Getting all of the open source tools set up for software radio experimentation can be a headache, but is a necessary prerequisite for any new students or researchers starting to dive into the area. Machine learning has some similar pains, getting things set up can take a bit of time and effort and cost additional ramp-up time for researchers.
We introduced PyBombs several years ago as a tool for the software radio community to help make getting dependencies, gnuradio builds, gnuradio modules, gnuradio apps and working environments set up in an easy, repeatable, portable, and user friendly way easy. Now, as SDR is growing to heavily leverage and encompass ML techniques, we need even more tools which are often more diverse and widely applicable than the GNU Radio centric SDR ecosystem. To help provide a baseline and a rapid method for getting up and running in this area with things that are easily added through pybombs and things that are easier otherwise, we’ve put together a set of Docker images which can be built and launched extremely easily and repeatably with pretty everything anyone needs to start doing some serious damage in the field. Hopefully these will be of great help for anyone getting going in this area!
Thanks to nvidia-docker, if you install the full-GPU version of the image, you can still leverage GPUs for much faster NN training times as well! Instructions for building and using these images can be found on github at: https://github.com/radioML/dockerRML
Update: Still wrestling with docker hub, currently the build times out and requires too much storage space for building through docker hub, *except* for the ML-only version. “docker pull radioml/ml” will pull a version of this image without GNU Radio or GPU support.