19:00:55 2022 +0100 call interrupt_event_handler in cancel_async diff --git a/src/librtlsdr.c b/src/librtlsdr.c index 67e30c4..b949692 100644 --- a/src/librtlsdr.c +++ b/src/librtlsdr.c @@ -1948,6 +1948,9 @@ int rtlsdr_cancel_async(rtlsdr_dev_t *dev) if (RTLSDR_RUNNING == dev->async_status) { dev->async_status = RTLSDR_CANCELING; dev->async_cancel = 1; + + libusb_interrupt_event_handler(dev->ctx); + return 0; }