This addresses a defect introduced in 6e75abf which causes the
_consecutive_failures count to get reset with every failure, rather
then upon a successful return value.
This change is intended to make the bladeRF source/sink implementations
slightly more resilient to any transient issues in a flow graph.
For poor choices of buffers/transfers or under high CPU load, an RX or
TX operation might time out. Instead of immediately reporting WORK_DONE
and bailing out, an error message is now printed and the device will
attempt to continue transmitting/receiving samples.
After 3 consecutive errors have occurred on an RX/TX operation, the
device will report WORK_DONE; in this situation, something is likely
very wrong.
libbladeRF provides accessors for rational sample rates, which the
integer sample rate functions use under the hood. Therefore, there's no
need to check if the requested rate contains a fractional portion and
switch between the two sets of functions.
Common parameter handling has been moved into bladerf_common::init().
The buflen parameter is now in units of samples, not bytes. This
deviates from the other gr-osmosdr items. However, with the requirement
that buffers be in multiples of 1024 samples, this makes specifying this
parameter a bit easier. The user shouldn't need to know we're operating
on SC16Q12 values under the hood, and have to calculate accordingly.
To avoid confusion when both a bladeRF source and sink are in a flow
graph a [bladeRF source/sink] prefix has been added to output. Error
number have been replaced with bladeRF string representations of these
error values.
Firmware flashing has been removed. The bladeRF-cli or bladeRF-flash
tools are the preferred route for firmware upgrades.
The gr_complex FIFO is no longer used on the TX side, so it doesn't
really make sense to have it in bladerf_common. The associated items
have been moved into bladerf_source, and some renaming has been done in
bladerf_sink to tidy up.
Pending further performance tests of the bladerf_source, the _fifo
member (boost::circular_buffer) may need to be replaced.
Removed the use of an intermediate sample FIFO in the sink
implementation. Note the the FIFO has not been moved out of
bladerf_common --> bladerf_source_c in this commit.
work() now handles converting samples from complex to SC16_Q12, and filling
"transmit-ready" buffers. The callbacks are now only responsible for
marking the provided buffer free, and returning the next buffer.
It appears that a small deadlock issues remains in this changest, which
can be induced by:
1: Using a small sample rate (160Khz)
2: Switching back and forth between sinusoid <-> GSM burst
In this case, it appears that work() is blocked waiting for a buffer to
become free. More investigation here is required...
This is based on the original work (https://github.com/Nuand/gr-osmosdr)
done by folks at nuand LLC for the gr3.6 branch of gr-osmosdr.
The following modifications have been done in this commit:
* port to gr-osmosdr master codebase (gr3.7)
* moved shared properties to bladerf_common
* added & verified IF filter bandwidth setters
* set LMS6002D registers with values taken from FAQ 5.27
* print device information (serial/versions) on startup
* added fpga= and fw= device arguments to program MCU/FPGA
* added bladerf=# dev. arg. to select a specific bladeRF
* grc gain field controls RF path VGA for RX/TX
* grc BB gain field controls BB path VGA for RX/TX
Usage example:
osmocom_fft -a "bladerf,fpga=/tmp/hostedx115.rbf"
The following RX named gain stages are available:
LNA: 0 to 6 dB, in 3dB steps
VGA1: 5 to 30 dB, in 1dB steps; nonlinear mapping done inside the lib
VGA2: 0 to 60 dB, in 3dB steps; not recommended to be used above 30dB
The following TX named gain stages are available:
VGA1: -35 to -4 dB, in 1dB steps, BB side
VGA2: 0 to 25 dB, in 1dB steps, RF side
Thanks a lot to the team of nuand LLC for this major contribution.
Jon Szymaniak
Dimitri Stolnikov